What a Low-Scoring Steam Page Looks Like: 25 Real Cases

Twenty-five of the 373 Steam pages in our sample scored below 50. The lowest scored 29. This post is what those pages have in common.

The headline finding is not the one we expected. Low-scoring pages are not mainly bad at art: their median capsule subscore is 45 against 72 for everyone else, a 27-point gap. They are worst at writing. Median description subscore in the bottom 25 is 35, against 65 for the other 348 pages — a 30-point gap, the widest of any element, on the one part of a store page that costs nothing but an afternoon to fix.

Every page below is anonymized. These are small teams, most of them unreleased, and naming them would be pointless cruelty that teaches nobody anything. We describe the genre, the year, and the mistake. The other teardowns in this series name their subjects because those subjects are large commercial games that can absorb the attention.

Two honesty notes. These are our product’s rubric scores, not a Valve metric and not a prediction that any of these games will fail. And the 373 pages are ones people chose to run through the free analyzer, which is a convenience sample: it probably over-represents developers who already suspected something was wrong.

The severity number is the one to sit with. Our analyzer marks an issue critical when it thinks the page is failing at something structural rather than being merely suboptimal. Across the whole sample there are 283 critical issues, and the bottom 25 pages account for 52 of them: 7% of the pages hold 18% of the critical problems. Bad pages are not evenly bad, they are concentrated.

The screenshot gap is also mostly a volume gap rather than a quality gap. Nearly half the bottom 25 ship six or fewer images, against roughly a fifth of everyone else. Our sample data shows the median screenshot subscore jumping ten points between the 6-to-8 group and the 9-to-12 group and then flattening, which we work through in the Palworld teardown. Getting from five images to nine is the single cheapest score movement available to a page in this group.

Thirty points is a big gap and the individual reports explain it consistently. Four patterns come up again and again.

The description that repeats itself. A coming-soon precision platformer scoring 43 had an About section of 1,143 characters that restated the short description almost word for word, then described its levels as “fun zones” without saying what the player does in them. A co-op roguelite scoring 37 had a short description reading “Team up with friends in a roguelite adventure!” and an About section opening “Team up with friends in this action-packed roguelite.” Its entire About section was 460 characters.

The description made of adjectives. “Action-packed,” “unique challenges,” “immersive experience.” Our analyzer’s note on one of these was that the page uses vague language “without explaining what makes this game special,” which is the polite version of: the copy would work equally well for four thousand other games.

The description that hides the genre. A horror crafting game scoring 42 opened with “Run your own toy workshop in this horror-themed crafting simulation game.” The report’s read: that ordering makes it sound like a cosy crafting game with horror bolted on, so horror fans do not recognise it as theirs and crafting fans are put off. When your game is a hybrid, the first noun decides which audience self-selects.

The description in the wrong language. The lowest-scoring page in the whole sample, at 29, had its entire store page written in a single non-English language with no English version, for a game aimed at a global audience. Three of its issues were marked critical. This is rarer than the others but it is fatal, and the fix is mechanical.

If you only do one thing after reading this, rewrite your first two sentences. The short description guide has the patterns, and the description writing guide covers ordering for the long section.

Seven of the 25 pages carry a capsule issue with essentially the same title: the capsule lacks visual impact and genre clarity. Three examples, all anonymized:

• —A coming-soon 2D action game (43 overall, capsule 22). Black title text on a flat cornflower-blue field, with two character sprites so dark and small they are nearly invisible at the corners. The small capsule was a compressed version of the same near-empty composition. This page also shipped no trailer at all, one of only 21 in the whole 373-page sample.
• —A horror crafting game (42 overall, capsule 25). Both the header and small capsule were the game’s title on a flat wood texture. Nothing else. The report’s line: a player browsing Steam has no idea this is a horror game.
• —A puzzle game with a mascot (44 overall, capsule 45). Cute animal logo on a pastoral background, with no puzzle grid, no tiles, no gameplay element anywhere in the frame. It reads as a children’s platformer.

The common failure is not ugliness. All three are competently made images. The failure is that the capsule answers “what is this called” and never answers “what is this.” A shopper scanning a browse row is doing genre triage, not evaluating art. If your capsule could sit on a book cover without changing, it is not doing store page work.

The counter-technique, from the Balatro teardown: fuse the genre object into the wordmark. Balatro replaces a letter with an actual playing card, so the title and the genre signal occupy the same pixels. Test what survives at 120x45 with the capsule validator.

This is the most common single structural error in the group and the easiest to fix, so it deserves its own heading.

Steam gives you two text fields that appear within a few hundred pixels of each other. They have different jobs. The short description stops the scroll. The About section converts the person who stopped. A page that puts the same sentences in both has spent its second field buying nothing.

One page in this group, a released simulation game scoring 37, was flagged for a detailed description “duplicated verbatim” with no additional differentiation. Another, at 44, repeated its short description at the top of the About section and then included an unattributed pull-quote introduced with “Let me quote...” and no link or source.

The structure that works is boring and reliable: short description states what the game is; About section opens with the specific promise the short description implied, then proves it with concrete detail, then lists features, then handles the objection your genre always raises. The Hades page follows exactly that shape, and its only description criticism was that it left the objection handling for last.

Three cases where the images actively argued against the copy:

• —A couch-multiplayer sports game (40 overall). The short description promised players fighting over a ball in local multiplayer chaos. All five screenshots showed one or two tiny characters standing far apart in mostly empty levels. Our analyzer marked it critical: the images do not contain the product the copy sells.
• —A business simulation (37 overall). Twenty-one screenshots, almost none of which showed any business management, staff, finances or operations. A player who came for the management sim had no visual proof it exists. Note that this page had more than twice the median screenshot count and still failed, which is the clearest possible evidence that gallery size is not the variable that matters.
• —A co-op roguelite (37 overall). Every screenshot showed a solo character in a sparse environment, on a page selling co-op. The first screenshot was an upgrade menu, in a different language from the page copy.

The test is mechanical and takes five minutes. Write down the three promises your short description makes. Then open your gallery and point at the image that proves each one. If you cannot point at one, that promise is unsupported, and unsupported promises are where refund requests come from. The screenshot checker covers order and first-frame legibility; this particular check has to be done by a human who knows what the copy claims.

Tag failures in this group split into two opposite modes.

Too few. A released simulation game was using only 6 of 20 available tag slots. A puzzle game was using 9 of 20. Unused tag slots are not neutral, they are discovery you have declined. Steam has no way to show your game to an audience you never described.

Contradictory. A casual simulation game scoring 41 carried Horror, Investigation, Mystery, Thriller, Choices Matter and Story Rich alongside Casual, Idler, Walking Simulator and Education. Our analyzer’s note called the list “self-contradictory” and it is hard to disagree: a shopper reading that cannot form a single idea of what the game is, and neither can the recommendation system. Another page carried FPS, Third-Person Shooter and Third Person simultaneously, which is a claim about camera perspective that cannot be true twice.

A third mode is subtler and worth checking on your own page today: a released game still carrying the Early Access tag. One page in this group released in 2022 and still listed Early Access four years later. Buyers read that as an unfinished game and skip it.

The rule that covers all three: your tag list should be twenty different ways of saying one true thing about your game. The tag optimizer tests variants against what people actually search in your genre.

The last cluster is not really a design problem, it is a maintenance problem, and it is over-represented in low scores because low-scoring pages tend to be pages nobody has re-read since launch.

Examples from the group: an Early Access tag on a game released four years earlier. Third-party watermarks left on screenshots so the images look unofficial. A description ending mid-feature-list with no closing line, on a page that has a playable demo the copy never mentions. A short description of 72 characters, when the sample median is 232, so the page forfeits most of the space Steam gives it.

None of these need a designer, a writer, or money. They need somebody to open the page as a stranger would and read it top to bottom once.

If your own page is somewhere in this range, the sequence that moves the most for the least effort:

1. 01Rewrite the first sentence of your short description so it contains a genre noun and a verb. This is the widest gap in the data and the cheapest thing on this list.
2. 02Make the About section stop repeating the short description. Start it where the short description ended.
3. 03Get to nine screenshots, and make sure at least one proves each promise your copy makes.
4. 04Fill all twenty tag slots, remove any tag that contradicts another, and check that Early Access reflects reality.
5. 05Put a genre object in the capsule. Not a better illustration, a recognisable thing. If the capsule works as a book cover, it is not working as a capsule.
6. 06Read your own page in a private window, as somebody who has never heard of your game, and delete whatever you would have skipped.

For the shape of the whole distribution rather than its bottom, see our audit of 373 Steam store pages. For the top of the range, the Hades teardown covers the highest-scoring page in the sample at 87, and the Stardew Valley teardown covers a page that has held an 80 for a decade.

You can run your own page against the same rubric on the homepage, or see where other pages landed on the leaderboard.