Mirakle Research · Listing histories · 29 September 2026
The hyperoptimizers: what 211 Shopify apps keep changing, and whether it pays off
Between 13 June and 20 September, 211 Shopify apps rewrote their own listings again and again, some of them every four or five days. We call them hyperoptimizers. This is an account of what they changed, read first as counts and then as sentences, and of how far their reviews can be tied to any of it.
What we found. The editing is real, mostly verbal and steady. Nine price editors return exactly to a price they had already tried; in the two we read closely, the package around that price had changed. Read as sentences, the edits make offers narrower and more specific: a named competitor's customers, a named task, a named safeguard. There is no store-wide AI wave. Around each app's first edit, review counts rise for 49 eligible apps and fall for 26, and pooled they grow faster than those of quiet listings in the same categories; the pooled rating on the 25 apps with enough ratings barely moves, 4.90★ to 4.94★. But the sentences that would explain a rise mostly arrived after the windows they would have to explain, and aligned to the wording itself the one apparent reward, for AI, is almost all one app. Listings that are being worked on gain reviews; which edit, if any, produced one is not identified.
Step 1 · Where this started
The first signal was our own parser
The question began narrow. Had two apps owned by one developer, Wide Bundles and WideReview, changed their listings to chase reviews around a competitor's product launch? Their June change-log rows looked like a rewrite.
Their complete available listing history, 27 April to 29 September, records zero pricing changes and exactly four copy/feature-bullet rows. All four are an exact reconstruction of the old combined text, matching our own 9–10 June parser repair (commits 3bfb2c9 and 0dc831b), not a merchant-facing edit. Their dated review count fell across the suspected window, from 7 in April–May to 2 in June–July. Mat De Sousa's own newsletter describes a pricing test in October 2023 and further tests in January 2024: first-party accounts of tests our listing history starts too late to see. The null result is about this database's window; it does not contradict his account.
The failure shaped everything after it. A change log records what our parser saw, and a parser repair looks exactly like a merchant's edit. So the screen below opens on 13 June, after that repair, and every edit it counts has to survive cleaning rules written to reject our own artefacts. The question that survived was broader: across the whole store, what does repeated listing work look like when it is real?
Step 2 · The screen
A screen that never looked at reviews
We read every description, tagline and pricing-tier row recorded for 29,410 stored apps over 100 days, 13 June to 20 September. Of 17,696 raw rows, 14,140 survived cleaning: a copy edit needs six changed word tokens and a nonempty baseline, a plan edit needs a genuinely different canonical plan list, and either needs two days before the next same-type row, so a flicker cannot count. 6,526 apps kept at least one edit.
An app qualifies as a hyperoptimizer with at least three distinct copy-edit days, two plan-edit days and five edit days in total. 211 apps under 197 developers do. The word is a label for measured behaviour, not a claim that any of these businesses runs a disciplined test process. We cannot see intent, staffing, or whether a change was tested rather than decided and shipped. We can see what changed, how often, whether a later state exactly repeats an earlier one, and what happened to reviews afterward. Review counts, ratings and reversals played no part in setting the thresholds, and 80 of the 211 have no unedited dated review in the window at all.
Step 1 left a specific worry. Cleaning rejects flicker; it cannot reject a parser change, which rewrites the listings it touches as if they had been edited, and that is exactly what fooled us in June. A store-wide change of that kind puts a large part of the store on one date, and the retained edits do not look like that: they fall on 98 of the 100 days, a median of 15 apps a day, and the busiest date, 9 September, holds 31 of the 211. That rules out a store-wide repair. It cannot rule out a narrower one, a change that reached only listings built on one template or was processed over several days, and nothing in a change log separates such a row from a merchant's edit. The counts in step 3 carry that residual risk. The close readings carry less of it: each was read as a before/after pair and changes what the listing claims, where June's repair was an exact reconstruction of text already there.
The other worry about any threshold is that it manufactured the group. So we tightened it until it hurt:
| Minimum changed word tokens | Minimum spacing to next row/boundary | Qualifying apps | With zero dated reviews |
|---|---|---|---|
| 6 | 2 days | 211 | 80 |
| 12 | 2 days | 171 | 70 |
| 6 | 7 days | 93 | 28 |
| 12 | 7 days | 75 | 23 |
The size of the group depends on the rule. Seven-day spacing also removes real fast iterations; it is a deliberately severe sensitivity, not a better definition, and none of these screens diagnoses a controlled test. But the group does not dissolve: under the harshest pairing, 75 apps still edit repeatedly. Together with the calendar check, that is the licence for what follows. From here on, “hyperoptimizer” names a pattern that survives stricter definitions and does not look like a store-wide artefact of our instrument, not one our thresholds conjured.
Who they are
Grouping the qualifying apps exposes portfolio activity. A company's total below counts app-days: changes to three apps on one date contribute three, not one. Developer ownership and names are those stored at extraction.
| Developer | Qualifying apps | App-edit days | Dated reviews |
|---|---|---|---|
| Pepin | 3 | 31 | 0 |
| momiji-shop | 5 | 30 | 0 |
| Pixelspiece solutions | 2 | 21 | 12 |
| Adot Technologies Inc | 1 | 21 | 6 |
| SetuBridge | 3 | 20 | 98 |
| Nexus Logic Systems | 1 | 20 | 3 |
| Buddify | 2 | 17 | 61 |
| WowApps.io | 2 | 17 | 19 |
| App HQ Pte. Ltd. | 1 | 16 | 70 |
| SHOP BOPIS | 1 | 15 | 2 |
| Petru Marchidan | 1 | 14 | 0 |
| WEBx47 | 1 | 14 | 0 |
Pepin's three qualifying apps account for 31 app-edit days and zero unedited review dates in the window. Its Marutto search listing raises Basic / Pro from $19.99 / $29.99 to $29.99 / $49.99 on 3 July, changes plan copy repeatedly and moves between Japanese and English presentation: plausibly deliberate localisation, or a change in what our scraper received; it cannot be labelled a keyword experiment from text alone. Right behind it, momiji-shop runs five small compliance-flavoured listings (a B2B terms guard, a green-claims guard, an address guard, a food-info guard, an “agentic catalog readiness” app) through 30 app-edit days and zero unedited reviews of their own. SetuBridge has three qualifying apps; Pixelspiece and Buddify each have two. These portfolios are useful subjects for ongoing observation. They do not establish that every app a company owns follows the same strategy.
Step 3 · What they change, and how often
Mostly words, about every ten days
With the group established, the first thing to measure is what its members actually touch. Across the 211 apps' 1,781 retained edits, 909 changed body or tagline copy and 358 changed a displayed price or trial length: about two and a half copy edits for every price edit. Another 514 changed a plan's name or bullet text without moving its price or trial. A copy edit needs no revenue decision behind it, which may be part of why it happens more often.
By app, 95 lean toward copy (more copy-edit days than plan-edit days), 68 lean toward plans and 48 split evenly. 170 of the 211, about four in five, touched a displayed price or trial at least once; the other 41 edited copy and plan bullets and never their price. A copy edit, up close, is rarely a synonym swap. Arvio's 18 June tagline is typical:
Before
Grow without another hire:AI thatdrafts your productedits, SEO, andemailwork for you.
After
AI store operator that edits products, optimizes SEO & runs email — all on your approval
The pace is the second measurement. The median hyperoptimizer's own mean gap between two retained edits is 10.5 days. Arvio averages one every 4.6 days for three months; Buddify, the slowest of the eight histories in the appendix, one every 14.3. A cadence like that could be one launch burst averaged over a summer. It is not: after the partial first fortnight the cohort edits at 19.3 a day in July, 19.4 in August and 17.1 in September's first twenty days, with more apps active in July and August than in June.
A cadence says the listings keep moving. It does not say whether a business is trying something and taking it back, or walking in one direction. Those leave different traces in a price list, and the price list is the one field we can compare exactly.
Step 4 · Does any of it look like a live test?
Nine returns to an earlier price, two read closely
A reversal is a later price/trial-set change whose captured price and trial pairs exactly match an earlier captured state for the same app: the listing returned to a state we had already recorded instead of moving to a new one. Of 170 apps that changed a price or trial at least once, 9 did this: Zapiet Eats ‑ Restaurants, Syncost: Profit Tracker & P&L, PromoLock: Discount Blocker, Stockful: Inventory Management, Shopicraft Build a Box BYOB, SizeMap Size Charts & Guides, CartWatch: Bot Detection Block, HomeVisioner: Room Visualizer, and Asks: AI Chatbot & Live Chat.
PromoLock: Discount Blocker is the clearest case. Between 8 July and 17 September it recorded eight retained price/trial-set changes, the most of any app in the cohort. The 8 July menu has $79 / $249 / $499 tiers; by 19 July Growth is $600/month, later falling to $300, then $299. Starter settles near $99 through several feature rewrites. These are change events, not eight unique states. On 17 September its two-tier state, $299 and $99 both on 3-day trials, exactly matches the pair captured on 5 September, twelve days earlier, even though the $299 tier had been renamed from “growth” to “plus” in between. The price returned before the name did.
Zapiet Eats – Restaurants shows the same pattern on a slower clock. It opens with a free tier and a $79.99 paid tier, cuts the paid tier to $49.99 on 15 July, briefly collapses both into one $49.99 “zapiet eats” plan on 18 July, then on 3 August reintroduces a free tier, relabelled “development stores”, alongside the same $49.99/14-day paid tier. The two-tier, two-price structure came back sixteen days after it briefly disappeared, and the returning free tier addresses development stores; it is not a general free plan.
| App | Retained price/trial changes | Reversal detected |
|---|---|---|
| Zapiet Eats ‑ Restaurants | 4 | 2026-08-03 |
| Syncost: Profit Tracker & P&L | 3 | 2026-08-18 |
| PromoLock: Discount Blocker | 8 | 2026-09-17 |
| Stockful: Inventory Management | 3 | 2026-09-04 |
| Shopicraft Build a Box BYOB | 3 | 2026-09-01 |
| SizeMap Size Charts & Guides | 4 | 2026-08-25 |
| CartWatch: Bot Detection Block | 3 | 2026-07-31 |
| HomeVisioner: Room Visualizer | 4 | 2026-08-19 |
| Asks: AI Chatbot & Live Chat | 3 | 2026-08-21 |
Nine of 170 is not evidence that testing is rare. A reversal only fires when the same exact price/trial pair recurs, which a business searching for the right price by trial and error need never produce; it can keep moving in one direction, the way Arvio, AppHQ and True COGS do. A reversal is the strongest signal this record offers that a change was provisional, and it has two duller explanations. The first is our instrument again: a scraper that is sometimes served an older copy of a page would record exactly this pattern. The spacing rule already sets aside states that last under two days, and PromoLock's match passes a harder test: the 17 September menu is not a stale copy of 5 September's page, because its $299 tier carries a different name. The second needs no test at all to survive: a price put back because the change was a mistake looks, in a price list, exactly like a test withdrawn. So a reversal says a change was provisional. It does not say why.
The two close readings below complicate the signal from the other side. In both, the price came back and the cards and sentences around it had moved on.
PromoLock: Discount Blocker · WEBx47
PromoLock: the price comes back; the promised job keeps changing
The July 2 rewrite gives a technical discount blocker a financial job. The old body explains code patterns, gift-card rules and error messages. The new last sentence asks the merchant to judge each promotion by the margin it consumes. On July 8 the tagline becomes a sequence of actions: create, measure, block. The body adds named loss categories, “sale loss, stack loss, and per-code breakdowns.” This is a more specific explanation of the value being sold, not merely another way to say “discount app.”
Before
PromoLockgives merchants precise control overdiscounts and gift cards at checkout. Blockallcodes site-wide,restrict to specific patterns,or set custom rules to handle any type of cart — includingsale itemsdiscount control.Apply dynamic variables sorulesadapt in real time,and show custom error messages that stay on-brand — no theme changes,no performance impact.
After
PromoLock rules how discounts and gift cards behave at checkout. Block or allow discount codes site-wide or by pattern, stop codes stacking, and control discounts on sale items. Create your own discounts with the limits and flexibility to protect your margin. Set redeem rules on individual gift cards, such as locking a card to one customer or requiring a minimum cart. See the exact margin each offer costs you, not just the revenue it brings.
That promise sits beside changing customer segments. The July 8 plans name Shopify Basic/Grow, Advanced and Plus stores. Later cards segment by monthly GMV. By September 17, Starter says “Non-Plus stores, up to $250k GMV/mo” while Plus says “Plus stores or $250k+ online GMV/mo.” The larger price is attached to more traffic, checkout controls and fraud protection. The copy tells us whom the tiers are meant to serve, which a price vector cannot.
Before
Control every offer at checkout. Know the exactmargineach one costs.
After
Create discounts that protect margin. Measure what slips through. Block what bleeds you.
The September 17 reversal still holds: $99 and $299, each on a three-day trial, recur from September 5. But the $99 card has changed from “Custom block rules using dynamic values” to “Unlimited custom discount block rules”; its loss-analytics and coupon-leak language has changed too. An identical price/trial pair is not an identical product package. The experiment signal survives, with a narrower interpretation.
Before
Block discounts by code orpattern,createprofit-first offers,andanalyze losses onyourtop codes
After
Block discounts by code or rules, build profit-first offers, analyze & fix losses on top codes.
Titles participate in the same search. From August 27 the observed name switches to “PromoLock: Discount Blocker,” returns to “PromoLock: Block Discount Loss” on September 6, then switches back the next day. The function and the financial pain take turns at the front of the listing.
Reviews around its first retained edit, 2026-07-02. No dated reviews on either side.
Before 2026-06-04–2026-07-01; after 2026-07-03–2026-07-30; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. There are no dated reviews in either original comparison window. The wording supports an interpretation of positioning and segmentation; it cannot show that either pitch converted, improved margins, or raised ratings.
Zapiet Eats ‑ Restaurants · Zapiet
Zapiet Eats: a narrower customer, an award claim, and a restricted free tier
The first retained rewrite drops the opening “Stop losing 30%” argument and narrows the intended setting. Instead of a list of restaurants, cafes, bakeries, QSRs and popups, the body spends scarce space on “in-store coffee shops inside retail and apparel stores.” The problem is now running a kitchen inside a Shopify retail operation: POS, preparation times, delivery zones and menus by time or location. This is a concrete customer segment.
Before
Stop losing 30% of every order to third-party food ordering apps.Zapiet Eats helps restaurants, cafes,coffee shops,bakeries,QSRsandpopup food businesses turn their own website into a direct ordering channel. OfferASAP orscheduled pickup and delivery,manageordersin real timefrom POS orAdmin,setdelivery zones,show accurate time estimates, controlmenu availability andcollectreviews automatically.
After
Zapiet Eats helps restaurants and coffee shops take direct pickup and delivery orders through Shopify, without third-party commission fees. Built for busy kitchens, takeaway counters and in-store coffee shops inside retail and apparel stores, it lets customers order ASAP or schedule for later, while your team manages orders from Shopify POS or admin, sets prep times and delivery zones, controls menu availability by time, day or location, and collects customer reviews automatically.
The July 25 opening adds “Shopify Build Award Winner 2026” before the functionality. That is an authority claim in the listing; this study does not independently verify the award. The August 3 tagline broadens the searchable tasks to restaurant ordering, food delivery and bookings, retaining the no-commission proposition. The replacement of a percentage-based loss warning with “0% commission fees” keeps the economic claim but changes how the reader encounters it.
Before
Zapiet Eats helps restaurants and cafes take direct pickup anddeliveryordersthrough Shopify, withoutthird-partycommission fees.
After
Shopify Build Award Winner 2026 – Pickup, delivery, dine-in and reservations through Shopify, without 3rd party commission fees.
The price/trial structure returns on August 3 after the July 18 one-tier state. Reading the plan text prevents calling this a general free-product relaunch: “development stores” is the restored free tier. The June free card already said “Development stores stay free forever”; the new label puts that restriction at the front. This return is sixteen days after July 18. The old report’s six-week wording was a calendar error.
Before
Onlineorderingfor restaurantsandin-storecoffee shops,withnocommission fees
After
Restaurant ordering, food delivery and bookings for coffee shops and cafes with 0% commission fees
The body mentions automatic review collection both before and after the June rewrite. Its presence is therefore not evidence that the rewrite newly introduced that mechanism. The award sentence appears after the original June 25–July 22 postperiod, and the August copy arrives later still.
Reviews around its first retained edit, 2026-06-24. Dated reviews 0 → 1; mean — → 5.00★ on 0 → 1 rated.
Before 2026-05-27–2026-06-23; after 2026-06-25–2026-07-22; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The original comparison is zero reviews before and one after, with no baseline rating. Neither the award claim nor the August price reversal can explain that earlier review. The stronger finding is the shift in customer and task framing.
A price vector told us PromoLock came back. Its plan cards say the product sold at $99 had changed, and its titles took turns between the function and the pain. Counting edits by kind treats a rewrite as a unit. From here the unit is the sentence.
Step 5 · Reading the words
No AI wave in a fixed panel; named specifics grow
PromoLock asks a merchant to see the margin lost through a discount. Prediko asks a Stocky user to imagine bringing suppliers and purchase orders along when switching. Buddify alternates between the size of its catalogue and the relationships inside its network. Those are different reasons to choose an app, and counting them as copy edits made them look much more alike than their sentences are.
The words come from the same change log: 1,179 description/tagline rows and 885 feature rows across the same 211 apps, plus category records, pricing-card wording and a shorter title history. A title names a category in a few words; a tagline chooses the first reason to care; a body explains the workflow; bullets can qualify an allowance or move a feature into a paid plan. We read them separately before combining any keyword counts. The change log stores changed fields only, so an absent description key beside a tagline edit means no description change was captured in that row, not that the body was erased.
| Stored field | Rows with a changed field | Apps represented | Nonempty endpoint pairs |
|---|---|---|---|
| Description | 1,022 | 211 | 211 |
| Tagline | 812 | 204 | 204 |
| Feature bullets | 885 | 209 | 209 |
| Feature tags | 0 | 0 | 0 |
| Highlights | 0 | 0 | 0 |
| Primary category | 33 | 29 | 29 |
| Categories | 0 | 0 | 0 |
| Category feature groups | 283 | 140 | 138 |
| Works with | 0 | 0 | 0 |
| Pricing tiers | 1,061 | 211 | 195 |
Rows can carry multiple fields, so column totals are not distinct events. Zero integration-field changes means none were logged in this window; integrations still appear in changing prose. Titles come from a separate observation stream. Every listed app has a nonempty before/after body chain, but not every app has a changed tagline or feature list.
This reading also keeps edits the frequency screen threw away. Prediko's first “Replaces Stocky” phrase and Wiser's insertion of “selleasy” are too small or too brief to qualify as retained edits, and they are among the most informative rows in the log. They explain the already-selected apps; they do not enlarge the cohort.
Is the cohort converging on one vocabulary?
The first question to put to the words is the one a keyword trend would answer: is everyone moving toward the same language? Any trend over a growing set of listings has a rival explanation in the denominator, since newer apps may simply arrive already speaking the fashionable words. So we fixed the panel. For 146 apps already observed when the screen opened, the body, tagline and feature bullets can be reconstructed at five boundaries with a nonempty old and final value for every field and no broken links between successive states. The first old state is carried back within the queried change window; it is not a screenshot taken on 13 June.
In that panel, AI wording appears in 52 listings at the start, peaks at 62 and ends in 55. That does not resemble a word that was absent and then became universal. What grows is narrower: the selected tool and channel names rise from 20 to 36, and approval, preview and undo wording from 17 to 27. The two largest falls are among the broadest claims in the codebook: 24/7 and instant wording 32 → 24, automation 68 → 61.
We defined these 15 families after reading the corpus (certification wording never appears), and that many will produce some rises and some falls by chance, so the panel describes these 146 listings rather than testing a trend. What it does show is that the rises and the falls point the same way, toward what can be named and away from what can only be asserted. AI, the broadest word of all, is the exception that keeps this modest: it ends 3 listings up on where it started.
| Literal vocabulary family | 13 Jun opening | Through Jun | Through Jul | Through Aug | Through 20 Sep |
|---|---|---|---|---|---|
| AI wording | 52 | 52 | 59 | 62 | 55 |
| Agent / agentic | 9 | 8 | 10 | 10 | 9 |
| AI discovery vocabulary | 13 | 17 | 18 | 17 | 13 |
| Automation / autopilot | 68 | 60 | 57 | 59 | 61 |
| No-code wording | 24 | 27 | 31 | 28 | 26 |
| Free entry / service | 4 | 4 | 3 | 7 | 6 |
| Approval / preview / undo | 17 | 19 | 21 | 29 | 27 |
| Compliance / named standards | 7 | 6 | 5 | 5 | 5 |
| Profit / margin | 10 | 12 | 15 | 13 | 10 |
| 24/7 / instant | 32 | 36 | 31 | 28 | 24 |
| Stocky | 1 | 1 | 2 | 3 | 3 |
| Black Friday / BFCM | 0 | 0 | 0 | 1 | 3 |
| Named tools / channels | 20 | 23 | 29 | 31 | 36 |
| Sidekick | 0 | 0 | 1 | 3 | 4 |
| Certification wording | 0 | 0 | 0 | 0 | 0 |
Two other instruments read the same way. Body-only measurement gives a slightly larger panel, and AI goes from 42 to 44 of 153 apps. Across all apps' individual first/last body pairs it goes from 57 to 63, with 22 entering and 16 leaving; that comparison spans different dates and includes newer apps, so it is an endpoint inventory rather than a calendar trend. Three cuts, one reading: AI language churns in both directions with a small net change.
These families measure language, not tested integrations or safeguards, and a counter cannot read. AddOn Suggest adds AI to say it avoids “AI guesswork.” Green Claims Guard's certifications are documents the merchant can retain, not necessarily a certification of the app. Approval can refer to an order, an outfit recommendation or a catalog write. Instant can describe an alert, a setup step or a customer-service response. There is no defensible count of newly certified apps or guaranteed response times here, which is why the codebook is called a lexical inventory.
AddOn Suggest: Product Upsell · Essenify
AddOn Suggest: adding the word AI to argue against it
The early pitch says the merchant hand-picks add-ons “not an algorithm.” July 30 then adds “No fancy AI Algorithms to mess up your product catalogue.” August 6 sharpens that distinction into a headline with “no AI mess” and a body promising recommendations “without AI guesswork or bundle discounts.” This is opposition to an AI-led proposition.
Before
Your customers are ready to spend more, they just need the right product recommendations. Hand-pick add-on productsandaccessories as cross-sell upsells on any product page, like a frequently bought together widget, but fully under your control. A clean,native app block showsyourpicks instantly.You choosetheadd-ons, not an algorithm. Nocode,no developer,live in minutes.Works withallOnline Store 2.0 themes,zero page impact. Great for boosting AOV with natural recommendations.
After
Maximise AOV and Product Visibility with simple single-click add-ons, no complicated bundle offers to eat your margin. Just simple showcasing of which product has the best relevant product matches available. No fancy AI Algorithms to mess up your product catalogue. Simple for merchants to set up, Easy for customers to buy, Addons that really work without fancy product bundles. Addons that work for all categories that you can think of: subscriptions, gifts, warranties, shipping protection
By August 12 the AI reference has disappeared again. The body describes product and cart placements, Plus-only checkout placement, no-code setup and clean uninstall. In September it becomes even more concrete: shoppers select add-ons and their variants and add everything in one click. The contested word was a temporary contrast; the persistent promise is merchant control and a simple buying action.
Before
Maximise relevant product visibility with product recommendation addons to increaseAOV
After
Free, one-click upsells, add-ons & cross-sells that boost AOV — no bundles, no AI mess.
Before
No code:letcustomers pick & addmultiple add-onsto cart fromtheproduct page. Free plan.
After
No code: customers pick add-on variants & add to cart from product page. Free plan.
Reviews around its first retained edit, 2026-06-26. No dated reviews on either side.
Before 2026-05-29–2026-06-25; after 2026-06-27–2026-07-24; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. This app is excluded from the incident-AI outcome group because the added wording rejects AI. It remains in lexical mention counts, where negation is deliberately not interpreted. The distinction is why the close reading matters.
English expressions also miss untranslated equivalents. The Japanese invoice listing changes a claim of reducing accounting work to zero into a claim of reducing it, while moving free-plan language and named accounting systems into its tagline. Those changes are visible in the archive but are not all captured by the English regular expressions. No multilingual semantic prevalence claim is made.
An average over 146 listings hides where the movement is concentrated. The family that grew was named things. The next step follows one name through the listings that used it.
Step 6 · A campaign with a calendar
One name, several promises: the Stocky migration
Stocky was Shopify's own inventory and purchase-order app, and Shopify's migration documentation confirms it was retired on 31 August. In the fixed panel the name goes from 1 listing to 3 of 146, a negligible share. Followed through every captured field of every cohort app, including apps that arrived after the panel was fixed, nine listings name it at some point. Four already do in June, more than two months ahead; three add it in July, one in August, and the last, Restock Forecast, two days after the retirement. That is a campaign with a calendar, run across the whole approach to one date, even though it never reaches the cohort as a whole.
The rival reading is a shared template: several developers copying the same “Stocky alternative” line. The sentences rule that out. Prediko adds replacement language in July and a migration mechanism in August. Alfred offers to run alongside Stocky. Binly moves from a deadline and assisted setup to API history and then to saved exports. Replenio names the relationships it claims to reconstruct from purchase-order files. The competitor's name is common to the pitch; the proposed work differs. Shopify's documentation corroborates the calendar; it does not verify any of these migration promises.
Prediko Inventory Management · Prediko
Prediko: a Stocky migration campaign, inside a much longer edit sequence
The June 24 event anchoring the existing outcome analysis is mostly a rewrite of explanation: ampersands become “and,” “POs” becomes “purchase order,” “BOMs” becomes “Bill of Materialss,” and two emoji disappear. The misspelling is then corrected in a small June 26 row that did not qualify as a substantial copy edit. AI inventory forecasting was already in the old body. The ratings comparison is not a before/after test of newly adding AI.
Before
AI inventory forecasting, planning&purchase order management in one app📈. Save time&replace sheets with advanced inventory management that syncs with Shopify in real time. Get automated inventory alerts&replenishment recommendations to createPOs in secondsand avoid stock outs or over stock. Track raw materials, manageBOMs, bulk edit SKUs or suppliers,and export reports(20+ templates) 📄.The transferfeature optimisesyourinventory by balancing stock across multiple locations.
After
AI inventory forecasting, planning and purchase order management in one app. Save time and replace sheets with advanced inventory management that syncs with Shopify in real time. Get automated inventory alerts and replenishment recommendations to create purchase order and avoid stock outs or over stock. Track raw materials, manage Bill of Materialss, bulk edit SKUs or suppliers and export reports. Inventory Transfer feature optimises inventory by balancing stock across multiple locations.
July 11 adds “Replaces Stocky” at the end. July 15 adds ERP and vendor terminology; July 18 turns a transfer function into something that “replaces Stocky,” then adds “Stock Take coming soon.” On July 30 that promise becomes “Stock Take & Barcode Label is now live.” The log can distinguish a future-feature promise from a later availability claim. It cannot test the released feature.
Before
Inventory Transfer feature optimises inventory by balancing stock across multiple locations.
After
Inventory Transfer feature optimises inventory by balancing stock across multiple locations - Replaces Stocky.
By August 18, replacement is no longer a trailing keyword. It moves to the second sentence and names the work of switching: “suppliers and purchase orders migrate automatically.” The feature bullets also make migration explicit. The useful reading is a campaign aimed at an inventory user facing a move, with successive attempts to specify what comes along.
Before
AI inventory forecasting, planning and purchase order management in one app.ERP with advanced inventory management that syncs with Shopify in real time.Get automatedinventory and replenishment alerts to create purchaseorderand avoid stockoutsorover stock. Track raw materials, manage Bill of Materials, bulk edit vendors, SKUs or suppliersand export reports.Inventory Transfer feature for stock management across multiple locations replaces Stocky.Stock Take & Barcode Labelisnow live.
After
AI inventory forecasting, demand planning and purchase order management in one app. Inventory Transfers move stock across multiple locations and replace Stocky: suppliers and purchase orders migrate automatically. Automated inventory and replenishment alerts to create purchase orders and avoid low stock or overstock. Track raw materials, manage Bill of Materials and export inventory reports. Stock Take & Barcode Label now live. ERP with inventory management that syncs with Shopify in real time.
The original rating average rises while the review count halves. The Stocky mention arrives on July 11, partway through the June 25–July 22 after-window. The live stocktake wording, migration sentence and August feature bullets all arrive after that window. Even a persuasive product-fit explanation would not make those later sentences the cause of the earlier rating difference.
Before
- AI Demand Forecasting:
LeverageAI to plan sales by product, SKU & season - Inventory Replenishment: Reorder with smart alerts & a dynamic buying table
SmartPurchase Orders:Create, track & bulk editPOs fullysynced with WMS- Raw Materials: Create bill of materials to track & forecast raw material needs
AdvancedInventoryReports:Export CSV/PDFwith100+ filters & 20+ templates
After
- AI Demand Planning & Forecasting: Use AI to plan sales by product, SKU & season
- Inventory Replenishment: Reorder with smart alerts & a dynamic buying table
- Purchase Orders: Migrate from Stocky, create, track & bulk edit synced with WMS
- Raw Materials: Create bill of materials to track & forecast raw material needs
- Inventory Transfers: Move stock across multiple locations with min/max top-ups
Reviews around its first retained edit, 2026-06-24. Dated reviews 14 → 7; mean 4.43★ → 5.00★ on 14 → 7 rated; visible at extraction 13 → 7.
Before 2026-05-27–2026-06-23; after 2026-06-25–2026-07-22; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. A smaller set of seven uniformly five-star reviews replaced a fourteen-review group with a lower average. The copy shows what the pitch became, but gives no reviewer-level link between that pitch and those reviewers’ experiences. “Probably why the stars rose” would overstate the evidence.
Binly ‑ Stocktake & Reorder · Prokop Studio
Binly: the migration method changes with the calendar
Binly already names Stocky in its first captured old copy. July 3 adds a deadline, a $19 replacement pitch and free white-glove migration. This is not a new invention of the competitor keyword; the deadline turns an existing replacement proposition into a timed offer.
Before
Count stock,raisePOs,and reorder from your phone. Leaving Stocky? Import your data in minutes.
After
Stocky shuts down Aug 31, 2026. Binly is the $19 replacement: phone stocktakes, POs, free migration.
August 28 promises an API import of suppliers and full purchase-order history, including receive dates and costs. The next day stresses that this is “not a partial CSV.” On September 2 the pitch switches to the CSV exports the merchant saved. The concrete object changes from an API key to exported files. The dates and verbs fit a migration campaign more closely than a generic increase in AI language.
Before
Stocky shuts down on August 31, 2026 — Binly is the affordable replacement: Stocky's essentials without the enterprise price or extra hardware.Open a stocktake and scanwith the phone in your pocket; counts and variancein units and dollarsappear live. Raisepurchase orders, receive deliveries, and reorder before you sell out. Transfer stock between locations, print barcode labels, and find dead stock.Free white-glovemigration:email usyour Stockyexportandwe set everything up.
After
Stocky shuts down on August 31, 2026 — Binly is the affordable replacement: Stocky's essentials without the enterprise price or extra hardware. Scan stocktakes with the phone in your pocket; counts and variance appear live. Count a shelf or one supplier at a time. Raise POs, receive deliveries, and reorder before you sell out — supplier minimums applied. True migration: paste your Stocky API key and Binly imports suppliers and full PO history with receive dates and costs.
By September 3 the opening still sells continuity of history, but the tagline gives space to month-end COGS and QuickBooks/Xero. The product is now being explained in terms of the workflow after moving. Shopify’s migration documentation corroborates the retirement date; it does not validate Binly’s import completeness.
Before
StockyclosesAugust 31.Binlybringsyournumbers with it:paste your Stocky API key and it imports yoursuppliersandfull PO history, receive dates and costs, not a partial CSV. Count stock with your phone, live variance in units and dollars. Raise and receive POs. Reorder from realsalesvelocity andrealsupplier lead time, with the date to order by. Weighted-average and landed costkeep margins honest,a late invoice over the order is flagged,andmonth-end COGSclosesthebooks.
After
Stocky shut down on August 31 — Binly keeps your history alive. Import the CSV exports you saved: suppliers, full PO history, receive dates and costs. Count stock with your phone. Raise and receive POs, print barcode labels for what arrived — Dymo file or from the browser — and generate EAN-13s for items without one. Reorder from real velocity and supplier lead time. Weighted-average and landed cost, late-invoice flags, month-end COGS. Ask the AI assistant — it answers from your own numbers.
No eligible first-event comparison. The app lacks a complete observed baseline or postperiod. It has 1 unedited dated review during the screen.
What this cannot show. Binly was first observed too late for the original first-event baseline. Its short history must not be made to look successful by comparing new reviews to an invented zero preperiod.
Alfred Purchase Orders & Stock · Alfred Commerce LLC
Alfred: replacement, parallel running, then scoped stocktakes
Alfred’s July 9 rewrite puts “Stocky is shutting down” first and offers to run alongside it. The old body had already described forecasting, POs, stocktakes, transfers and alerts. Much of the functional vocabulary survives; the reason to choose those functions now is new.
Before
Forecastdemand. Streamline purchase orders generation. Receive& transferinventory with ease.
After
The purpose-built Stocky replacement: forecast demand, plan POs, receive & transfer stock.
August adds API migration and then a detailed stocktake safeguard: zeroing stays within the chosen count scope, and uncounted items can be reviewed before they are zeroed. That is a specific risk-reduction promise for a merchant moving an inventory process. On September 3, “shuts down” becomes “shut down.” The calendar moves inside the sentence.
Before
Stockyis shuttingdown. Alfred is a purpose-built replacementthat does more. Forecast demand from real sales velocity with stockout dates. Build purchase plans by demand, min/max or manual, then email or export(PDF/CSV)purchase orders.Manage suppliers,receive by barcode,manualorPO/shipment quantity,run stocktakes, andlog credit memos.Transfer and rebalancestockacross locations,with low-stock and inventory alerts. MigrateStocky data via APIand run in parallel until the deadline.
After
Stocky shuts down August 31. Alfred is a purpose-built replacement. Forecast demand from real sales velocity with stockout dates. Build purchase plans by demand, min/max or manual, then email or export purchase orders. Run scoped stocktakes: count specific products, a product set, or a whole location — zeroing stays inside the scope you chose. Review uncounted items before any are zeroed, count blind, and value variances at cost. Receive by barcode, transfer stock, migrate Stocky data via API.
Reviews around its first retained edit, 2026-06-17. Dated reviews 1 → 2; mean 5.00★ → 5.00★ on 1 → 2 rated.
Before 2026-05-20–2026-06-16; after 2026-06-18–2026-07-15; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The first-event comparison of one review to two is too small for a rated comparison under the report’s minimum. Later copy changes cannot turn that into evidence of a successful migration campaign.
Replenio — Purchase Orders · Veltrio Labs LLC
Replenio: the history left behind becomes the product
The July copy sells purchasing, FIFO or weighted costing, freight and duty allocation. It refers indirectly to “a retiring purchasing app.” August 9 names Stocky and leads with the problem of retaining supplier relationships and historical purchase orders. The merchant is no longer asked to buy a general purchasing loop; the first job is to recover the structure inside an export.
Before
Shopify gives every variant one static cost per item. Repleniokeepsthereal one:every receipt recalculates unit costwith FIFO or weighted costing,andfreight, dutyandfees areallocated across the lines that carried them. It runs on your own purchasing data: purchase orders, supplier records with lead times and minimum order quantities, barcode receiving, stock takes. Switching from a retiring purchasing app? CSV import rebuilds suppliers, unit costs and reorder points in minutes.
After
Stocky closes on 31 August 2026. Shopify's transition guide says historical purchase orders can't be imported, and that suppliers can't be exported from Stocky at all — only the purchase order export still carries them. Replenio reads that export and rebuilds your suppliers, product links and dated unit cost history. Take it before the shutdown. After that, every receipt recalculates unit cost, with freight and duty allocated across the lines that carried them.
On August 20 the statement about suppliers changes materially. The earlier sentence says suppliers cannot be exported at all; the new version says the supplier export does not identify which product came from which supplier. The claimed missing object is now the product–supplier link. That is a correction in the explanation, not an interchangeable synonym. Neither version is adopted here as a verified technical fact about Shopify.
Before
Stocky closes on 31 August 2026.Shopify's transition guide says historicalpurchase orders can't be imported, andthat suppliers can't be exportedfromStocky at all— only the purchase order exportstill carriesthem. Replenio reads that export and rebuilds your suppliers, product links and dated unit cost history. Take it before the shutdown. It also tracks what each supplierhasinvoiced, what's paid,and what's overdue. Every receiptthenrecalculates unit cost, freight and duty included.
After
Stocky closes on 31 August 2026. Historical purchase orders can't be imported into Shopify, and Stocky's supplier export never says which product came from which supplier — only the purchase order export pairs them on every line. Replenio reads that export and rebuilds your suppliers, product links and dated unit cost history. Take it before the shutdown. It also tracks what each supplier invoiced, what's paid and what's overdue. Every receipt recalculates unit cost, freight and duty included.
This is competitive positioning through an information-loss problem. Prediko promises automatic migration, Alfred parallel running, and Replenio reconstruction from a file. The shared keyword conceals different offers.
No eligible first-event comparison. The app lacks a complete observed baseline or postperiod. It has 1 unedited dated review during the screen.
What this cannot show. Replenio has no eligible original first-event review comparison. These are recorded product claims and a change in their precision; the study did not run an import or inspect the application.
Named-tool language is being selected and displaced as well as added. Sidekick goes from 0 to 4 in the fixed panel. PageFly replaces a broad AI-audit sentence with a list of things to ask Sidekick for; Estimated Delivery Date adds Sidekick AI to its rules explanation. EZstock adds a Sidekick sentence on 20 August and removes it the next day in favour of Stocky migration. Black Friday/BFCM goes from 0 to 3 in the same panel: Slide Cart and Discounty put the seasonal job in their September taglines, preparation by a few relevant apps rather than a store-wide change of vocabulary.
Naming Stocky reaches a competitor's customers at the moment they have to move. The next edits reach for something one step further removed: the words a merchant might type while searching for a neighbouring product.
Step 7 · Competitors, categories and search terms
Words aimed at neighbouring searches
On 7 September Wiser changed one word in a sentence about scaling personalisation and added a comma: “easily” became “selleasy”. The frozen store catalogue contains an upsell app named Selleasy.
beforescale personalization
easily by cross sell & upsell bundles.
afterscale personalization, selleasy by cross sell & upsell bundles.
The inserted word does not describe a feature in the surrounding sentence, which is what separates it from the many edits that add a capability. It is consistent with targeting a neighbouring app's searches. The record cannot establish the author's intention, a ranking effect, or that the spelling was not an error: there are no search-volume, ranking or traffic records here with which to test that reading.
PIMW Product Personalizer shows the same pressure as a sustained pattern. Its body and tagline pass through 30 captured states, the first carried back from before 9 July and the last recorded on 19 September, and in every one the two fields together hold between 552 and 597 characters. The listing does not grow, and it cannot: across the cohort, 622 of the 1,022 captured bodies sit between 480 and 500 characters and only 4 run longer, the signature of a field Shopify caps at 500. A body written to the cap cannot gain a word without losing one, so each new name for the product (personalizer, customizer, configurator, builder, designer, product options) has to displace something already there. What the figure shows is how PIMW spends a fixed budget.
Orichi briefly lists other feed-product names in an integration sentence, then replaces the list with a feed-health explanation. These are direct observations of competitor and category language. Each is consistent with an attempt to be found for neighbouring searches; none can be tested as one here.
Wiser AI ‑ Upsell & Cross Sell · Expert Village Media Technologies
Wiser: category vocabulary, then a competitor-shaped word
Wiser repeatedly enumerates the places and mechanisms of an upsell: frequently bought together, cart drawer, checkout, post-purchase, related products, add-ons. August and September edits exchange near-neighbour phrases rather than merely shorten prose. A broad product is being named in the several ways a merchant might search for it.
Before
AI Recommendations with Frequently Bought, Related,InCart Upsell, Post Purchase Upsell&more
After
Personalized AI Recommendations with Frequently Bought, Related, Cart Upsell & Post-Purchase Upsells
The September 7 edit is unusually revealing. “Scale personalization easily by cross sell & upsell bundles” becomes “scale personalization, selleasy by cross sell & upsell bundles.” Selleasy is the name of another stored upsell app. The inserted word does not explain a feature in the surrounding sentence. It is consistent with search-term targeting, although the record cannot establish the author’s intention, a ranking effect or whether the spelling was an error.
Before
Track performance directly in your Shopify admin, access APIs for custom integrations, and scale personalizationeasilyby cross sell & upsell bundles.
After
Track performance directly in your Shopify admin, access APIs for custom integrations, and scale personalization, selleasy by cross sell & upsell bundles.
The June tagline removes AI and July brings it back. The observed title also loses AI briefly in September. A binary “ever added AI” flag would miss the movement between the technology label and category terms.
Reviews around its first retained edit, 2026-06-25. Dated reviews 5 → 2; mean 5.00★ → 5.00★ on 5 → 2 rated.
Before 2026-05-28–2026-06-24; after 2026-06-26–2026-07-23; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The original comparison has five reviews before and two after, all five stars; it fails the three-ratings-per-side rule. The Selleasy insertion is in September, long after that window. Competitor wording is observed; its commercial effect is not.
PIMW Product Personalizer · Pixelspiece solutions
PIMW: many names for the same customisation task
PIMW’s July tagline grows from product personalisation with live preview into POD, jewelry, apparel and furniture, then photo upload, engraving and custom text. The body repeatedly substitutes designer, personalizer, customizer, builder, options and configurator. That is a semantic neighbourhood, not a single keyword.
Before
Let customers personalize and customize productswithunlimited options andlive preview
After
Real-time product personalizer with live preview for POD, jewelry, apparel & furniture
The August 1 sentence “custom product personalizer customizer with instant live preview” reads like a cluster of searchable product labels. On August 23 it becomes a more concrete buyer action: buyers see their design before checkout; the tool sets dynamic pricing and conditional fields. September returns to repeated variants of product options and customization. The sequence alternates between describing use and naming the category.
Before
Live Productpersonalizerturns anyproductpage into a liveproductcustomizer. add infinite variant options with Image, text, fonts, color swatches, dropdowns and checkboxes.Shoppers watch alive previewupdate instantly on the product image. Built for print on demand, engraving, monogram, custom gifts andpersonalised jewelry. Set option pricing and price add-ons, use conditional logic to show or hide options, and pass every personalization to your order as line item properties.
After
live preview product personalizer for custom product with product Builder. add infinite product options variant options with Image, text, fonts, color swatches, dropdowns and checkboxes. custom product personalizer customizer with instant live preview . Built for print on demand, engraving, monogram, custom gifts and personalized products. Set option pricing and price add-ons, use conditional logic to show or hide options, and pass every personalization to your order as line item properties.
The claim of an AI product-option generator appears in an August 3 body, surrounded by a long product-options list. Treating that one phrase as the whole intervention would discard the much larger rewrite around it.
Before
Sellcustomisedproductslivecustomers design themselves.real timeproductcustomizer. add custom text,monograms,engraving &photouploads. Unlimited product variant live image & color swatches, dropdown menus,buttons,fonts & uploadanddesignfields. productoptions &customizer with conditional logic& dynamic pricing for every personalized add-on. For engraved jewelry & print on demand.Turnproduct personalizationinto more sales & easy custom orders with product options visualizer
After
Sell custom products your customers design themselves. Add unlimited product options variants options: engraving, monogram text, photo upload, swatches, dropdowns and custom fields. Buyers see their design in live preview before checkout. This product customizer sets dynamic pricing, shows or hides fields with conditional logic, and builds option sets in seconds with AI. From personalized gifts to laser engraving, product personalization live customization that convert. No code.
Reviews around its first retained edit, 2026-07-09. Dated reviews 3 → 1; mean 5.00★ → 5.00★ on 3 → 1 rated; visible at extraction 2 → 0.
Before 2026-06-11–2026-07-08; after 2026-07-10–2026-08-06; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The original first-event comparison falls from three reviews to one. A new window aligned to the August AI mention rises from one to five, which is still too small for the rating rule and overlaps further copy and plan edits. Neither window isolates the keyword.
OC Multi Product Feed LLMs.txt · OrichiCommerce
Orichi: integrations and neighbouring brands in the same sentence
The July 8 body ends “We integrate with Orichi product feeds, Simprosys Google Shopping feed, Datafeed, Nabu, CED , multifeeds.” The list mixes the app’s own name with names and phrases associated with other feed products. A lexical integration count cannot establish what these connections actually do.
Before
Establish unlimited XML data feeds to seamless integrate with Facebook Product Feed, Instagram Shop, Pinterest Catalog and Bing Shopping Feed(Microsoft Shopping). Offer robust customization capabilities across all data to align with diverse catalogue ads(catalog ads). Maintain real-time data updates to ensure smooth synchronize between your Shopify store and multiple social platforms. We integrate with Orichi product feeds, Simprosys Google Shopping feed, Datafeed, Nabu, CED , multifeeds
After
Establish unlimited XML data feeds to seamless integrate with Facebook Product Feed, Instagram Shop, Pinterest Catalog, Chatgpt ads and Bing Shopping Feed. Offer robust customization capabilities across all data to align with diverse catalogue ads(catalog ads). Maintain real-time data updates to ensure smooth synchronize between your Shopify store and multiple social platforms. We integrate with Orichi product feeds, Simprosys Google Shopping feed, Datafeed, Nabu, CED , multifeeds
By July 24 that sentence is gone. The space instead explains feed-health errors and in-app fixes; the headline still ranges across named sales and advertising destinations. “LLMs.txt” is added to the tagline on July 29. This is a move among search vocabularies and a more specific operational claim, not evidence that every named destination or peer app is an implemented integration.
Before
Establishunlimited XML data feedsto seamlessintegrate with Facebook Product Feed, Instagram Shop, AWIN, Pinterest Catalog,Chatgpt ads andBing Shopping Feed.Offer robust customization capabilities across alldata toalign with diverse catalogue ads(catalog ads).Maintain real-time dataupdatesto ensure smooth synchronize betweenyour Shopify store andmultiple socialplatforms.We integrate with Orichi product feeds, Simprosys Google Shopping feed, Datafeed, Nabu, CED,multifeeds
After
Create unlimited XML data feeds that integrate seamlessly with Facebook Product Feed, Instagram Shop, AWIN, Pinterest Catalog, ChatGPT Ads, Bing Shopping Feed and more. Customize every data field to match the requirements of different catalog ads. Real-time updates keep your Shopify store and all connected platforms in sync automatically. Feed Health flags errors and warnings before platforms reject your products, fix them in-app.
Before
Create multiple product feeds forFacebook-Insta, Pinterest, Google Shopping, Bing, OpenAI, AWIN...
After
Create multiple product feeds for Facebook, Pinterest, Google Shopping, Bing, OpenAI, AWIN, LLMs.txt
Reviews around its first retained edit, 2026-07-09. Dated reviews 0 → 3; mean — → 5.00★ on 0 → 3 rated.
Before 2026-06-11–2026-07-08; after 2026-07-10–2026-08-06; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. A listing can name an ecosystem without proving a tested integration. This report preserves the words and the removal, and makes no allegation about their truth or origin.
Cleverific Order Edit & Upsell · Cleverific, Inc
Cleverific: AI in the body while it leaves the title
The July 2 change reframes order editing as both a support-saving tool and a revenue surface. AI upsells are said to be trained on the merchant’s own sales data, add into the original order, and cost money only when an upsell converts. The same-day bullets add automatic winner detection for offer tests. This is the app advertising an A/B-testing capability; it is not evidence that its developer A/B-tested the listing.
Before
With Cleverific,yourcustomerscanmake self-serve changeswithout contactingsupport.Easily handle any other order edits such as editingbillingaddress,exchanging items,adjustingproduct options,addingcustomizations,or modifyingtaxes. Simplify your workflow with a single solution to edit orders,offer self-serve order editing, automateordereditingwithFlow, recover abandoned carts, and enhance draft orders forsales, custom orders, wholesale orders, backorders,preordersandmore.
After
With Cleverific, customers make self-serve changes to their orders and support tickets disappear. Edit anything else in seconds: billing addresses, exchanges, product options, customizations, taxes, and more. And every order becomes a revenue opportunity with AI upsells trained on your real sales data and customer behavior, converting in one click right after checkout and landing in the original order. You only pay when an upsell converts.
The July 24 body removes the AI framing and spells out edit controls and personalised post-purchase recommendations. August further explains that a separate order or checkout is unnecessary. In September, AI returns to the body just as it leaves the observed title. Different surfaces are doing different jobs: category clarity in the title, a technology claim in the longer explanation.
Before
- Let customers edit
address,add productswith self-serve order editing Order editing forproduct options, shipping, billingaddress, pricing, taxesOrder editing automations to add gifts with purchase or split bundlesCraft detailed custom orders & enhanced invoicing using draft ordersEdit ordersandsend payment requests in one step, recover abandoned carts
After
- Let customers edit addresses, swap items & more with self-serve order editing
- Edit any order detail: product options, shipping, billing, pricing, taxes
- AI builds upsell offers from your real sales data and customer behavior
- One-click post-purchase upsells add products to the original order
- A/B test offers with automatic winner detection and upsell revenue analytics
Before
Save time with self-service order editing & increase average order value with personalized post-purchase upsells.
After
Save time with self-service order editing & increase average order value with personalized AI post-purchase upsells.
Reviews around its first retained edit, 2026-07-02. No dated reviews on either side.
Before 2026-06-04–2026-07-01; after 2026-07-03–2026-07-30; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The original comparison contains zero reviews on both sides. There is no review trajectory here with which to distinguish the competing pitches.
Categories and titles move too
The category record contains 33 primary-category transitions across 29 apps. Stockup's move from inventory alerts to order splitting accompanies a move into Order editing. PIMW moves from Custom file upload to Product variants; Arvio from Analytics to Workflow automation. The text helps explain why a category might change. We cannot tell whether the developer chose the assignment or Shopify changed it.
Nested category-feature records need more caution. The capture often represents labels as chains of nested keys, so we compare the set of labels as well as the object: 2 changed objects have exactly the same labels. Even where labels differ, reordering, capture behaviour and taxonomy changes remain alternatives to deliberate category jockeying. The archive publishes both states and the derived set differences.
The change detector has no title rule. Successful listing observations supply a shorter name history beginning 2026-08-04: 238 literal transitions across 111 apps, including punctuation and spacing changes. It does not extend the original editing-day screen.
| App slug | Detected | Previous observed title | New observed title |
|---|---|---|---|
| edit-order | 2026-09-11 | Cleverific AI Upsell & Editing | Cleverific Order Edit & Upsell |
| frequentlify | 2026-08-17 | Simesy Product FAQs & FAQ Page | Simesy Product Image Translate |
| frequentlify | 2026-09-12 | Simesy Product Image Translate | Simesy Product FAQs & FAQ Page |
| buddify | 2026-08-11 | Buddify: Partner Dropshipping | Buddify Community |
| buddify | 2026-09-06 | Buddify Community | Buddify:Wholesale Dropshipping |
| buddify | 2026-09-11 | Buddify:Wholesale Dropshipping | Buddify: Collective Commerce |
| promolock | 2026-08-27 | PromoLock: Block Discount Loss | PromoLock: Discount Blocker |
| promolock | 2026-09-06 | PromoLock: Discount Blocker | PromoLock: Block Discount Loss |
| promolock | 2026-09-07 | PromoLock: Block Discount Loss | PromoLock: Discount Blocker |
| nexus-ai | 2026-09-01 | Nexus AI | Shopicraft Build a Box BYOB |
| spotlight-try | 2026-08-29 | Spotlight | Spotlight Recommendations |
| spotlight-try | 2026-09-17 | Spotlight Recommendations | SizeMap |
When the instrument cannot tell which product it is watching
Some listings change more than their vocabulary. The Simesy FAQ listing temporarily describes a withdrawal-button product and later an image translator, and its observed title changes too. An exact cross-app text search did not find another change-log owner for either body. This supports an identity warning, not a claim that the developer invented unrelated features. Its cohort membership is preserved, and its incident claim events are excluded from the outcome analysis in step 10. Stockup and Spotlight show milder versions of the same problem.
Simesy Product FAQs & FAQ Page · Simesy
Simesy: when the text appears to describe another product
The June 25 change replaces an FAQ-and-tabs body with an EU-withdrawal-button description. Five days later the FAQ body returns, with a terminal full stop. On August 17 the same app ID receives an image-translation description, and September 12 restores FAQ copy again. These are changes of described product, not small expansions of a feature list.
Before
Organize product description into tabs and accordion. Reduce support with product FAQ and FAQ page.
After
Implement EU withdrawal button for EU Directive 2023/2673. Quick setup in few minutes without coding
The title observation stream also changes from Simesy Product FAQs & FAQ Page to Simesy Product Image Translate on August 17 and back on September 12. That corroborates what this collection pipeline saw in more than one field. It is not an independent page capture and cannot rule out an association or source-page problem.
Before
Enable your customers to help themselves with dedicated FAQ page. Display FAQ onproductpage or anywhere (homepage, CMS page, collection page...) soyourcustomers can find the information they need.Many options to customizethe looktomatch your brand design. Fully responsive and auto adapt based on device width. Organizeproductdescription into tabs and accordionautomaticallyusing heading tags. Create shared product tabsandaccordions across multiple productsandcollections.
After
Simesy Product Image Translate lets you translate product image files into multiple languages. Localize product images for each market by uploading your own translated images. Use AI to translate product images automatically while preserving style and font matching. The app integrates with your store, works with any theme, and requires no coding.
An exact-text search across the whole production change log found these two bodies only under this app ID, in their arrival and departure rows. That check did not locate a second listing to which the text could confidently be reassigned. The cause remains unresolved.
Before
Translate product images&alt text into other languages.Autoproductimages translatwithAI
After
Reduce customer support with Product FAQs & FAQ Page. Oganize product page with tabs & accordions
Reviews around its first retained edit, 2026-06-25. No dated reviews on either side.
Before 2026-05-28–2026-06-24; after 2026-06-26–2026-07-23; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The app stays in the original verified cohort and in literal wording inventories. Its incident keyword changes are excluded from the new outcome study. Calling these edits opportunistic compliance or AI marketing would attribute a motive before establishing that the listing was really the intended product.
Order Splitting & Fulfillment · Stokow
Stockup: the named job changes from alerts to order splitting
June copy describes low-stock alerts and reorder quantities. On July 8 it becomes “AI Split Order,” with rules for splitting orders and routing shipments, simulations against past data and a claim of zero errors. A July 18 primary-category change from Inventory optimization to Order editing follows the new described job.
Before
Stockup helps you eliminate stockoutsbysending instant low stock alerts. Set up custom low stock alertrulesfor specificlocations, vendors,collections,orvariants in seconds.Choose between real-time low stock notifications or scheduled summaries sent to your team.With smart reorder calculations, each low stock alert tells youtheexact quantities neededtoreplenish your inventory,saving you time on manual tracking.
After
AI Split Order streamlines your fulfillment by automatically splitting orders into separate orders based on your workflow rules. Effortlessly split items and route shipments to different locations using tags, vendors, or SKUs. Our intelligent order routing works 24/7. AI analyzes your recent order history to recommend the best presets. You can safely simulate new rules against past data before activation to ensure precise automation with zero errors, making multi-vendor fulfillment simple.
The August title stream starts with AI Split Orders & Fulfillment, then drops AI. Taken together, the body, category and title form a more coherent change of positioning than a stray keyword edit. The persistent slug still says stockup, so a present-day name alone would conceal the earlier promise.
Before
Inventory optimization
After
Order editing
Reviews around its first retained edit, 2026-06-13. No dated reviews on either side.
Before 2026-05-16–2026-06-12; after 2026-06-14–2026-07-11; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The product may have pivoted, been repurposed, or changed how it presented an expanded feature set. The log does not establish which. Zero reviews on both sides of the original anchor cannot choose among those explanations.
SizeMap Size Charts & Guides · XCI labs
Spotlight to SizeMap: a different customer problem under one slug
June through August describes discovery and recommendation feeds. September 17 replaces that with a visual layer over a store’s size chart. The observed title becomes SizeMap on the same date. The before/after is too large to read as a variant of a single sales sentence.
Before
Spotlightadds arecommendation sectionto yourproduct pages that keeps showing relevant products as shoppers explore.It automatically curates what comes next usingyourproduct images and product details, keeping recommendations connectedtowhat the shopper is viewing.Products continuetoappear until your catalog runs out,without manual collections or discovery paths. Spotlight works alongside your existing storefront.
After
SizeMap adds a visual layer to the size chart already on your store. Keep your existing measurements while giving shoppers a clearer way to understand them. Map measurements to visual guides, show selected-size values, compare sizes, and track guide views and size selections in the Report.
The old body tries to extend browsing after a product page; the new one tries to make measurements understandable before a purchase. Both may be aimed at conversion, but they intervene in different buyer problems. Keeping them in the same app-level series is correct for the database ID and potentially misleading for a product-level effect study.
No eligible first-event comparison. The app lacks a complete observed baseline or postperiod. It has 0 unedited dated reviews during the screen.
What this cannot show. This example motivates retaining exact histories and identity caveats. It is not evidence of a false feature claim; the study did not inspect the running product before and after.
Search vocabulary makes an offer findable. The remaining edits change what the offer is: how much of the work the app claims, and what the merchant keeps control of.
Step 8 · What the offer promises
Narrower, inspectable promises
The approval, preview and undo wording that grew in the fixed panel has a shape in the sentences. ListSmith leads with a before/after diff and approval. AccessComply retreats from preventing lawsuits to supported checks and reversible changes. SoftCookies makes a person supervising the automation part of the offer. Arvio writes down which small catalogues do not need it. In each, the promise becomes more specific about what the merchant can inspect, undo or hand to a person, which is a different direction from a bigger claim.
Four examples chosen after reading are an illustration, not a count. The count is step 5's, and it was taken over a fixed panel rather than chosen: approval, preview and undo wording 17 → 27, while 24/7 and automation claims fell. The readings below show what that number is made of.
Arvio: AI Store Operator · Adot Technologies Inc
Arvio: agency, approval and the cost of asking
Arvio’s early proposition is an operator that does work instead of handing over a report. Approval, before/after previews and undo are already present. Several July taglines switch between bulk-edit vocabulary, a daily manager and an optimiser that finds lost sales. The recurring promise is relief from identifying and performing catalog work.
Before
Arviois an AI operator that runs the busywork of growing yourstore.Ask in plain words,or let it check your store daily and hand you a ranked to-do list of what's costing you sales: weak titles,missing SEO,mispricedorunsellable products, risky discounts, slow stock.It drafts the actual edits — in bulk — analyzes best-sellers and competitors, andcan run Klaviyo campaigns. You approve every changebeforeit goes live, nothing is ever deleted,andany edit is one click to undo.
After
Arvio gives Shopify merchants a natural-language workspace to investigate store issues and review available proposed actions. Store Path Diagnosis can show issues, supporting evidence, check time, and areas Arvio could not check. Use Recheck or View all issues to continue investigating. Where an action is available, inspect its content and action-specific status before confirming. Available actions and reversal behavior vary by workflow. Review AI output before confirming changes.
The August 3 body briefly describes a more bounded investigative workspace, with areas it could not check and reversal behaviour varying by workflow. The following day returns to the stronger claim that any edit is one click to undo. Reading the raw sequence exposes that oscillation; the original two-day spacing rule excludes some of it.
Before
Arvio is an AI store operatorforstore SEO, prices, stock and content.Ask it to rewrite product descriptions, fix weak titles,alt text, metafields, schema and blogs, or run bulk product edits across your whole catalog.It reads more than Shopify—shopper behaviour on your storefront,plus connected email and ads—so a dead-looking product may be bot traffic. Each day itrankswhat costsyouprofitanddrafts fixes. Approveonechangeorthe wholecatalog; nothing goes live withoutyou.
After
Priced below cost. Can't be bought. Titles nobody searches for. No meta description. Empty alt text. You don't have hands to check all of it. Arvio does — every product, every night, with nothing for you to ask or prompt — then ranks what's costing you sales most and writes the actual fix for each one. You approve or reject them one by one, and any edit undoes in one click. If your catalog is small and rarely changes, you don't need us.
The September 5 body says a small, rarely changing catalogue does not need the app. That is an explicit suitability boundary. September 9 then qualifies approval: price changes and other key edits wait, where earlier copy said every change waited. This is a change in the described division of work, not merely a softer synonym for automation.
Before
Priced below cost. Can't be bought. Titlesnobody searches for. Nometadescription. Emptyalt text.You don't have hands to checkall of it. Arvio does—everyproduct,every night,with nothing for you to ask or prompt—then ranks what's costingyousales most and writestheactual fix for each one.You approve or reject them one by one, andany editundoes in one click. If your catalog is small and rarely changes, you don't need us.
After
Your catalog is full of small problems nobody has time to find: titles nobody searches for, missing meta descriptions, empty alt text, products priced below cost, products nobody can buy. Arvio ranks all of it by what it's costing you, then writes each fix — product titles, descriptions, ecommerce SEO, alt text — across your whole Shopify catalog. In the morning you skim the list. Price changes and other key edits wait for your approval, and anything Arvio does undoes in one click.
Reviews around its first retained edit, 2026-06-18. Dated reviews 4 → 0; mean 5.00★ → — on 4 → 0 rated; visible at extraction 3 → 0.
Before 2026-05-21–2026-06-17; after 2026-06-19–2026-07-16; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. Arvio’s high activity, changing price ladder and original four-to-zero review comparison remain as established. The copy describes a search for the right scope and trust promise; the observation record does not establish which operational guarantee the app implemented.
ListSmith · S.E.A. Enterprises LLC
ListSmith: a safeguard becomes the opening sentence
The July tagline adds approval before publishing to a content-generation pitch. August brings the safeguard to the front of the body: a before/after diff, approval and one-click revert. Flat pricing and no theme-code injection sit next to the control promise. The proposed reason to choose the product is the cost of an unwanted edit as much as the quality of generated text.
Before
Bulk-generate on-branddescriptions, SEO, and alt text. One-click revert on every change.
After
AI product content with a safety net: descriptions, SEO, and alt text you approve before publish.
The later body adds AI-shopping data audits and a free monthly allowance while retaining the safety opening. This bundles three reasons to try it: discover a problem, inspect a proposed correction, and limit the cost of experimenting.
Before
ListSmithwrites and maintains product content at scale: descriptions, SEOmetatitles anddescriptions,andimage alt text. It adds a safety net: a clear before/after diff on every changeandone-click revert of any bulk edit. It stays on brand by learningyour voice from yourexistinglistings, and uses vector search to surfaceduplicate and thinproductcontent. Built for catalogs of 50 to 5,000 productsand fully embeddedin yourShopifyadmin.
After
ListSmith can't wreck your store: every AI change shows a before/after diff, applies only when you approve it, and reverts in one click. No code is injected into your theme, and pricing is flat — no credits, no overages. It writes descriptions, SEO titles and meta, image alt text, and internal product links; learns your voice from your listings; finds duplicate and thin content; and audits alt text for EU Accessibility Act readiness. Built for catalogs of 50 to 5,000 products, in your admin.
No eligible first-event comparison. The app lacks a complete observed baseline or postperiod. It has 0 unedited dated reviews during the screen.
What this cannot show. The guarantee that it “can’t wreck your store” is quoted as marketing language, not endorsed as a tested property. The review record is too thin to assess that promise’s effect.
AccessComply: SEO, Speed & ADA · AccessComply
AccessComply: from lawsuit prevention to an inspectable change
The old July tagline promises to stop lawsuits. The July 21 replacement offers supported checks, approval, saved originals and rechecking, and explicitly says the results do not prove conformance or legal outcomes. The promise becomes narrower and more inspectable.
Before
Stop ADA,EAAandBFSG lawsuits. Real WCAG 2.2 fixes in your theme code. Free scan + 3 freefixes.
After
Find SEO, speed, and accessibility issues; preview fixes safely and keep saved originals.
August moves between legal-standard names and the practical symptoms of a storefront: slow pages, missing information, blocked shoppers. WCAG and EAA references reappear on August 10, then leave the body again on August 18. Standards vocabulary is being tested against a purchase-friction explanation; it is not spreading monotonically.
Before
AccessComplyprotects your Shopify store from ADA,EAA (incl. Germany's BFSG),AODAandSection 508 lawsuits. We scan against WCAG 2.1 + 2.2 AA fortheissues that get stores sued: missing alt text, low contrast,broken headings,unlabeled buttons,keyboard gaps.Unlike overlay widgets, fixes are written directly into your theme code: backed up,verified by a re-scan,reversible. Start free: 3 scans + 3 real code fixes per month,no credit card. Scheduled monitoring catches new issues early.
After
AccessComply checks reached storefront pages for supported SEO, speed, and accessibility issues. Results stay separated so you can see what each check measured. Review supported theme or content changes before approval; the app saves the original and checks the result afterward. Schedule scans by plan, download Growth Scanner PDFs, publish an accessibility statement, and send work to a developer. Results do not prove rankings, field Core Web Vitals, conformance, or legal outcomes.
The final September body makes the workflow concrete: each issue links to its page; the merchant previews and approves a fix; the original is saved and the page checked again. It also says “not an overlay” in the tagline. That locates the claimed intervention in the theme rather than merely applying a general compliance label.
Before
AccessComply findswhatstopsapurchase on reached storefront pages:slowerpages,missingproductinfo, andblockedshoppers.Findings stay separated by check.You approve everyfix: no overlay widget,real correctionsto your theme, with theoriginal savedand each changere-checked afterit applies. AI drafts imagealt textandaccessiblelabels for your review.Schedulescans,downloadPDF reports,and publish anaccessibility statement.
After
One scan of your live store checks what a shopper hits: pages that load slowly on a phone, product pages search engines and AI agents cannot read, and buttons, images and menus some shoppers cannot use. Every issue is linked to the page that caused it. Preview each fix, approve it, and AccessComply writes it to your theme — original saved first, page re-checked after. AI drafts image descriptions and labels for your review. Scheduled scans, PDF reports, a publishable accessibility statement.
Reviews around its first retained edit, 2026-07-21. No dated reviews on either side.
Before 2026-06-23–2026-07-20; after 2026-07-22–2026-08-18; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. No conformance audit or legal assessment was performed for this study. The first-event review comparison is zero to zero. A more careful promise is visible in the text; a measured improvement in safety or conversion is not.
SoftCookies · Petru Marchidan
SoftCookies: the person supervising the automation becomes the offer
The July 26 body lists sync coverage, real-time records, matching and configuration. Within days it becomes a short account-manager promise, with almost all integration detail removed. August iterates that sentence from someone who fixes issues to someone who helps fix them. The certainty of the service promise changes.
Before
Seamless integration between Shopify andyouraccounting or invoicing system. It does not break,skip,nor duplicate transactions.
After
Focus on your business while our dedicated Account Manager administers, maintains, monitors, and optimizes your service from end to end.
On August 29 the technical explanation comes back underneath the person: QuickBooks, sales and refunds, duplicate prevention, logs and mapping. September’s tagline says a real human watches the automation; the body later names a dedicated bookkeeper. This is an integration being sold as supervised work.
Before
Focus on your business while yourdedicated account managerhandles setup,monitorsfor issues andhelps fixthem.
After
Your dedicated account manager sets things up, watches for issues, and fixes them. SoftCookies connects Shopify and QuickBooks, syncs sales and refunds in real time, matches products and customers, prevents duplicates, and keeps a detailed import history log so you can see what was synced and when. Smart mapping and advanced configuration give you more control over how your data is recorded. Get powerful automation with a real person watching over it...
The loop matters. A first-versus-last comparison would show only a longer list of capabilities and miss the month in which the service relationship almost entirely displaced them.
Before
Syncsales, payments & refundswithQuickBooks,whileyour account manager takes care of everything
After
Export sales, payments & refunds to QuickBooks while a real human keeps an eye on the automation.
Reviews around its first retained edit, 2026-07-23. No dated reviews on either side.
Before 2026-06-25–2026-07-22; after 2026-07-24–2026-08-20; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. No dated reviews appear in either original window. The narrative can identify what is being sold, but cannot infer that the service framing increased demand.
The offer behind the keyword
The same specificity reaches the offer. A free plan becomes an order allowance with a unit; a broad AI claim becomes named tasks; a dated rule becomes an evidence workflow. These are the edits a price/trial-only count intentionally misses.
Retention Hub: Loyalty Program · Inventive Media Pty Ltd
Retention Hub: “free” changes meaning with the allowance
The July 6 body combines $49 per month, a seven-day trial and “Install free.” That wording does not by itself describe a continuing free usage tier. On July 20 the captured cards become Free, Starter, Growth and Scale, with order-volume limits rather than feature exclusions.
Before
Retention Hub helps Shopify merchants earn morerepeat revenuewith native loyaltyand referrals.Reward customers forpurchases and actionslike sign-ups or social follows,let themredeempoints for discounts, andrunrefer-a-friendprograms that turn advocates into new buyers. Brandedwidgets and dedicatedloyalty pagekeep rewards visible from storefront to thank-youand customer accounts, building stronger relationships and higher lifetime value without heavysetup.
After
Most stores lose repeat revenue when customers buy once and never return. Retention Hub is an all-in-one loyalty app — points, referrals, VIP tiers, and bonus campaigns. Customers earn on purchases and actions, redeem at checkout, and share refer-a-friend links. Branded loyalty page, sitewide widget, and product page points preview. Works with checkout and customer accounts. $49 USD/month. 7-day free trial. Fast setup, no developer needed. All features included — no hidden fees. Install free.
The July 29 tagline says free for 100 orders per month. On August 3 the allowance becomes 250, and the free plan’s bullet changes on the same date. The prices stay put. This is a measurable expansion in the described package that the price/trial-only count intentionally misses.
Before
Everything you need for customer loyalty. Free for100orders/month. All features included.
After
Everything you need for customer loyalty. Free for 250 orders/month. All features included.
August 20 then adds a Shopify Plus restriction to redemption at checkout in the free card. Free access, order limits and platform eligibility are separate parts of the offer. Collapsing them into a “free” keyword would lose the meaningful change.
Reviews around its first retained edit, 2026-07-06. No dated reviews on either side.
Before 2026-06-08–2026-07-05; after 2026-07-07–2026-08-03; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The paired tagline and plan bullet corroborate the stated allowance change within the listing. They do not establish actual billing, enforcement, conversion, or long-term free access.
PageFly ✦ AI Page Builder · PageFly | AI Landing Page Builder & CRO
PageFly: a broad AI claim becomes a named workflow
The July 16 body adds an AI page and section generator, alongside a drag-and-drop editor and a template count. It also claims an AI audit that prescribes fixes. In August the audit sentence is replaced by specific Sidekick tasks: finding templates, help articles and published pages. The integration becomes something the reader can picture asking for.
Before
Go beyond themes—create conversion-readylanding & product pagesfast, no coding required.
After
AI-powered page builder to design high-converting landing & product pages and boost store CRO.
Later copy lists page types and mentions a cart drawer in beta, then adds Claude to the named assistant surface. These are different claims: a generator, analytics and testing, assistant access, and an explicitly unfinished feature. A single “AI added” classification groups them too broadly.
Before
Our smart AI audit scans your pages to prescribe prioritized,actionable fixes for speed,SEO, andcopy.
After
In Sidekick, ask PageFly to search page and section templates, find help articles, and list your published pages.
This is the largest positive contribution to the new AI-aligned review total. The July 16 window has 22 reviews before and 50 after. The original report used the earlier July 11 retained event and found 25 to 40. Both are valid descriptions of different windows; neither should silently replace the other.
Reviews around its first retained edit, 2026-07-11. Dated reviews 25 → 40; mean 4.92★ → 4.98★ on 25 → 40 rated; visible at extraction 25 → 35.
Before 2026-06-13–2026-07-10; after 2026-07-12–2026-08-08; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. Several features and later edits share the after-window. The cohort’s apparent AI increase is heavily concentrated here, so it does not establish a general benefit from adding AI language.
Green Claims Guard · momiji-shop
Green Claims Guard: a dated rule, then an evidence workflow
The July 10 body names rules and dates and promises that each finding cites a provision. It explicitly says the tool never rewrites the text or claims legal outcomes. July 26 puts the asserted deadline first and later clarifies “Records wording, not legality.” The proposed output is a cited record for review, not a certificate that the shop is compliant.
Before
Scan green marketing claims. Keep dated evidence records for review.
After
Continuously scans EU EmpCo (27 Sep 2026) and UK CMA wording, citing sources — never AI guesses.
The September 16 body compresses into a free scan and upgrade permissions. The substantive rule explanation no longer appears in that field even though the tagline retains the deadline. This is why a body-only compliance count can fall without establishing that the app abandoned the positioning.
Before
Re-scans your catalog, pages, and blog posts after every edit, plus a weekly sweep, againstthe EUEmpCo Directive(2024/825,applies 27 Sep2026) andthe UK CMA Green Claims Code. Every finding citesits UCPD Annex I point or CMA principle, so you see why, not just that.Keep an evidencevaultof certifications, exporta timestamped audit trail, andtry our method free: greenclaims.momiji-space.com/green-claims-checker.html.Read-only: never rewrites copy or claims legal outcomes.
After
From 27 September 2026 the EU bans generic environmental claims and offset-based climate claims outright. Germany has already legislated this (UWG, BGBl. 2026 I Nr. 43). Scans products, collections, pages and articles for green-claim wording in English and German, plus US origin claims (16 CFR Part 323). Every finding names the provision it cites. Evidence vault, CSV export, monitoring re-scans, and a printable compliance report on the Agency plan. Records wording; does not assess legality.
Before
From 27 September 2026 the EU bans generic environmental claims and offset-based climateclaims. Germany has legislated this (UWG, BGBl. 2026 I Nr. 43).Scans products, collections, pages and articles in English and German, plusUSorigin claims (16 CFR Part 323). Every finding names the provision it cites. Try it first: the free checker at greenclaims.momiji-space.com scans any page. Evidence vault, CSV export, monitoring,printable report on Agency. Records wording, not legality.
After
Free scans claims with cited rules. Upgrade for DE/US packs, monitoring, CSV (Starter+).
No eligible first-event comparison. The app lacks a complete observed baseline or postperiod. It has 0 unedited dated reviews during the screen.
What this cannot show. All legal statements here are quotations of listing claims, not this report’s legal guidance or verification of the named rules. The portfolio’s zero-review context is preserved.
That is what the words support: tasks named, switching costs addressed, control offered, terms specified. A listing is edited to change what happens next, and the one next thing we record is a review. The last two steps ask whether any of this shows up there.
Step 9 · Reviews, first pass
Reviews around each app's first edit
We can compare equal 28-day windows for 132 of the 211 apps: each needs to have been observed at least 28 days before its first retained edit and to leave a complete post-period. The remaining apps are ineligible, not failures. Of those eligible, 49 rise, 57 are unchanged and 26 fall in unedited dated review count. Unchanged usually means very little review activity, not a measured null effect.
More rises than falls is not yet a finding about editing, because reviews move for reasons of their own. So each app gets three controls from its current primary category that have no raw description/pricing rows during the whole screen and were observed before the baseline, chosen by nearest preperiod review volume and app age without looking at postperiod counts. Because quietness and hyperoptimizer membership use later activity, this remains a descriptive comparison, not a prospective experiment.
| App | App before → after | Controls before → after | Relative multiple | Rating before → after |
|---|---|---|---|---|
| Arvio: AI Store Operator | 4 → 0 | 10 → 1 | 0.0× | 5.00 → — |
| Amp Slide Cart Upsell Drawer | 17 → 20 | 68 → 80 | 1.0× | 5.00 → 5.00 |
| PIMW Product Personalizer | 3 → 1 | 6 → 3 | 0.667× | 5.00 → 5.00 |
| Buddy AI: Agentic Commerce | 12 → 27 | 32 → 19 | 3.789× | 4.83 → 5.00 |
| Buddify: Collective Commerce | 24 → 9 | 65 → 63 | 0.387× | 4.58 → 4.89 |
| SearchPie: SEO, Schema & Speed | 28 → 22 | 159 → 115 | 1.086× | 5.00 → 4.96 |
| Prediko Inventory Management | 14 → 7 | 47 → 34 | 0.691× | 4.43 → 5.00 |
Pooled across all 132, the hyperoptimizers' dated reviews go 474 → 516 (+9%) while their control appearances go 1,501 → 1,408 (−6%). A pooled sum can be carried by a few large pairs, so we cut by size measured before the edit, never by what happened after it: dropping the 3, 5 or 10 pairs with the largest control baselines, or the 5 or 10 largest apps. The cohort grows in every cut (+12% to +42%) and its controls never do (−24% to −1%). Pair by pair the shape is the same: the app's count rose in 49 pairs and fell in 26; its control trio's rose in 16 and fell in 47.
That is an association, not yet a finding about editing. Both groups were chosen on what they did later. A hyperoptimizer kept editing into September; a listing nobody touched for 100 days includes listings nobody is running, and control trios falling 47 times to 16 is what that rival predicts. Editing is also what an active business does alongside support, advertising and product work that no listing records. So the comparison says that listings being worked on gained reviews faster than listings left alone. It cannot say that the listing work was the reason, and the rest of this step and the next test the one version of that reason the record can reach: the words.
Relative multiple = (app after ÷ app before) ÷ (pooled controls after ÷ pooled controls before). We leave it undefined when the required denominator is zero or three controls are unavailable. It has no uncertainty interval here and is not an estimated treatment effect. A value above one can coexist with fewer reviews, as SearchPie shows.
Stars, on the smaller group that has enough of them
Review count is not the only way reception can move. Requiring at least three rated dated reviews on both sides of the first retained edit leaves 25 apps: a smaller, noisier group, but one where an average means something. 9 rise, 7 fall and 9 are unchanged. Pooled, summing every star and dividing by every rated review rather than averaging app averages, the group runs 4.90★ before its first retained change and 4.94★ after. 9 against 7 is a split chance produces routinely, and the pooled change is three hundredths of a star on ratings already near the ceiling. Stars do not move enough here to carry an argument either way.
If the listing work is the reason, the close readings offer candidates: Prediko's Stocky pitch, Buddify's community, Zapiet's award. We tested them the only way the record allows, by date. A sentence can only explain a review window it arrived inside.
Mostly they arrived too late. Prediko's detailed migration sentence comes on 18 August, four weeks after its after-window ended on 22 July; only the trailing “Replaces Stocky” lands inside it, on 11 July. Buddify's community positioning arrives on 11 August, and its window closed on 15 July. Zapiet's award claim appears on 25 July, three days after its post-window ended, and its free tier returns later still. Buddy AI's trial changes are September. Wiser's “selleasy” is September. The one exception is PIMW's AI option generator, inside its window, and PIMW's reviews fell from three to one. The rewrites that make the best reading cannot explain the rating history they sit beside.
A second rival treats the count and the rating as two results. Prediko and Buddify both lose reviews and gain stars at once: Prediko 14 → 7 reviews and 4.43 → 5.00★, Buddify 24 → 9 and 4.58 → 4.89★. A smaller, uniformly satisfied after-group is one fact seen two ways, and neither half alone is the outcome. Buddy AI, the strongest count movement in the cohort at 12 → 27, carries a different caution: restricted to reviews still visible at extraction it is 11 → 13.
Buddify is the case where both rivals meet, so its full reading belongs here rather than with the vocabulary.
Buddify: Collective Commerce · Buddify
Buddify: scale, two-sided commerce, then belonging
The June 17 tagline replaces numerical scale—200,000 products and 2,500 verified brands—with the relationship between the two sides: source products and let other stores sell yours. The body follows it, dropping the approximately 20% partner-price claim and introducing brand partnerships, wholesale buyers and cross-selling. That changes which reason to join is most legible.
Before
Discover 200,000+ discounted products from verifiedShopifybrands,all with circa 20%+ partner pricing. Import in one click,chat directlywithsuppliers,negotiate wholesale deals, andbuild relationships thatgrowyour business—not just another dropshipping store.
After
Buddify helps Shopify stores discover products, build brand partnerships, find wholesale buyers, and create new revenue streams through cross-selling. Connect with verified brands, automate inventory sync, and grow without inventory or commissions.
The numbers return on July 1. They then disappear again on July 9 in favour of wholesale, dropshipping and brand partnerships. These are not successive improvements along one axis. The sequence moves between scale as proof, the mechanics of buying and selling, and a broader account of the network.
Before
Connect withotherbrands, sourceproducts,and let other stores sell yours.
After
Connect with 2,500+ brands, source 200,000+ products & grow through B2B partnerships.
On August 11 the tagline becomes “The community where brands help each other grow.” The body and bullets stress trusted brands, co-marketing and new sales channels; the observed title changes to Buddify Community the same day. In September, category terms are appended to the tagline before the “Collective Commerce” title arrives. Community language and discoverable commerce nouns coexist uneasily.
Before
Buildwholesale, dropshipping and brandpartnerships withone global commerce network. Buddify connects suppliers and retailers so you can sourceproducts,let other storessellyours,discover new business partners and grow through shared inventory andco-marketing. Expand your catalog, increase distribution and automate order syncing,inventoryandfulfillment—all without payingsalescommissions.
After
Buddify Community helps brands grow together. Build partnerships with trusted brands, discover discounted partner products, buy and sell wholesale, collaborate on co-marketing campaigns, and create new sales channels by letting other Shopify stores sell your products. Everything happens inside one trusted community designed for long-term growth.
The first-event postperiod ends July 15. Its lower count and higher average therefore accompany the June/July move away from a discounted catalogue, not the August community positioning. The original prose placed the community move in September; the captured body and title establish August 11.
Before
The community where brands help each other grow.--Dropshipping & Wholesale & Find Products - Collab
After
Buddify Collective Commerce: The community where brands help each other grow.
Reviews around its first retained edit, 2026-06-17. Dated reviews 24 → 9; mean 4.58★ → 4.89★ on 24 → 9 rated; visible at extraction 23 → 4.
Before 2026-05-20–2026-06-16; after 2026-06-18–2026-07-15; event day omitted. Ratings with fewer than three reviews on a side are context only and stay out of the pooled comparison.
What this cannot show. The fall from 24 to 9 reviews and rise from 4.58 to 4.89 stars are preserved. Currently visible reviews fall from 23 to 4. A different promise might attract a different audience, but no exposure, acquisition or reviewer text links that mechanism to the observed change.
We separately retained listing totals at five calendar boundaries and counted current visible review rows. 200 of 211 hyperoptimizers have current inventory equal to their latest listing total. The remainder may reflect asynchronous collection or incomplete records. Even agreement only checks today's stock: it cannot validate the historical timing of every review.
The first-edit anchor has a structural problem as well as a timing one. Each app's first retained edit may be a pricing or plan-bullet change, so an AI phrase added in August cannot explain a June comparison. The honest next move is to align the window to the wording itself.
Step 10 · Reviews, second pass
Reviews around the wording itself
We keep the original result and add two views. The first asks what was added at the original anchor, splitting the eligible apps by whether their changed body newly mentioned a family on that date. “Other” includes apps that already used the term, added it later, or never used it, so it is not a clean untreated group. Its use is to test whether a proposed explanation even coincides with the original window.
| New body wording at original event | Apps adding | Their reviews before → after | Other apps | Other reviews before → after | Adding apps with ≥3 ratings each side |
|---|---|---|---|---|---|
| AI wording | 3 | 0 → 2 | 129 | 474 → 514 | 0 |
| Automation / autopilot | 6 | 26 → 16 | 126 | 448 → 500 | 1 |
| No-code wording | 6 | 2 → 5 | 126 | 472 → 511 | 0 |
| Approval / preview / undo | 3 | 0 → 0 | 129 | 474 → 516 | 0 |
| Free entry / service | 2 | 1 → 2 | 130 | 473 → 514 | 0 |
| Named tools / channels | 3 | 12 → 12 | 129 | 462 → 504 | 1 |
| Stocky | 0 | 0 → 0 | 132 | 474 → 516 | 0 |
Very few original anchors introduce these terms, and none of the apps newly adding AI body wording at that date has enough rated reviews on both sides for the rating rule. The pooled rise on the 25 apps cannot be described as a demonstrated reward for adding AI.
The second view aligns the window to the first actual mention. For each family we take an app's first body transition from no mention to a mention, provided its first captured old body lacked the term, with at least two days to the next body row and a complete observed 28-day baseline and postperiod. Each app contributes at most once per family. Rapid first mentions are excluded rather than replaced by a more convenient later event. Negative-AI positioning (AddOn Suggest) and the unresolved Simesy product swap are excluded from the relevant comparisons. These filters measure incident wording, not verified feature adoption.
| First incident body mention | Eligible / candidates | Reviews before → after | Up / equal / down | Zero both | Median count change | Rated apps | Pooled stars before → after |
|---|---|---|---|---|---|---|---|
| AI wording | 16 / 31 | 46 → 79 | 5 / 9 / 2 | 8 | 0 | 3 | 4.82 → 4.97 |
| Agent / agentic | 3 / 8 | 27 → 27 | 1 / 0 / 2 | 0 | −1 | 1 | 5.00 → 4.96 |
| AI discovery vocabulary | 7 / 14 | 2 → 9 | 3 / 4 / 0 | 3 | 0 | 0 | — → — |
| Automation / autopilot | 18 / 39 | 30 → 26 | 6 / 10 / 2 | 9 | 0 | 1 | 4.58 → 4.89 |
| No-code wording | 17 / 23 | 32 → 38 | 5 / 10 / 2 | 9 | 0 | 2 | 4.96 → 4.95 |
| Free entry / service | 7 / 21 | 39 → 40 | 2 / 4 / 1 | 4 | 0 | 1 | 5.00 → 5.00 |
| Approval / preview / undo | 10 / 22 | 5 → 11 | 3 / 5 / 2 | 5 | 0 | 1 | 5.00 → 5.00 |
| Compliance / named standards | 1 / 5 | 0 → 1 | 1 / 0 / 0 | 0 | +1 | 0 | — → — |
| Profit / margin | 9 / 13 | 13 → 6 | 1 / 6 / 2 | 6 | 0 | 1 | 5.00 → 5.00 |
| 24/7 / instant | 10 / 19 | 19 → 33 | 5 / 5 / 0 | 5 | +0.5 | 1 | 5.00 → 4.96 |
| Stocky | 3 / 6 | 10 → 14 | 2 / 0 / 1 | 0 | +2 | 1 | 4.89 → 5.00 |
| Black Friday / BFCM | 1 / 1 | 1 → 1 | 0 / 1 / 0 | 0 | 0 | 0 | — → — |
| Named tools / channels | 13 / 27 | 92 → 92 | 4 / 5 / 4 | 2 | 0 | 3 | 4.94 → 4.82 |
| Sidekick | 3 / 7 | 75 → 76 | 1 / 1 / 1 | 0 | 0 | 2 | 4.99 → 4.85 |
| Certification wording | 0 / 0 | 0 → 0 | 0 / 0 / 0 | 0 | — | 0 | — → — |
Candidates are distinct apps with no initial body mention and a subsequent first addition. Eligible counts exclude insufficient observation, late or rapid changes, and the stated semantic/identity exceptions. Apps and reviews recur across families, so rows cannot be added. Stars pool only apps with at least three rated reviews each side. Every eligible and excluded app row, date and control identity is in the snapshot.
The largest apparent rise is AI: 46 to 79 reviews across 16 eligible apps. It is also the family where the pooled number is least like its parts.
8 of the 16 have zero reviews on both sides, and the median change is 0. PageFly alone goes from 22 to 50. Without it, the other 15 apps move from 24 to 29. Only 3 AI-adding apps meet the rating rule. A pooled rise concentrated in one established app is not a cohort-wide payoff, and PageFly's own after-window also contains a page generator, analytics and testing, Sidekick access and a beta feature: the close reading in step 8 shows how many different claims share it.
Nor is the surrounding listing held fixed for anyone: 10 of the eligible AI cases have later body or feature edits in the after-window, and 10 have later pricing rows. Pricing rows include plan copy as well as economic changes. The plausible mechanisms are bundled together in time.
The last rival is the store itself: reviews in a category can rise for everyone. For each incident mention we look for up to three other cohort apps in the same frozen primary category, observed early enough and never mentioning that family in their body during the screen, chosen by nearest baseline review volume without using postperiod counts. Controls share the event date. They can be reused, can edit other wording, and are chosen partly using future non-mention, so these are descriptive checks, not causal estimates.
| Wording family | Apps with 3 controls | Matched app reviews | Control review appearances | Distinct control apps |
|---|---|---|---|---|
| AI wording | 5 | 17 → 26 | 10 → 19 | 9 |
| Agent / agentic | 1 | 22 → 26 | 5 → 5 | 3 |
| AI discovery vocabulary | 4 | 2 → 2 | 2 → 9 | 12 |
| Automation / autopilot | 2 | 0 → 4 | 13 → 13 | 6 |
| No-code wording | 4 | 13 → 12 | 14 → 30 | 12 |
| Free entry / service | 4 | 1 → 2 | 33 → 30 | 12 |
| Approval / preview / undo | 4 | 3 → 9 | 20 → 31 | 10 |
| Compliance / named standards | 0 | 0 → 0 | 0 → 0 | 0 |
| Profit / margin | 3 | 2 → 1 | 9 → 3 | 9 |
| 24/7 / instant | 5 | 2 → 5 | 18 → 24 | 9 |
| Stocky | 2 | 10 → 12 | 10 → 6 | 3 |
| Black Friday / BFCM | 1 | 1 → 1 | 1 → 3 | 3 |
| Named tools / channels | 3 | 23 → 28 | 9 → 10 | 9 |
| Sidekick | 3 | 75 → 76 | 7 → 9 | 9 |
| Certification wording | 0 | 0 → 0 | 0 → 0 | 0 |
For AI, the 5 matched apps rise from 17 to 26 while their control appearances rise from 10 to 19. PageFly has no three same-category controls and is not among them. The matched subset does not show an advantage in proportional review growth. Other families have too few matched or rated apps to support a stable comparison. Stocky is a convincing shared copy campaign, but its incident-mention outcome sample is small and omits apps already using the name.
Step 11 · The boundary
What the evidence supports
Repeated listing work is real, survives stricter definitions, and is mostly verbal: 909 copy edits to 358 price or trial edits, at a steady pace across three months. Nine price editors returned exactly to a price they had already tried, and in the two we read closely the product around that price had changed.
Read as sentences, the edits move in one direction more than any other: toward the specific. A named competitor's customers at the moment they must leave. A named task, a named tool, a named safeguard, a named person. A free plan with a unit attached. Some edits reach for neighbouring searches directly, by inserting a competitor-shaped word or rotating a product's synonyms through a body that never grows. AI language, which a trend count would have made the headline, churns in both directions in a fixed panel.
The review record does not say which of this works. Around each app's first edit, the hyperoptimizers gained reviews faster than quiet listings in their categories, which is what a business being worked on looks like from outside and not yet what an edit does; their stars barely moved. The close readings that would explain the rise mostly arrived after the windows they would need to explain, and the one that arrived inside its window sits beside a fall in reviews. Aligned to the wording itself, the one apparent reward, for AI, is almost entirely one app, and the matched subset does not beat its controls.
So what we have is the outline of many small decisions: their dates, their words and their direction, without the decisions themselves. The edit histories are not standing in for a mechanism we failed to find. They are the form in which a developer's search for the right sentence is visible from outside. They show that the search happens, what it tries and how fast it moves. They do not show what it found. Installs, exposure, ranking, conversion and revenue are unobserved, and this is insufficient evidence for a difference between kinds of copy, not evidence that all kinds have equal or zero effect.
Appendix A · Eight full histories
What eight hyperoptimizers changed, and what followed
The cases below combine the highest activity, multiple products under one developer, and established listings with enough reviews to compare. This is an illustrative selection, not a representative subsample; the full cohort and every eligible comparison remain available below and in the data. Counts refer to unedited displayed review dates, including stored reviews later removed or archived; ratings refer to the same set, restricted to reviews with a recorded star value.
Arvio: AI Store Operator · Adot Technologies Inc
A changing proposition, then a new pricing ladder
21 edit days14 copy14 plan6 price/trialevery 4.6 days6 dated reviews, avg 4.83★
Arvio is the most frequent editor under this screen: 14 copy-edit days and 14 plan-edit days, 21 in total, an edit every 4.6 days on average from 18 June to 18 September. June's tagline moves from avoiding another hire to an operator that edits products, SEO and email with approval. July adds a limited-time $9.90 plan alongside two uncosted “estimated monthly cost” tiers. By August the ladder settles into Starter $9.90, Growth $39 and Pro $99, all with five-day trials, a genuine repricing rather than a relabelling, though the early ‘estimated monthly cost’ fields never reveal an actual dollar price.
Reviews. Its first comparison falls from four dated reviews to zero, and only six land across the full 100 days (all five-star, where rated). The fastest, most repriced listing in the cohort is a clear activity case, not evidence that the work paid off in review volume.
| Calendar context | May | Jun | Jul | Aug | Sep 1–20 |
|---|---|---|---|---|---|
| Unedited dated reviews | 0 | 5 | 2 | 1 | 0 |
Calendar months provide context for later edits; they are not event-aligned effects. September has only 20 days. A dash means the entire month precedes our first observation; earlier reviews can be collected retrospectively. First retained event: 2026-06-18. Before: 2026-05-21–2026-06-17. After: 2026-06-19–2026-07-16. The event date is excluded.
True COGS: Cost of Goods Sold · Nexus Logic Systems
The price can stay put while the package changes
20 edit days14 copy11 plan1 price/trialevery 4.47 days3 dated reviews, avg 5.00★
True COGS rewrites the cost-accounting problem and the promises attached to its plans on 14 of 20 edit days, at a similar pace to Arvio (one retained edit every 4.5 days). On 24 August, Free / Standard $29 / Pro $79 becomes Free / Starter $29 / Growth $49 / Scale $79. Feature-based tiers give way to receipt-line allowances (25, 300, 1,500, unlimited), and later bullets describe FIFO costing, landed-cost math and a CPA packet. The free allowance, the unit being limited, and the paid value proposition all move; only one of its eleven plan-edit days actually touched the displayed price.
Reviews. Only three unedited reviews are dated in the 100-day window, all five stars. We first observed the app on 19 June, two days before its first retained edit, so it has no eligible 28-day baseline; a zero-filled preperiod here would be misleading, not informative.
| Calendar context | May | Jun | Jul | Aug | Sep 1–20 |
|---|---|---|---|---|---|
| Unedited dated reviews | — | 0 | 0 | 2 | 1 |
Calendar months provide context for later edits; they are not event-aligned effects. September has only 20 days. A dash means the entire month precedes our first observation; earlier reviews can be collected retrospectively.
Amp Slide Cart Upsell Drawer · App HQ Pte. Ltd.
A price cut, a higher tier, and seasonal copy
16 edit days9 copy9 plan6 price/trialevery 6.27 days70 dated reviews, avg 4.93★
AppHQ's Slide Cart adds an ‘original since 2019’ claim in June and changes Growth from $49.99 to $39.99 on 26 June. By 3 August, Starter is $14.99 and Pro $29.99. On 24 August the whole menu changes shape, to order-volume tiers at $19.99 / $29.99 / $39.99 plus a development-store plan, so the price-cut sequence ends in a different package, not simply a cheaper one. September's tagline explicitly prepares for Black Friday. This is one of the stronger examples of an established listing repeatedly adjusting both copy and offer.
| Detected | Captured new price/trial state |
|---|---|
| 2026-06-26 | free forever: free; growth: $39.99/month (7-day free trial) |
| 2026-06-30 | free forever: free; pro: $49.99/month (7-day free trial); starter: $39.99/month (7-day free trial) |
| 2026-07-03 | free forever: free; pro: $49.99/month (7-day free trial); starter: $29.99/month (7-day free trial) |
| 2026-07-20 | free forever: free; pro: $39.99/month (7-day free trial); starter: $19.99/month (7-day free trial) |
| 2026-08-03 | free forever: free; pro: $29.99/month (7-day free trial); starter: $14.99/month (7-day free trial) |
| 2026-08-24 | 0-200 orders: $19.99/month (7-day free trial); 201-500 orders: $29.99/month (7-day free trial); 500+ orders: $39.99/month (7-day free trial); development stores: free |
Reviews. Dated reviews rise from 17 to 20 after the first retained edit, essentially matching its three quiet-listing controls (68 → 80, a relative multiple of 1.0). Ratings do not move either: 5.0 before, 5.0 after, on 17 and 20 rated reviews. The window starts at the first copy edit, not the later price cuts, so the design cannot isolate any one of them; later monthly counts (19 in June, 25 in July, 22 in August) show successive cuts not obviously tracking a rising count.
| Calendar context | May | Jun | Jul | Aug | Sep 1–20 |
|---|---|---|---|---|---|
| Unedited dated reviews | 21 | 19 | 25 | 22 | 10 |
Calendar months provide context for later edits; they are not event-aligned effects. September has only 20 days. A dash means the entire month precedes our first observation; earlier reviews can be collected retrospectively. First retained event: 2026-06-13. Before: 2026-05-16–2026-06-12. After: 2026-06-14–2026-07-11. The event date is excluded.
PIMW Product Personalizer · Pixelspiece solutions
Pixelspiece expands keywords and the offer
14 edit days10 copy6 plan3 price/trialevery 5.15 days9 dated reviews, avg 5.00★
PIMW Product Personalizer moves from general real-time personalisation toward a much denser vocabulary: POD, jewelry, apparel, furniture, engraving, monograms, photo uploads and conditional options. On 10 July it adds an Unlimited tier at $24.99 alongside Free, $7.99 and $13.99, a genuine new price point rather than a rename. Later trial fields disappear and return at three days; an absent field is a captured absence, not a verified withdrawal. Pixelspiece's Tailor size-guide app also qualifies, making this a repeated pattern across two of the company's products.
Reviews. PIMW has nine unedited dated reviews in the 100 days, all five stars, but the first-event comparison falls from three to one (both still five stars where rated). The one post-window review is no longer visible at extraction. The later calendar picture differs (5 in August against 2 in July), so the first-event fall is not a verdict on every subsequent edit; both comparisons are too small to attribute an effect either way.
| Calendar context | May | Jun | Jul | Aug | Sep 1–20 |
|---|---|---|---|---|---|
| Unedited dated reviews | 2 | 2 | 2 | 5 | 0 |
Calendar months provide context for later edits; they are not event-aligned effects. September has only 20 days. A dash means the entire month precedes our first observation; earlier reviews can be collected retrospectively. First retained event: 2026-07-09. Before: 2026-06-11–2026-07-08. After: 2026-07-10–2026-08-06. The event date is excluded.
Buddy AI: Agentic Commerce · Buddify
The promising rise with an important sensitivity
10 edit days6 copy4 plan3 price/trialevery 10.11 days45 dated reviews, avg 4.87★
Buddy AI rewrites its proposition in June from five agents that sell automatically toward a store-specific AI team, then voice, recommendations and support: six copy-edit days set against only four plan-edit days, the most copy-led of the eight. Its September paid tiers keep $49 / $79 / $199 while trials move from 14 to 30 to seven days; that later trial sequence cannot explain reviews dated in the earlier June–July comparison window.
Reviews. All stored unedited dates rise from 12 to 27 while controls fall from 32 to 19, a raw 2.25× and relative 3.789× increase, the strongest count movement in the cohort. Restricting to reviews still visible at extraction changes the comparison to 11 → 13. The average rating also rises, 4.83 → 5.0 on 12 and 27 rated reviews. Monthly counts fall from 31 in June to 13 in July to 8 in August, so the first-window rise is not sustained monthly acceleration. The positive association merits attention, but its size depends heavily on reviews that later disappeared from view; it does not establish why they disappeared or that anything improper happened.
| Calendar context | May | Jun | Jul | Aug | Sep 1–20 |
|---|---|---|---|---|---|
| Unedited dated reviews | 3 | 31 | 13 | 8 | 4 |
Calendar months provide context for later edits; they are not event-aligned effects. September has only 20 days. A dash means the entire month precedes our first observation; earlier reviews can be collected retrospectively. First retained event: 2026-06-13. Before: 2026-05-16–2026-06-12. After: 2026-06-14–2026-07-11. The event date is excluded.
Buddify: Collective Commerce · Buddify
The same developer, the opposite direction
7 edit days5 copy2 plan2 price/trialevery 14.33 days16 dated reviews, avg 4.94★
Buddify's collective-commerce app also changes both description and plans, on a much slower cadence than its sibling: an edit every 14.3 days, the slowest of the eight. On 11 August the body replaces the supplier-and-wholesale framing with a community in which brands help each other grow. The two products show the same developer editing more than one listing, in different categories and at different stages.
Reviews. Its first-event comparison falls from 24 dated reviews to nine, while the same-category controls barely move (65 → 63). Counts still visible at extraction fall further, 23 → four. The average rating rises anyway, 4.58 → 4.89 on 24 and nine rated reviews: a smaller, quieter audience that rated the app somewhat better, not a growing one. The contrast with Buddy AI rules out a simple story in which this developer's frequent editing always accompanies more reviews.
| Calendar context | May | Jun | Jul | Aug | Sep 1–20 |
|---|---|---|---|---|---|
| Unedited dated reviews | 0 | 34 | 1 | 0 | 0 |
Calendar months provide context for later edits; they are not event-aligned effects. September has only 20 days. A dash means the entire month precedes our first observation; earlier reviews can be collected retrospectively. First retained event: 2026-06-17. Before: 2026-05-20–2026-06-16. After: 2026-06-18–2026-07-15. The event date is excluded.
SearchPie: SEO, Schema & Speed · PieLab
SearchPie is a copy-led case
9 edit days8 copy2 plan0 price/trialevery 11.75 days84 dated reviews, avg 4.95★
SearchPie repeatedly reshuffles the SEO vocabulary in its body and tagline (AEO, structured data, JSON-LD, image ALT text, AI descriptions, Google Search Console) across eight of its nine edit days, with zero retained price or trial changes. Its two qualifying plan edits concern captured names or bullets only: a plan edit is not automatically a price cut.
Reviews. Dated reviews fall from 28 to 22; the control group also falls, from 159 to 115, so the relative multiple (1.086) sits above one despite an absolute decline. The average rating slips slightly, 5.0 → 4.96 on 28 and 22 rated reviews. Both quantities matter: ‘outperformed controls’ would conceal that this app received fewer reviews and a touch less praise per review.
| Calendar context | May | Jun | Jul | Aug | Sep 1–20 |
|---|---|---|---|---|---|
| Unedited dated reviews | 39 | 24 | 30 | 32 | 11 |
Calendar months provide context for later edits; they are not event-aligned effects. September has only 20 days. A dash means the entire month precedes our first observation; earlier reviews can be collected retrospectively. First retained event: 2026-06-15. Before: 2026-05-18–2026-06-14. After: 2026-06-16–2026-07-13. The event date is excluded.
Prediko Inventory Management · Prediko
Changed explanation, quieter but better-rated reviews
8 edit days5 copy5 plan0 price/trialevery 11.71 days32 dated reviews, avg 4.88★
Prediko expands abbreviations and rewrites inventory-planning language, then repeatedly changes captured plan bullets: five of its eight edit days touch plan copy, none touch the displayed price or trial. This is optimising the explanation and package, not the sticker price.
Reviews. The first-event comparison falls from 14 dated reviews to seven, against 47 → 34 for controls, a relative multiple of 0.691 and one of the weaker count outcomes in the case set. But the average rating rises sharply, 4.43 → 5.0 on 14 and seven rated reviews: the smaller after-group is uniformly five stars. Across the full 100 days it has 32 unedited dated reviews. Fewer reviews and a higher average can be the same fact seen two ways; neither alone is “the” outcome.
| Calendar context | May | Jun | Jul | Aug | Sep 1–20 |
|---|---|---|---|---|---|
| Unedited dated reviews | 9 | 13 | 9 | 10 | 8 |
Calendar months provide context for later edits; they are not event-aligned effects. September has only 20 days. A dash means the entire month precedes our first observation; earlier reviews can be collected retrospectively. First retained event: 2026-06-24. Before: 2026-05-27–2026-06-23. After: 2026-06-25–2026-07-22. The event date is excluded.
Appendix B · Evidence archive · public listing text only
Read beyond the selected examples
The 22 close readings above were chosen for an interpretable sequence, contrasting strategies or a measurement problem; they are not a random sample. Each pair resolves to a specific event and field in the frozen snapshot, with spelling, awkward phrasing and qualification preserved and words removed or inserted marked. Detection dates can lag publication. Here you can search every captured field in the same cohort, including full bodies, bullet lists, pricing tiers, category structures and observed titles, and choose an app to follow every row in order. The archive includes small edits, reordered lists and rapid changes, so it is intentionally broader than the retained-event calendar. No reviewer text or identity is included.
The interactive archive is omitted from print. All selected before/after pairs above are printed; the full searchable archive and raw JSON remain available at mirakle.dev/research/2026-09-28-listing-optimizers.
Appendix C · Method, limits and reproducibility
What the snapshot preserves
- Frozen evidence, separately timed reads. The original extraction used separate production READ ONLY SELECTs replayed locally, not a single repeatable-read transaction. The copy extension preserves every original metric and chart array, and adds four separately timed READ ONLY queries. The manifest distinguishes production read times from local replay times and records parent and analysis digests. No production data or deployed files were changed.
- Two scopes. Wide uses all its recorded listing history through extraction. The broad screen uses 13 June–20 September, avoiding the known June parser transition and the most recent week. Monthly supporting tables run April–20 September; September has only 20 days and is not compared as a full month.
- What is captured. Copy means body and tagline. The broad screen does not count standalone feature-bullet events. Plans include captured names, displayed prices, trials and bullets. Price/trial-set changes can be added tiers or different trial labels, not just price changes; feature bullets can themselves change the economic allowance even when the displayed price is unchanged.
- Cleaning. Unicode normalization, lowercase and collapsed whitespace; copy uses a word-token multiset difference, with both states present and bodies at least 30 characters. Plan cards are deduplicated and sorted; empty reads and equal canonical states are removed. The two-day spacing check uses the next same-type raw row, including noisy rows, and may discard valid fast edits.
- Cadence. An app's “mean gap” is the average number of days between two consecutive retained edit days, over its own edit calendar; the cohort figure is the median of those per-app means, not a single pooled average. Edits per day in step 3 divide each month's retained edits by its observed days (18 in June, 20 in September).
- Reversal. A later price/trial-set change whose captured price/trial pairs (ignoring plan name and bullet text) exactly match an earlier captured state for the same app. It detects a repeated economic state, so a renamed tier at the same price still counts, as PromoLock shows; it says nothing about whether the repetition was deliberate.
- Ratings. Per-review star ratings are read from Shopify's accessible star-group label, not a fill-colour heuristic, since a scraper change on 3 May 2026, six weeks before this screen opens, so no pre-fix rating enters a comparison. An app enters a rating comparison only with at least three rated dated reviews on each side of its first retained edit. The pooled figures sum stars and rated reviews across apps before dividing.
- Outcome. Count stored review rows once by displayed posting date, including deleted and archived rows. Exclude edited dates in the primary result because a revision can replace the original date. Missing historical reviews, mislabelled edits and acquisition delays remain possible. Zero means zero in our stored dated records.
- Comparison. First retained event per qualifying app; event date omitted; 28 days on each side. Controls are ranked by |log(1+pre reviews) difference| + 0.25 × |log(1+launch age in days) difference|, then stable app ID. These choices prevent interpreting the relative multiples as causal.
- Selection. The directory contains all qualifying apps, not only successful ones. Case studies are purposive and labelled as such. Review counts, ratings and reversals were not used to set the editing thresholds.
- Close-reading coverage. We read the complete changed-tagline sequences for 204 apps, then followed full bodies, bullets, offers, titles and categories for the 22 published cases and additional contrasting examples. Every available body chain was checked by code; lexical counts cover the entire eligible panel. The cases are editorial readings, not a blinded, independently coded annotation of every sentence. Keywords and exclusions were developed after inspection, so these are exploratory descriptions.
- Figures drawn from the archive. Three figures read copy_history directly in the browser: the Stocky lanes (a case-insensitive whole-word match on “Stocky” in any captured field, carried forward per field), PIMW's term grid (ten terms chosen for that figure, matched in body plus tagline) and PromoLock's price ladder (every captured pricing row, with the eight retained offer days marked). Before/after marks are a word-level longest-common-subsequence diff; every character of both states is printed unchanged.
- Additional fields and publication. The copy supplement includes small and rapid edits, standalone bullets and category changes, plus titles from successful observations beginning in August. It does not alter the original cohort definition. Public listing links and contact addresses inside captured text are replaced with [link] and [contact]; otherwise the paired wording is preserved. The article and presentation PDFs are direct print exports of the Rails surfaces; no parallel standalone HTML report is maintained.
- Order of this page. Earlier versions presented counts, then outcomes, then the words as a supplement. This version orders the same evidence as an argument, so the close readings sit in the step they support and the outcome analysis follows the words it tests. Nothing was removed; the eight quantitative histories moved to Appendix A.
- Attribution. We observe listing states and reviews, not who typed the copy, whether an A/B test ran, or why a review disappeared. Associations do not establish motive, misconduct, installs, revenue or conversion changes.
The public evidence includes the complete qualifying directory, edit calendar, monthly counts, kind and cadence breakdowns, reversal list, eligible comparisons with named controls and ratings, coverage, stock checks, threshold sensitivities, and full captured before/after fields across all 211 apps, the title stream, lexical codebook, calendar panels, exact case passages and claim-aligned review aggregates. It excludes reviewer names, store identities and review text.
Appendix D · Complete cohort
All 211 hyperoptimizers
Sorted by distinct edit days, then copy days and slug. “Offer” is the number of days with a retained price/trial-set change. “Reviews” means unedited displayed dates during the 100-day screen; “rating” is their average, where at least one is dated.