Steam Reviews Scraper + AI Analysis (No Login)

**Steam reviews scraper** — Scrape unlimited Steam reviews and detect review bombs, update backlash, and the sentiment gap between new and veteran players. No login required.

Scrape unlimited Steam reviews for any game by App ID via Valve's official public API — no login, no proxy needed. On top of raw reviews you get optional AI analyses: review bombs timeline, playtime segments, pain points, feature requests, and a headline summary.

Most Steam scrapers hand you raw rows. This one ships the analysis you were going to build anyway: review-bomb detection, playtime-segment sentiment, complaint themes — computed deterministically, no LLM tax.

---

Review bombs — the part nobody else does

Every studio eventually asks "what happened in that week?" This actor answers it directly, with the receipts:

{
  "window": "2024-05",
  "reviews": 2529,
  "negativeShare": 0.953,
  "deltaVsBaseline": 0.21,
  "negativeLanguages": [
    { "language": "english", "reviews": 2049 },
    { "language": "german", "reviews": 84 }
  ],
  "exampleReview": "With today's announcement that Take-Two has laid off the entirety of Intercept Games…"
}

That's real output (Kerbal Space Program 2, 31,874 reviews). The detector flagged the exact month the studio-layoff news broke, quoted the top-voted review from that window, and showed the bomb was English-speaking — not a regional pile-on.

Detection is statistical, not vibes: a window counts as a bomb only when its negative share beats the game's lifetime baseline by 10+ points at the Wilson lower bound, so a loud week with 12 reviews can't fake it. You also get the full month-by-month (or week-by-week) timeline to chart.

Valve hides reviews from periods it marked as off-topic review activity. Turn on includeOffTopicReviews to analyse those too.

Playtime segments — who exactly is unhappy

Steam reviews carry hours-played at review time. The actor splits sentiment by segment:

Segment	What it tells you
under 2h (refund window)	Your first impression — these reviewers can still refund
2–10h / 10–50h / 50–200h	The mid-game
200h+	Your veterans

Plus received for free vs paid, and early access vs after release splits. If newcomers sit at 45% positive while veterans sit at 89%, you have an onboarding problem, not a content problem — and now you can prove it.

No count cap

Apple caps its public feed at 200 reviews per storefront. Steam doesn't: this actor walks the cursor feed from newest to oldest until your limit, your sinceDate, or the end.

Measured: 31,874 of a game's 31,876 total reviews in one run (the feed itself withholds a couple), zero duplicates. sinceDate stops collection early — reviews arrive newest-first — so weekly incremental pulls stay small and cheap.

Analyses

Each is billed once per run, not per review.

Analysis	What you get
reviewBombs	Per-window timeline, flagged bomb windows with language mix and a top-voted example quote
playtimeSegments	Sentiment by hours-played, newcomers-vs-veterans verdict, free-vs-paid and EA-vs-release splits
painPoints	What negative reviewers actually complain about, scored by lift vs the full corpus, with example quotes
featureRequests	What players ask for, extracted from request-language reviews
summary	Headline numbers incl. store-wide totals and score label straight from Steam

Analysis is deterministic — no LLM in the loop. Same input, same output, every time. No hallucinated themes, no per-token cost passed on to you.

Use Cases

• Review-bomb crisis triage: catch a bomb the day it starts with the Wilson-bound-flagged window and a top-voted quote — the receipts you need before responding publicly.
• Onboarding vs. veteran diagnosis: split sentiment by playtime segment to see whether under-2h newcomers or 200h+ veterans are the unhappy ones — a different fix either way.
• Giveaway / bundle impact check: compare received for free vs paid reviewers to see whether a promo shifted your review average up or down.
• Feature roadmap input: mine featureRequests for what players are actually asking for, in their own words.

Steam reviews scraper — input

{
  "appIds": ["954850"],
  "maxReviewsPerApp": 50000,
  "languages": ["all"],
  "sinceDate": "2026-01-01",
  "includeOffTopicReviews": true,
  "aiAnalysis": ["reviewBombs", "playtimeSegments", "painPoints"]
}

appIds accepts numeric IDs or full store URLs. No IDs handy? Pass searchTerm instead and the actor resolves it.

Steam reviews scraper — output fields

Reviews (type: "review"), one dataset item each:

Field
appId appName language	which game, which language
review votedUp timestampCreated	the review
playtimeAtReviewHours playtimeForeverHours	hours played at review / now
steamPurchase receivedForFree refunded earlyAccess steamDeck	context flags
votesUp votesFunny weightedVoteScore authorName url	metadata

Analyses arrive as separate items with type: "analysis".

Pricing

Event	Price
Actor start	$0.00005
Per review	$0.0002 ($0.20 per 1,000)
Per analysis	$0.05 (summary: $0.03)

The full 31,874-review history above with all five analyses: about $6.60. A weekly 1,000-review incremental pull with two analyses: about $0.30.

Budget-aware: if your run's Max total charge (USD) can't cover both the reviews and the analyses you selected, the actor collects fewer reviews instead of starving the analyses. Measured: a $0.50-capped run collected 1,348 reviews and still delivered all five analyses — the run completes instead of aborting mid-way.

Why this one won't rot

It reads the same public review endpoint Steam's own store pages use — documented by Valve at partner.steamgames.com. No login, no headless browser, no proxy rotation, no CAPTCHA arms race. Nothing to break when a site ships a redesign.

No personal data beyond what Steam already publishes: reviewer names are public display names, and the actor adds nothing on top.

Notes

• Some pages legitimately return fewer than 100 reviews mid-stream — that's Steam's language-shard merging, not the end of the feed. The actor keeps walking the cursor; it stops only on an empty page or a repeated cursor.
• languages: ["all"] is the honest default for bomb analysis — bombs are often concentrated in one language, and filtering it out hides the story.

---

Built by Backyard Tools.