Steam Reviews Scraper — Exact Date Windows, No Key

Scrape reviews for any game on Steam. Give it an app id or a store URL and get back the review text, the verdict, the vote counts, and a full author block — playtime, games owned, review history, Steam profile. No API key, no login, no cookies.

Ask for a date range and get that date range. Steam's own day_range filter leaks badly — asking for 3 days returned reviews up to 3,515 days old, and a 30-day request let 33 of 60 through from outside the window — so this scraper does not use it at all. It walks reviews newest-first and enforces your window itself. Anything outside is never delivered and never charged, and the run log says how many were trimmed.

What you can filter on

Field	Type	Default	Notes
appIds	array	— (required)	Numeric ids (570) or store URLs (…/app/292030/…). Several games per run
maxItems	integer	1000	Hard cap across the whole run
language	select	all	Steam language code, or every language
reviewType	select	all	Positive only, negative only, or both
purchaseType	select	all	Bought on Steam, activated from a key, or either
onlyReviewsAfter	date	—	Start of the window, e.g. 2026-01-01
onlyReviewsBefore	date	—	End of the window; a bare date includes that whole day
includeOffTopic	boolean	false	Include bursts Steam classes as off-topic review bombing

All of these are verified live, not assumed. On The Witcher 3: negative returned 200 rows with none recommended, positive + german returned 100 with all recommended and all German, and non_steam_purchase returned 100 with none marked as a Steam purchase.

One caveat worth knowing. purchaseType only separates anything on a paid game. Every acquisition of a free-to-play title counts as a Steam purchase, so on something like Dota 2 all three settings return identical rows. That is Steam's behaviour, not a bug here.

A second one. Steam serves reviews newest-first, so a window well in the past means paging down through everything newer to reach it — a window from three weeks ago took 16 pages to fill 300 rows, where a recent one fills immediately. The upside of walking in order is that the crawl knows when it is finished: asking for a two-day window with a cap of 5,000 stopped after 6 pages and 191 reviews rather than running to the cap. The log reports both the pages walked and the rows trimmed.

What you get per review

Review: id, permalink, text, language, recommended or not, created and updated timestamps.

Engagement: helpful votes, funny votes, Steam's weighted vote score, comment count.

Context flags: bought on Steam, received free, written during early access, played mainly on Steam Deck, refunded.

Author: Steam id, display name, profile URL, avatar, games owned, reviews written, and playtime — at the time of the review, lifetime, and last two weeks — reported in hours rather than Steam's raw minutes.

Example input

{
  "appIds": ["292030"],
  "maxItems": 500,
  "language": "english",
  "reviewType": "negative",
  "onlyReviewsAfter": "2026-06-01",
  "onlyReviewsBefore": "2026-08-31"
}

Notes

• Steam's public review endpoint is used directly, so runs are fast and need no proxy.
• Reviews are deduplicated by recommendationid within a run.
• Cursor pagination stops when Steam stops advancing the cursor, rather than spinning on the last page.
• A run that finds nothing says why in its status rather than returning a silent empty dataset.