How GAMESLIKE ranks games

Player insights and aspect history

Player insights summarize a screened sample of complete English Steam reviews from a recent window. A model classifies a fixed set of gameplay aspects and checks the supporting sources for each retained paraphrase. These results have not been independently verified by a human.

Each aspect bar shows how far the sampled opinion leans positive or critical, weighted across that aspect’s positive, mixed and negative opinions. Neutral mentions, absence and unclassified reviews are left out of the lean. A small sample can move substantially, and the panel marks the ones it draws from.

Each successful analysis is a dated snapshot containing both its text and aspect values. Earlier results remain dated even when private supporting material expires, explicitly withdrawn results are removed, and an analysis is refreshed once enough new review activity has accumulated.

Comparisons use compatible methods and sample windows. Rolling windows may overlap, and changes describe the sampled reviews rather than proving that a patch caused an improvement or decline.

GamesLike Review

A GamesLike Review is an AI-assisted summary of a sample of Steam player reviews, not a first-hand playtest or a GamesLike score. It does not cover every review. Sample details and dates appear on the game page.

Where available, charts summarize opinions on individual aspects of a game and changes over time. These are estimates from sampled reviews; uncertain comparisons are omitted. More review activity does not necessarily mean better sentiment.

This feature remains inactive pending data-processing safeguards. Real Steam review text is not currently sent to an external model provider. See our Privacy Policy for data use and safeguards.

Groups and game nights

Group recommendations use games a participating person played for at least five hours and games they explicitly mark Loved. A Not for me choice vetoes that exact title for the session. Playtime is evidence of engagement, not proof of enjoyment, and a Group fit number is not an enjoyment probability.

Only selected attendees contribute taste, ownership and votes. Fit numbers and ownership coverage are shared with the group; contributing titles, hours and private preference lists are shown only to the person they belong to. Session choices require an explicit in vote from every attendee.

Capacity evidence is a small dated editorial set for Windows online co-op. Unknown capacity remains unknown unless a person explicitly includes unverified games. Cross-play, server availability, installation and actual play are not inferred.

Review discussions

We automatically summarize recurring topics and clear opinions from a sample of recent English Steam reviews. Each panel shows its sample size, dates, and filters. Samples may miss wider opinion, and automated analysis can miss context, so uncertain results are left out.

Completion estimates

We calculate a rough median from completion times mentioned in retained English Steam reviews. An estimate appears only when at least ten distinct reviews report a usable completed-run time; the info box shows how many contributed. Each review contributes once, and a reported range contributes its midpoint. This is a sample of reported completions, not every player's playtime.

Predictions, clear repeat runs, speedruns, DLC-only and completionist times are excluded. Automatic extraction can still miss context, and playstyles or optional content can change the time. We use local text analysis, rather than Steam's lifetime playtime counter or HowLongToBeat's measurements. HowLongToBeat links provide more detailed times and playstyle categories.

Changed inputs are normally recalculated at most once per day per game; unchanged inputs are skipped. Retained source removals and review objections withdraw affected estimates before recalculation. We publish the estimate and contributing count without individual review text, identities or review links.

Dataset credits

Archived game information includes the FronkonGames Steam Games Dataset (MIT), Donald Fountain’s Steam Dataset 2025 and Nik Davis’s Steam Store Games (both CC BY 4.0).

Historical price data includes the Steam Games Dataset by Dulakshi Wannigamage, Michael Barlow, Erandi Lakshika and Kathryn Kasmarik (CC BY 4.0). We clean and combine these records and use selected fields to fill gaps. Archive coverage varies; snapshots do not establish continuous price history. See the CC BY 4.0 license and dataset notices.