How We Track Steam Reviews
Every day at 03:00 UTC, we record four numbers for each of the 29,950 games in our active tracking pool: total review count, positive count, negative count, and Steam's score category ("Very Positive", "Mixed", and so on). Since May 2026 that has added up to more than 1.6 million snapshots.
Everything we publish — the weekly Review Watch, case studies like Slay the Spire 2's vanishing reviews or MECCHA CHAMELEON's climbing launch score — is built on those daily records. This page explains exactly how they're collected, what we derive from them, and where the limits are. It's the page our "note on method" sections point to.
Where the data comes from
All review data comes from Steam's public store endpoints — the same data anyone can see on a game's store page, fetched programmatically and politely (rate-limited well below Steam's published guidelines). We collect totals across all languages and purchase types. We don't scrape private data, and we don't use any privileged API access.
A game's positive ratio as we report it is positive ÷ (positive + negative), computed from the counts. We also record Steam's own score category label at snapshot time, which lets us detect the exact day a game crosses a threshold — say, from "Mixed" to "Mostly Positive".
How the tracking pool is built
We index roughly 166,000 Steam apps in total, but snapshotting all of them daily would be wasteful — most games get no reviews in any given week. So the pool has tiers:
Active (29,950 games as of July 6, 2026) — snapshotted daily. Games are promoted here once they have at least 50 reviews.
Observation — newly discovered games that haven't reached the 50-review bar yet. We check them daily and promote them the moment they qualify; games that stay silent for 30 days are demoted.
Indexed — the long tail of the catalog. Known to us, not yet tracked daily.
New games enter through several channels: a daily scan of newly registered Steam apps, weekly sweeps of Steam's top-seller and popularity charts, genre seed lists, and user activity on this site — when you search for or view a game on SteamSense that isn't tracked yet, that alone pulls it into the pipeline. In practice a new release with real traction starts accumulating daily history within a day or two of getting its first wave of reviews.
One honest caveat: discovery isn't perfect. TBH: Task Bar Hero — briefly one of the most-played games on Steam — sat in our index untracked for weeks because it fell between promotion rules (we've since fixed those rules and backfilled what Steam's public weekly histogram allows). When a game's tracking history starts later than its release date, that's why.
What we derive from the snapshots
Daily deltas. The difference between today's and yesterday's totals tells us how many reviews arrived — or disappeared — in each 24-hour window. Steam's store page only ever shows you the current state; the deltas are where the stories live.
Vanished reviews. When a game's total review count decreases between snapshots, reviews were removed. Steam publishes no deletion log, so subtraction is the only way to see this happen. Important: we can observe the change, not the cause. A drop can mean players deleted their own reviews, Valve moderation removed them, or off-topic review activity was purged. From the outside these are indistinguishable, which is why our write-ups describe removals without attributing individual deletions.
Anomaly signals. Each day we compare every game's review inflow to its own 7-day baseline. Sustained deviations produce signals — volume spikes, volume drops, unusual positive-ratio shifts, score category changes — which feed the Trends page and our weekly column. A game needs at least three days of history before we'll score it, so freshly tracked games show as "new entries" first.
AI review analysis. Separately from the counting pipeline, game pages on SteamSense offer AI-generated summaries of what reviewers are actually saying. Those are built by sampling a game's most helpful reviews (up to ~100, by Steam's own helpfulness votes) and running them through a language model for sentiment and topic extraction. Summaries are cached and refreshed periodically — they describe the sampled reviews, not every review ever written.
Why our numbers may differ slightly from the store page
Three reasons, all boring:
- Timing. We snapshot once a day at 03:00 UTC; the store page is live. For a game gaining thousands of reviews daily, the two will always be a few hours apart.
- Totals. We compute total = positive + negative. Steam's own "total reviews" figure occasionally differs from that sum (it can include reviews excluded from scoring), so we prefer the sum for internal consistency.
- Filters. The score you see on a store page can reflect your language and purchase-type defaults. Our numbers are all-languages, all-purchase-types.
None of these differences exceed rounding-error territory for analysis purposes, but if you diff our page against Steam's at a random moment, you'll spot them.
What this data can't tell you
Daily granularity means we can't see intraday shapes — a review bomb that starts and gets moderated within the same day looks like one net number to us. Tracking starts at discovery, so pre-discovery history is limited to what Steam's public weekly histogram preserves (weekly resolution, no deletion visibility). And counts are counts: they measure people who wrote reviews, which is a small, self-selected slice of any player base — as the MECCHA CHAMELEON case shows, sometimes as little as 0.3% of reported buyers.
We think the honest posture is to publish what the data shows, flag what it can't show, and let the gaps be part of the story.
FAQ
How does SteamSense get its data? From Steam's public store endpoints, collected once daily at 03:00 UTC for every game in the active pool, plus discovery scans for new games. No scraping of private data, no privileged access.
Does SteamSense use the official Steam API? We use Steam's publicly documented and publicly accessible web endpoints — the same data shown on store pages — under Valve's published rate limits. Valve does not sponsor or endorse SteamSense.
Why do your numbers differ slightly from the Steam store page? Snapshot timing (once daily vs. live), total definition (we use positive + negative), and filter defaults (we count all languages and purchase types). Differences are small and don't affect trend analysis.
How often is the data updated? Snapshots run daily. The Review Watch column is published every Monday. AI summaries on game pages refresh periodically.
Can I get a game added to tracking? Yes — search for it on steamsense.app. Viewing an untracked game pulls it into the pipeline automatically; if it has 50+ reviews it will be promoted to daily tracking on the next cycle.
Questions about the methodology? Contact us. Data available on every game page at steamsense.app.