The Shelf / App Store / App Review Response Engine
App Review Response Engine
Turns App Store and Play reviews into replies you approve and a fix list built from real quotes.
The job: turn the reviews people leave on the App Store and Google Play into a real routine — a ranked list of who to answer first, replies written and waiting on your approval, a fix list built from what users actually wrote, and an alarm when the one-stars pile up. It runs inside your own agent, on a schedule, after your app is out.
Most review handling goes wrong one of two ways. Ignore the reviews, and a bad release quietly drags your rating down while the people who would have come back move on. Or paste the same "sorry for your experience, please contact support" into every one — readers can tell, it makes an angry reviewer angrier, and both stores can treat it as spam. The obvious shortcut, a bot that posts on its own, is worse. It can repeat a reviewer's personal details, argue with them, promise a fix you cannot deliver, or drop a price into a reply — and a price in an App Store reply breaks Apple's rules outright.
What's on the tag:
- It pulls from both stores and keeps a copy of everything it has seen. Google only shows the last seven days. Miss that window and those reviews are gone for good.
- Reviews grouped by what they are about and how the person felt, each tagged with the version they were running and the exact words they used — so a complaint you already fixed never gets answered as if it were live.
- A ranking that floats the paying customers about to quit to the top of the queue. Those are the ratings worth saving.
- Two replies per review, one for each store, in the language the review was written in. Every draft is run past a filter that strips personal details, prices and promos, any ask to change a rating, and any fix you have not confirmed.
- A fix list your engineer can work straight from: a week of reviews as a table with the quote attached to each line, not a paragraph of summary.
- An alert when the one-stars spike that names the likely cause behind them, plus a plan for when to ask users to rate the app — at a moment the app just did something good for them, shown to everyone rather than only the people you think are happy. Showing it only to the happy ones is a rejection risk.
Who it's issued to: indie operators and small studios answering their own reviews, who want a set pass on a regular schedule instead of a scramble after every bad release. No per-person tool required.
Why not a free review-reply prompt? A free prompt writes a reply. It does not know that asking someone to change their rating can get you thrown off both stores, that a promo line breaks Apple's reply rules, or that Google's window shuts after seven days. This one was run start to finish on a batch of reviews — the field report shows the actual output — describes both stores' reply rules accurately in our own words, and is versioned: when a store moves a limit, the gear gets updated and your locker gets the new copy. Every reply is a draft. A human approves before anything posts.
FIELD REPORT real output, not a promise
From a documented run on 14 App Store reviews for PennyWise (a fictional budgeting app, v3.2), fed through the skill in draft-only mode — no ASC API key, so pulls/clustering/triage/drafts/roadmap all run and replies are staged for human copy-paste. Cold start: no reviews.json history, so no trailing baseline exists yet. Excerpted and trimmed to the strongest honest portions of the captured output.
Step 2 — Cluster (theme × sentiment)
| Theme | Sentiment | Count | Versions | Representative quote(s) (verbatim) |
|---|---|---|---|---|
| crash | negative | 3 | v3.2 (3) | "crashes every time I open the budgets tab. Unusable now."; "App crashes on the budgets tab since 3.2."; "Se cierra sola al abrir presupuestos. No sirve." |
| billing/paywall (bug) | negative | 1 | v3.2 | "Charged for Pro but the premium reports are still locked. Want a refund." |
| bug | negative | 2 | v3.2 (1), v3.1 (1) | "Face ID login stopped working after the update..."; "Sync is slow and sometimes double-counts transactions." |
| performance | negative | 1 | v3.2 | "the widget hasn't updated in days." |
| UX-confusion | negative | 1 | v3.0 | "Too many notifications, can't find how to turn them off." |
| missing-feature | neutral/positive | 3 | v3.2 (2), v3.1 (1) | "export to CSV would make it perfect"; "shared household budget feature"; "wish it synced with my credit union" |
| praise | positive | 3 | v3.2 (3) | "Love the weekly digest, keeps me honest."; "The new dark mode is beautiful."; "Best budgeting app I've tried, worth the Pro price." |
14/14 reviews accounted for.
Step 3 — Priority reply queue (1–3★ + churn-risk; 9 of 14)
| # | Review (excerpt) | Stars | Version | Signal | Action |
|---|---|---|---|---|---|
| 1 | "crashes every time I open the budgets tab... Unusable now" | 1★ | v3.2 | crash, current, churn ("unusable") | Reply now + engineering flag — P0 |
| 3 | "Se cierra sola al abrir presupuestos. No sirve." | 1★ | v3.2 | crash, current (dup, Spanish) | Reply now (Spanish) + engineering flag — P0 |
| 4 | "Charged for Pro but premium reports still locked. Want a refund." | 1★ | v3.2 | payment failure, refund, paying user | ESCALATED — draft + human sign-off — P0 |
| 5 | "Face ID login stopped working after the update..." | 2★ | v3.2 | auth regression, current, daily friction | Reply now — P1 |
| 6 | "Sync is slow and sometimes double-counts transactions." | 2★ | v3.1 | data-integrity bug + perf, fix unconfirmed | Reply now — P1 |
(v3.0/v3.1 reviews are NOT assumed fixed by 3.2 — no changelog/confirmation was supplied, so they stay in the open queue rather than getting a "fixed in vX" reply.)
Step 4 — Drafts (App Store, human-approval required before posting)
#1 (1★, v3.2, en):
"Thanks for reporting this — a crash on the Budgets tab after the 3.2 update is a real blocker, and we're sorry it's made the app unusable for you right now. We've reproduced the issue and a fix is actively in progress for an upcoming update. We won't stop until Budgets opens reliably again — appreciate you flagging it."
#3 (1★, v3.2, es):
"Gracias por avisarnos — sabemos que la app se cierra sola al abrir Presupuestos desde la actualización 3.2, y lamentamos que ahora mismo no te esté funcionando. Ya reproducimos el problema y estamos trabajando en una solución para una próxima actualización. Gracias por tu paciencia mientras lo resolvemos."
#4 (1★, v3.2, en) — ESCALATED — human sign-off required before posting:
"Sorry about this — being charged for Pro without the premium reports unlocking isn't okay, and we understand wanting a refund. We're looking into why the unlock didn't apply to your account. App Store purchases are refunded by Apple directly, so you can request one anytime at reportaproblem.apple.com. We're also digging into the unlock issue on our end."
#6 (2★, v3.1, en):
"Thanks for the detailed report — slow sync is annoying on its own, but transactions double-counting is the more serious part since it throws off your real budget numbers. We've flagged both, with the double-counting as the priority, and we're investigating now. Appreciate you calling this out specifically."
MUST-NOT filter check (all 14 drafts): no reviewer PII, no pricing/promo language, no solicited rating changes, no invented workarounds (the Face ID and sync drafts deliberately omit troubleshooting steps since none were buyer-confirmed), no over-promised dates, no hollow non-answers, no two near-identical drafts.
Step 6 — 1★-spike alert (cold start)
No trailing baseline yet — the relative 2× trigger can't be computed until the next run, and the report says so rather than inventing a trend. Raw this-run numbers: 4 of 14 reviews are 1★ (28.6%), 3 of those 4 (75%) are the same Budgets-tab crash cluster on v3.2. The absolute floor (default N=5) did not trip (4 < 5), but it's close and single-cause-dominated — recommendation is to expedite the P0 crash fix regardless of the formal alert state.
That is the shape of every run: a verbatim-quoted cluster table with full-count reconciliation, a churn-weighted triage queue that escalates billing to a human, drafts that stay honest about what's confirmed, an explicit MUST-NOT pass, and cold-start honesty about the metrics it cannot yet compute.
SERVICE RECORD living gear — updated as the factory learns
v1.0.0 — 2026-07-17
First issue. Post-launch App Store / Google Play review-response engine: multi-store pull, theme × sentiment clustering, 1–3★ triage, policy-safe two-variant reply drafting, quote-backed roadmap, 1★-spike alert, and rating-prompt timing spec. Policy guardrails (no auto-post, no solicited ratings, no promo in App Store replies) and the Google 7-day-window state-file requirement are baked in.
Every update ships free to owners — your locker always serves the latest version.
QUESTIONS
Does it post replies for me?›
No. It writes the replies and stacks them up. You read each one and approve it before anything goes live. If you give it a sign-in key for App Store Connect (Apple only lets the Admin or Customer Support kind post replies) or for Google Play, it walks you through the safe way to post. With no key at all it still does everything else — pulls the reviews, sorts them, ranks them, writes the drafts, builds the fix list — and you paste the approved replies in yourself.
How is this different from the pre-submit checklist?›
Different job, different day. A pre-submit checklist runs before you send the app to Apple, to get it through review. This one runs after your app is out, on a schedule, to keep your rating healthy and pull the real problems out of what people write.
Do I need a paid tool like AppFollow?›
No. It reads the stores' own free review feeds from inside your agent, so there is no per-person fee. And the rules — never post on its own, never ask anyone to change their rating, no prices or promos in an App Store reply — are built into the writing step rather than bolted on afterwards.
Will it handle reviews written in other languages?›
Yes. It works out what language a review was written in and replies in that language — a German one-star gets a German reply. If it cannot write confidently in that language, it flags the review and says why, rather than guessing or handing back a reply the author cannot read.