App Review Product Intelligence

Turn low-rating Apple App Store and Google Play reviews into a product-improvement brief whose every finding points back to real review IDs.

This standalone Apify Actor is built for founders, product managers, UX researchers, and competitive-intelligence teams. It collects auditable raw reviews, groups recurring evidence with a fixed taxonomy, ranks issues with a published formula, extracts explicit feature requests, and compares primary-product evidence with competitor evidence.

No runtime AI. No LLM calls. No embeddings. No external AI APIs. The method is deterministic TypeScript: the same review rows produce the same themes, scores, priorities, matched terms, and recommended investigate/validate actions.

Outputs

• Default dataset: unified raw Apple and Google reviews for auditing/export.
• REPORT: complete JSON findings, feature requests, comparison signals, method, thresholds, and limitations.
• REPORT.md: concise human-readable report with evidence links.
• SUMMARY: success/partial/failure status and per-app diagnostics.
• PROGRESS-{platform}-{appId}: latest bounded scrape progress.

Each raw review includes platform, primary/competitor role, app metadata, discovery provenance, review ID/URL/title/text/rating/date/version, helpful count, developer reply where available, locale, and scrape time. Apple fields that its public feed does not expose are null; Google review titles are null.

Input

In Apify Console, paste one store link or app ID per line—no JSON objects:

• Your app links are marked primary.
• Competitor app links are marked competitor.
• Discover competitors by search searches both stores and marks matches competitor.

Apple URLs and numeric IDs are auto-detected. Google Play URLs and package IDs are auto-detected.

API input uses the same simple string lists:

{
  "primaryApps": [
    "https://apps.apple.com/us/app/spotify-music-and-podcasts/id324684580"
  ],
  "competitorApps": [
    "https://play.google.com/store/apps/details?id=com.google.android.apps.youtube.music"
  ],
  "searchTerms": [
    "music streaming"
  ],
  "minRating": 1,
  "maxRating": 2,
  "maxReviewsPerApp": 100,
  "country": "us",
  "language": "en"
}

Search discovery uses Apple's public iTunes Search API and Google Play search. Direct apps keep their list role; discovered apps are competitors.

Defaults focus on actionable negative evidence: ratings 1–2, 500 matching reviews per app, 3 apps per search, and 10 total apps. Hard caps are 25 total apps, 10 apps per search, and 2,000 matching reviews per app. Apple is additionally capped to the 10 pages exposed by its public RSS endpoint. Requests time out after 15 seconds, temporary failures receive bounded retries, and 404 responses are not retried.

See examples/input.json.

Deterministic method

The Actor normalizes review title and text, then matches transparent phrases/regular expressions for:

• crashes/performance
• login/account
• subscriptions/payments
• ads
• update regressions
• usability/navigation
• sync/data loss
• notifications
• support
• privacy
• battery
• connectivity
• localization/accessibility
• missing features

It separately extracts explicit requests beginning with phrases such as please add, wish, needs, should have, would be better if, and missing.

For each app/theme:

score = round(
  share of negative reviews × 60
  + min(occurrences / 5, 1) × 25
  + min(helpful votes / 20, 1) × 15
)

Priority thresholds are critical 80–100, high 60–79, medium 35–59, and low 0–34. Confidence is high at 5+ matching reviews, medium at 2–4, and low at 1. Recommendations come from a fixed map and are deliberately framed as investigations or validations. The Actor does not claim root cause, sentiment understanding, or an invented fix.

Store limitations

• Apple customer-review RSS is public but bounded to at most 10 pages. Availability varies by storefront. It does not expose helpful counts or developer replies.
• Google and Apple control what their public endpoints expose and may omit, repeat, or reorder reviews.
• Locale selection changes the visible sample. Results are not all worldwide reviews.
• Lexical matching can miss paraphrases or match ambiguous language. Read the evidence quotes and raw rows.
• Competitor signals compare only the collected low-rating sample; they are not market-wide prevalence claims.

Partial success is preserved: one failed app does not discard successful apps. If no target resolves, every app fails, or zero useful review rows are produced, the Actor writes a diagnostic SUMMARY and fails rather than silently returning an empty success.