How to analyse Steam reviews to find what players love and hate
To analyse Steam reviews, pull every review in every language with its metadata, separate reviews that count toward the score from key activations and free copies, then segment by playtime, language and date. Tag each review with themes such as performance, controls, difficulty, value and content length, count how often each appears in positive and negative reviews, read the reviews behind each theme and rank fixes by frequency, how early they hit and cost to fix.
Why are Steam reviews the best free player research?
Every Steam review is written by someone who chose to play, usually paid, and then said in their own words whether they recommend the game. Each one comes with metadata surveys rarely get: how long the reviewer had played, whether they bought it on Steam, which language they wrote in and whether the game was in Early Access at the time. Read at scale, reviews tell you what keeps players and what drives them away, for your own game and for every competitor.
The headline score is the least useful part. "Very Positive" says players are broadly happy; it does not say why, or what the unhappy ones want fixed. That comes from breaking reviews into themes.
What data does a Steam review contain?
Valve exposes reviews through the IUserReviewsService GetAppReviews endpoint (the older appreviews endpoint is now marked deprecated). The fields that matter most for analysis are:
| Field | What it tells you |
|---|---|
voted_up | Recommended or not |
author.playtime_at_review | Minutes played when the review was written |
author.playtime_forever | Minutes played in total, including after the review |
steam_purchase | Whether the reviewer bought the game on Steam |
received_for_free | Whether the reviewer ticked the box saying they got it free |
written_during_early_access | Whether the review predates full release |
primarily_steam_deck | Whether the reviewer mainly played on Steam Deck |
votes_up | How many other players found the review helpful |
Two defaults catch people out. The endpoint returns only reviews from Steam purchases unless you change purchase_type, and it returns English only unless you set the language list to "all". It also filters off-topic review activity by default. It returns at most 100 reviews per page, so pulling a full history means following the cursor value page by page.
How do you analyse Steam reviews step by step?
- Pull every review, in every language. Store the full text and all metadata. For a game with tens of thousands of reviews this takes a few minutes of paging.
- Separate the reviews that count. According to Valve's review documentation, only reviews from accounts that purchased on Steam count toward the score; key activations are visible but excluded. Valve made that change in September 2016. Flag key activations and reviews marked as received for free, then report the score both with and without them.
- Segment by playtime. Steam allows refunds within 14 days and under two hours of play, so reviews written under two hours often reflect a refund decision: performance, onboarding, controls or price. Reviews written after 20 or more hours are about depth, balance, endgame and content. Use bands such as under 2 hours, 2 to 10, 10 to 50 and over 50.
- Segment by language and date. A theme that dominates Simplified Chinese or Russian reviews but not English ones often points to translation quality or regional pricing. Splitting by date shows whether a patch helped and isolates sudden spikes.
- Tag every review with themes. Build a short codebook of 10 to 15 themes (see below) and assign each review one or more. Keep a theme for "other" and read it periodically to find themes you missed.
- Count and weight. For each theme, calculate the share of positive and negative reviews that mention it. Weight by helpful votes as a second view, because a complaint 500 players agreed with matters more than one nobody saw.
- Read the actual reviews behind each theme. Numbers tell you where to look; the words tell you what to do. Pull five to ten representative reviews per theme.
- Turn themes into a ranked fix list. Score each negative theme on how often it appears, how early it hits (lower playtime means more refunds), and how cheap it is to fix.
The ReviewPulse review analysis runs this process automatically for any Steam game: themes players love and hate, split by hours played and language, the exact reviews behind each theme, issues to fix first and a score without free-key reviews.
Which themes should you track in game reviews?
| Theme | Typical phrases | Usually shows up |
|---|---|---|
| Performance and stability | "stutters", "crashes", "FPS drops" | Under 2 hours |
| Controls and feel | "clunky", "floaty", "great game feel" | Under 2 hours |
| Onboarding and difficulty | "confusing", "no tutorial", "too hard" | Under 10 hours |
| Price and value | "worth every penny", "too expensive for the content" | Any |
| Content length and replayability | "finished in 4 hours", "one more run" | 10 hours and up |
| Balance and progression | "grindy", "overpowered build" | 10 hours and up |
| Story and writing | "characters", "ending" | Any |
| Art, music and atmosphere | "beautiful", "soundtrack" | Any |
| Multiplayer and netcode | "desync", "empty lobbies", "fun with friends" | Any |
| Localisation | "bad translation", "no Chinese" | Non-English reviews |
| Developer support | "abandoned", "devs listen" | Long-running games |
Adjust the codebook to your genre. A co-op horror game needs separate themes for proximity voice and monster design; a city builder needs one for traffic simulation.
How do you handle review bombs and sudden spikes?
Look at reviews over time before reading anything into a score drop. A sudden spike of negative reviews about a business decision is a different problem from a slow decline about gameplay. In May 2024, Helldivers 2 saw a surge of negative Steam reviews after Sony announced that PC players would need to link a PlayStation Network account; Sony made linking optional three days later. Those reviews said a lot about the account policy and almost nothing about the game's design.
Valve's own rules help here. Its documentation says that when a surge of reviews is unrelated to the product, Steam can remove "the abnormal period of volume" from the score, and reviews flagged as off-topic have their visibility reduced. When analysing, tag spikes separately so they do not distort your theme counts.
What can rival reviews tell you?
Running the same analysis on your three closest competitors is often more valuable than running it on your own game, especially before launch. Their negative themes are opportunities; their positive themes are the minimum players expect. If two rivals are repeatedly criticised for short campaigns and missing controller support, those are features you can lead with. The rival player insights view does this across a competitor set at once, and rival socials and community talk adds what players say outside Steam.
Can a game recover its Steam reviews?
Yes, though it takes sustained work. No Man's Sky was heavily criticised at its 2016 launch; after years of free updates, its all-time Steam rating moved to "Very Positive", as TweakTown reported in late 2024. Steam shows both a recent score (past 30 days) and a lifetime score, so fixes show up in the recent score first.
Valve also notes that a review score of Mixed or better is not a factor in algorithmic visibility; only scores below 40% make a game less likely to be featured. The score still shapes whether a shopper buys, though, so fixing early-playtime problems pays off twice.
How should developers respond to reviews?
Valve's documentation lets developers post an official response under any review, but advises against responding to every review or using responses to argue with players. A practical rule: respond when you can say something specific and useful, such as "fixed in patch 1.0.3" or "here is how to change the keybinding", and let the patch notes do the rest. Players notice when the top complaint in your review themes disappears from new reviews after an update. That is the clearest sign your analysis worked.
Frequently asked questions
Do free key reviews count toward a Steam review score?
No. Since September 2016, reviews from accounts that activated a game with a Steam product key are still shown but do not count toward the review score. Only reviews from Steam purchases are included.
How many Steam reviews do I need for a useful analysis?
A few hundred reviews is enough to see the main themes. Below about 100, read every review by hand instead of counting themes, because a handful of reviews can swing the percentages.
Why segment Steam reviews by playtime?
Because players at different playtimes complain about different things. Reviews under two hours often reflect refund-window problems such as performance or onboarding, while reviews after many hours focus on balance, endgame and content.
Does a low review score hurt Steam visibility?
According to Valve, review scores of Mixed or above are not a factor in algorithmic visibility. Games below 40% positive are less likely to be featured. The score still affects whether shoppers buy.
Try it on a real game
Paste any Steam link, app ID or game name and get rivals, review themes, pricing by country and a store page teardown in minutes. Free to start.