Apple App Store Reviews Scraper & Rating Intelligence

A single public Apify Actor for collecting public Apple App Store reviews by country storefront. It uses Apple iTunes Lookup for app metadata and Apple's public customer-review JSON feed for review records. Acquisition is HTTP-only: no browser, login, cookies, configured proxy, paid API, AI service, or residential proxy.

Input

Provide app_id or an apps.apple.com app_url. The Python contract requires at least one and rejects conflicting IDs. countries defaults to ["US"] and accepts at most 10 unique two-letter storefront codes. max_reviews defaults to 100 and is capped at Apple's observed maximum of 500 recent reviews per app/country. include_rating_intelligence defaults to true. The Input Schema intentionally leaves app_id and app_url optional because Apify Input Schema does not support using anyOf to express this rule.

{
  "app_id": "310633997",
  "countries": ["US", "GB"],
  "max_reviews": 100,
  "include_rating_intelligence": true
}

Output

Every valid review is written as one Dataset item with app metadata, country, stable review_id, title, text, 1–5 rating, review version/date/author, helpful vote count and sum when Apple exposes them, storefront rating metadata, and warnings. Missing optional evidence remains null; the Actor does not invent developer replies or other fields.

OUTPUT summarizes app metadata, requested/completed countries, reviews delivered, pages fetched, deterministic rating intelligence, warnings, Dataset ID, and the report reference. REPORT.html provides a responsive summary, star distribution, rating shares, country comparison, review table, warnings, and methodology. Empty feeds and nonexistent apps produce valid empty outputs.

Deterministic rating intelligence

The Actor calculates delivered-review average, 1–5 star distribution, counts and shares for negative (1–2), neutral (3), and positive (4–5) reviews, plus reviews by version. Lowest-rated version and highest/lowest country are returned only when each compared group has at least three delivered reviews; otherwise those values are null with an explicit insufficient-sample warning. No AI sentiment, topic inference, or unsupported benchmark is performed.

Pagination and reliability

Apple's public feed was observed to return 50 reviews per page for pages 1–10; page 11 returns HTTP 400. The Actor never requests beyond 10 pages or 500 reviews per country. Countries are isolated, at most four storefront requests run concurrently, every request has a 10-second timeout, and only transient failures (429, 500, 502, 503, 504, connection errors, and timeouts) receive two moderate retries. Deduplication uses country + review_id, never review text.

The feed is a legacy public surface without an SLA. Storefront data, current versions, ratings, counts, and review availability can differ by country. Helpful vote fields may be zero or absent.

Billing and free use

The prepared PPE event is review-result. One event is charged only after a valid review is written to the Dataset. Metadata, empty feeds, invalid apps, failures, discarded rows, duplicates, OUTPUT, and REPORT are not billable. The price remains unset until Cloud cost validation. The evergreen sample requires no secrets and fetches at most 100 reviews from one storefront, keeping ADWIC comfortably below five minutes under the validated acquisition profile.