Fewer Than 0.5% of Steam Reviews Get a Developer Response. Here Is What Happens to the Ones That Do.

Only 12.7% of Steam reviews are ever updated at all

This is the number that sets expectations correctly, and no competitor leads with it.

"Just 12.7% of Steam user reviews are ever updated."

Seven in eight reviews are written once and never touched again. A developer response roughly doubles the odds — "reviews that have developer responses are 2x as likely to be updated" — but doubling a small number gives you a slightly less small number.

Which means most of your score movement will come from new reviews, not recovered ones. Responses are a lever on the margin: a good one, a cheap one, but a marginal one. What makes it worth pulling is that flips count double — a flipped negative leaves the negative column and joins the positive column, where a new positive only adds one and grows the denominator.

The arithmetic, worked

Say you have 500 reviews at 66% positive: 330 positive, 170 negative. Mixed. You respond to all 170 negatives.

• 21.6% come back and edit: about 37 reviewers.
• 55.9% of those edits go positive: about 21 flips — the 12% figure, applied to 170.
• New score: 351 of 500 = 70.2%. You cross into Mostly Positive by two tenths of a point.

A genuine tier change with no new reviews — and the best case. It assumes you replied to every negative, no edit went the other way, and no new negatives arrived while you worked. Shade it down and the realistic version reads: clearing your negative backlog is worth a few percentage points, and a few percentage points is exactly the width of a tier boundary when you are sitting near one.

Valve has never published the thresholds. We derived them from the labels Steam's own review API returned across 747 games in July 2026:

Label	Positive %	Minimum reviews
Overwhelmingly Positive	95%+	500
Very Positive	80%+	50
Positive	80%+	10-49
Mostly Positive	70-79.99%	10
Mixed	40-69.99%	10
Mostly Negative	20-39.99%	10
Negative / Very Negative / Overwhelmingly Negative	under 20%	10 / 50 / 500

The 80% line is exact. The highest "Mostly Positive" we observed was 79.984%; the lowest "Very Positive" was exactly 80.000%. One review crosses it. What each band does to a shopper is broken down in Steam review score tiers explained.

Now the counterweight. If you cannot flip your way up, climbing costs this many consecutive positive reviews:

Move	100 reviews	500	1,000	5,000
Mixed 65% → Mostly Positive (70%)	17	84	167	834
Mostly Positive 75% → Very Positive (80%)	26	126	251	1,251
Mostly Negative 35% → Mixed (40%)	9	42	84	417
Very Positive 88% → Overwhelmingly Positive (95%)	140	700	1,400	7,000

Those columns assume a perfect cohort. Realistically it is worse: at 500 reviews and 75% positive, Very Positive takes 126 straight positives, or 251 new reviews arriving at a 90% rate, or 501 at 85%. Below 85% incoming it is unreachable, because a cohort at or under your target can never pull you to the target.

Two things follow. Cost scales linearly with the reviews you already have, so acting at 300 is worth roughly ten times acting at 3,000. And Steam's Recent score uses a rolling 30-day window, so it moves far faster than lifetime — the realistic short-term lever, and the one the stuck-at-Mixed playbook is built around.

The caveats the analyst published about his own numbers

This is the part that gets stripped out every time these figures get repeated, and it is the reason to trust them at all. His own methodological note: "There is some fuzzy math here… we can't look at whether a given review has flipped, we can only infer based on surrounding variables." Steam's API does not expose edit history. Nobody watched a review change. Everything above is inference from a snapshot.

Four more:

• The 55.9% is a filtered subset. It covers roughly 1,000 games with at least 50 developer responses that produced an update. That is not Steam-wide, and games with 50+ developer responses are unusual games.
• Selection bias, admitted and pointed the other way. Responses in smaller games are more likely to get updates and those skew positive, so the author's view is that the figure likely understates impact. Either way it is a bias, not a control.
• The "reviews drift negative without interaction" finding is cross-sectional. It compares different groups of reviews, not the same reviews over time, and the author disclaims causality: "we can't rule out the idea that users who update reviews simply leave lower review scores… It is likely that reviews trend negative over time." The gap is roughly 8-10 percentage points depending on the cut. Read it as an association, not a decay curve.
• The dataset is about 11,000 games. Steam now ships close to 20,000 per year. This is a large sample of an older Steam.

The analysis also does not examine response quality, response timing, long-term effects, or sales impact — so anyone telling you it proves a particular reply style works is extrapolating past the evidence. None of which makes the numbers useless. It makes them what they are: the best public measurement anybody has taken of a thing almost nobody does, with the error bars visible.

Valve's counterweight, which is real

Read the Steamworks documentation and you will find Valve arguing, in writing, for less of this rather than more.

"Though it may be tempting, not every review needs to be responded to. A developer response will frequently draw more attention than the original statement, potentially turning a small issue into a much larger community discussion. It's also not a good idea to use this feature to refute customer opinions."

"Don't argue with reviewers. Let your community answer each others' questions."

That is a genuine tension with the data above and it should not be smoothed over. The data says responding to negatives correlates with reviews improving. Valve says a response amplifies the review it attaches to and often makes things worse. Both are true, because they are describing different reviews.

A response to a fixable, specific complaint — a crash on a particular GPU, a missing keybind, a save bug you patched — is new information, and amplifying it helps, because the amplified version now contains the fix. A response to an opinion — the combat feels floaty, the price is too high, the art style is not for me — is a rebuttal, and amplifying a rebuttal is how a two-line review becomes a forum thread.

The rule that satisfies both: respond substantively where there is something to say that the reviewer does not already know. Everywhere else, do not. That is not a compromise; it is the only reading under which Valve and the data agree. The triage version, category by category, is in the response framework; the sentence-level patterns are in the response templates.

The free recovery mechanism nobody mentions

Buried in the same Steamworks page is a feature we have not found mentioned in a single competing article:

"If a user leaves a negative review for a game and subsequently continues to log substantial playtime, Steam may prompt them to update their review."

Valve re-prompts negative reviewers who keep playing. A recovery channel that costs nothing, requires no reply, and rewards exactly one behaviour: shipping something that keeps people playing after they were annoyed. It is also the strongest evidence available that scores are recoverable, because it is Valve saying so about its own client.

One more mechanic while you are in there: only accounts that purchased the game on Steam count toward the score. Key activations, press copies, free weekends and Family Sharing produce reviews that display but do not count — which is why the denominator in your head is usually wrong.

What a better score does not buy you

This article previously claimed that "Steam's review tier system controls how the algorithm treats your game across the entire store." Valve says otherwise, in writing, on two documentation pages:

"As long as your game's reviews are Mixed or above (40%+), review score is not a factor in algorithmic visibility. If your game dips below 40% into Mostly Negative, your game will be less likely to be featured."

The same page rules out wishlists, store page traffic and conversion rate as placement factors too. What drives it is whether people are buying and playing, plus language support and tags.

So the 40% line is a real cliff, and above it your score buys no algorithmic placement whatsoever. Chris Zukowski, measuring independently for years, lands in the same place: "the Steam algorithm does not 'penalize' games for having 'mixed' reviews. They only hide your game if your review scores are Mostly Negative or lower."

If you have read that responding to reviews will get Steam to promote your game — including on this site, before today — that was wrong.

What it does buy you

Three things, all of which run through humans rather than the algorithm.

Conversion on traffic you already have. Gamesight's Q1 2026 performance marketing report found that moving from Mixed to Very Positive "can as much as triple" a game's advertising conversion rate. Three caveats travel with it: it is drawn from paid acquisition, not organic store traffic; free-to-play titles in the same data showed no measurable impact, which the report attributes to the absence of a paywall; and the primary document is gated behind a lead form, published by an ad-tech vendor selling ad technology to game marketers. Directionally useful, and a vendor report. Both true.

The shopper's decision. Less glamorous, better sourced. A GameDiscoverCo player survey in July 2026 found 40.5% of players always check reviews before buying and 30.8% said a Mixed-or-below score would negatively influence their purchase. Zukowski, from the other side of his own argument: "developers who have a game that oscillates between 'mixed' and 'positive' say that they can see their sales dip when it hits 'mixed.'"

Human curation. Press, streamers, publishers, festival juries and bundle curators use the tier label as a first-pass filter. None of it is algorithmic; all of it is gated on the same number.

Reviews per sale: the figure to stop quoting

The old rule of thumb — one review per 40 to 60 copies sold, median 58 — is a 2020 measurement that was already breaking inside the article that produced it. GameDiscoverCo measured 237 games in August 2020 and got an average of 63, median 58. Split by release year, the same dataset showed the ratio collapsing: pre-2017 games averaged 81, 2019 releases 57, 2020 releases 41. The cause is not mysterious. In late October 2019 Steam started prompting players to review games from inside the client, and reviews per sale rose permanently.

The 2022 revision, across 200+ games released 2020-2022, is the current figure — a range, not a constant:

Lifetime units sold	Copies per review
Under 1,000	~20x
1,000-9,999	~36x
10,000-49,999	~49x
50,000-99,999	~59x
100,000+	~48x
All 2022 releases	~29x median

No credible primary update has been published since. If you see "the Boxleiter number averaged around 63 in 2025," you are looking at the 2020 figure with a fresh date stamped on it.

For your revenue math: a small game reviewing at 20x and a mid-size game at 59x are not in the same business, and neither is at 58x. Multiply by the row matching your sales band and treat the output as an order of magnitude, not a forecast. For your own numbers rather than a worked example, run a free review audit — live score, tier position, distance to the next threshold.

What developers who actually do this report

The platform-wide data is inference. The first-hand accounts are small samples with real names attached, pointing the same direction.

Nicholas Laborde, in Game Developer in 2016, describes the cost of a Mostly Negative launch — "it's hard not to let the stress creep in, especially as our rating plummets" — and then the outcome, once he started talking to reviewers individually:

"nearly every single player I spoke with who left a negative review had changed their review either from negative to positive"

Nine years later, Digital Cybercherries ran the same play publicly after DON'T SCREAM TOGETHER launched at roughly 23% positive. Their post-mortem, via PCGamesN in December 2025: "People do not lose trust because a game has problems. They lose trust when developers pretend those problems are not there. So we decided to face everything head on." They shipped same-day hotfixes and replied individually — "going through each review from players affected and letting you know that the fix is now live" — and reported that "some even updated their reviews after seeing how quickly we responded." They were equally honest about the cost: "Seeing the negative reviews has been really hard… I'm not going to pretend it hasn't hit us emotionally."

Then the case that shows the downside, because Valve's warning is not hypothetical. In May 2026, Square Glade Games replied to negative reviews of their van-life sim with this:

"Feel free to send a support request to the Steam support to get a full refund on your purchase. If you do so, we would appreciate if you would update or remove your negative review."

It read as buying off criticism, generated more negative reviews than it removed, and ended in an apology: "the 24 hours since the launch have been incredibly overwhelming. While we started with a lot of excitement, the reality of a launch brings a lot of pressure and intense emotions."

That is the whole mechanism. Nobody writes their response policy in advance, so the first replies get written in the worst possible emotional state by someone awake for thirty hours. The wider picture is in games that respond vs. games that don't.

Why almost nobody does it

Four reasons, and only one of them is laziness.

The advice tells you not to. The most-linked guide to responding to Steam reviews, from Game Developer in 2022, opens the door and closes it again: "The only downside to replying to a negative review is you. If you feel you're the type of person who can't take criticism well… then I highly recommend staying away from negative reviews." That excludes precisely the developers who most need a written process. You do not need to be calm. You need a template and a rule about what you do not reply to.

The emotional cost is not abstract. Taylor McCue, on itch.io: "it was pretty upsetting to read and made me afraid to keep releasing games." Nobody measures that outcome, because it shows up as a game that never gets made.

The feedback loop is invisible. You reply, the player quietly edits two days later, and Steam tells you nothing — there is no native alert when your score moves. The corpus is unreadable at scale too: the store page surfaces the most helpful-voted reviews, not the most frequent complaints, so developers fix the loud thing and miss the common thing.

The window is short. Chris Hanney of Creature, at the GDC 2023 Community Management Summit: "There is sometimes a precious half hour of time between a player experiencing a complication in play time and posting a negative review of your game." The review is the second thing a frustrated player does. The first is a forum post or a support ticket, and nobody watches that channel fast enough.

How to act on this

Strip out everything unfalsifiable and this is what is left standing.

• Fewer than 0.5% of Steam reviews get a developer response, half of those from 79 games. The bar is on the floor.
• 21.6% of responded-to negative reviewers come back and edit, against 8.5% of positive reviewers — so your effort goes roughly two and a half times further on negatives.
• About 12% of responded-to negatives end up positive. Not 21.6%, not 55.9%. Twelve percent, derived, with fuzzy math the author flagged himself.
• Only 12.7% of reviews are ever updated. New reviews do most of the work; responses are a lever on the margin, and a flip moves the ratio twice as far as a new positive.
• Your score buys no algorithmic visibility above 40%. It buys conversion, curation and the shopper's benefit of the doubt.
• Valve is right that not every review needs a reply. Respond where there is something the reviewer does not already know. Skip opinions.

Start with the negatives describing a problem you have already fixed. Highest-yield category, the one where a response carries real new information, and the only one where Valve's warning about amplification works in your favour — because what gets amplified is the fix.

Then write the policy down before you need it. The developers who blew this up publicly were not careless. They were tired, at launch, with no rule to fall back on.