Steam Price vs Store Page Quality: Data From 154 Games

Across the 154 released paid games in our analyzed set, store page scores rise with list price up to about $15 and then stop. Games listed under $5 carry a median score of 59.5 against 70.0 for everything $5 and above (p less than 0.0001), and every objective measure moves with it: 40% less description text, two and a half fewer screenshots, and a critical issue on 77% of pages against 45%. Above $15 the relationship is gone: the rank correlation between price and score is +0.026 with p = 0.85.

Raising your price will not improve your capsule. This is an association in a convenience sample, and the obvious lurking variable is money. A team that can charge $20 usually had more of it, and more time, and spent some of both on the store page. Nothing below is a causal claim, and if you take one action from this post it should be about your page, not your price tag.

Our companion post on pricing psychology argues from buyer behavior that a sub-$5 tag reads as a warning label. This is the other half of that: some of the stigma is earned, and here is what the pages actually look like.

The price our analyzer records is whatever Steam was charging at the moment of analysis, discount included. That number is unusable for this question, and it took a re-export to notice.

51 of the 165 released pages in the set (30.9%) were on sale when we scored them. Discounts ran from 10% to 100%. Twelve of them cross the $5 line purely because of the sale: a $44.99 strategy game at 90% off, a $19.99 narrative game at 85% off, Half-Life 2 at $9.99 list showing $1.99. Every one of those would have been filed as a budget game by a naive read of the price column.

The difference is not cosmetic. Using the discounted price, sub-$5 games score 64.0 against 71.0, a 7.0-point gap at p = 0.0010. Using the list price the same comparison gives 59.5 against 70.0, a 10.5-point gap at p less than 0.0001. The sale price both shrinks the effect and moves games into the wrong bucket. Everything below uses list price from Steam’s price_overview.initial field.

Scope: 373 pages analyzed, of which 165 were released. 11 of those are free-to-play and 154 carry a US list price. Coming Soon pages are excluded because exactly 1 of the 208 had a price at all, which is its own finding about pre-launch pages.

Read this table down the columns, not across the rows. The four subscores are separately calibrated and two of them carry known rubric defects (see the limits section), so “description 46 is lower than screenshots 69” is not a claim this instrument supports. “Description 46 in the cheap band is lower than 71 in the $20 band” is, because every page was scored by the same distorted ruler.

A permutation omnibus across the five bands gives p = 0.0016 on overall score and p = 0.0012 on description. Spearman rank correlation between list price and overall score is +0.476 (p less than 0.0001, n=154).

Note the top row. $40-plus games score below the $20 to $39.99 band, 71 against 76. That difference is not significant on 8 games (p = 0.18), so treat it as “the curve flattens”, not “premium pricing hurts”. But it is not a straight line, and anyone drawing one through this data is over-reading it.

Splitting at $5 gives 30 games against 124. The gap is not confined to the parts of the score that rely on the model’s taste. It shows up in things you can count without any rubric at all.

The capsule row is reported as a mean deliberately. The rubric parks 59 of these 154 pages on a capsule subscore of exactly 72, so both medians are 72 and the median test returns a meaningless p = 1.0000. The mean gap is real and significant. Any time you see a tied median on a rubric score, suspect a mass point.

Trailers behave the same way: median 1 against 2 (p = 0.55, tied ranks), mean 1.3 against 2.6 (p = 0.0030).

The character count is the cleanest number here because no judgment enters it. A sub-$5 page carries roughly 60% of the long-description text of a $5-plus page and about 44% of a $20-plus page. The description writing guide covers what to put in the gap; the short description guide handles the 300 characters at the top.

This is half the finding and it is the half that gets dropped when this kind of result gets repeated. The gradient is real at the bottom and absent at the top.

Each row drops the cheap games and re-runs. By $15 there is nothing left. Among the 58 games listed at $15 or more, price and store page score are unrelated.

The band-to-band tests agree. $5 to $9.99 against $10 to $19.99 is a 4.5-point difference at p = 0.069, which does not clear the bar. $10 to $19.99 against $20-plus is 5.0 points at p = 0.018, which does. $20 to $39.99 against $40-plus reverses sign on 8 games at p = 0.18.

So the honest shape is a floor effect, not a slope: something distinguishes the sub-$5 band from everything else, the $5 to $20 range is a mild and mostly non-significant climb, and above $15 the signal is gone. If you are pricing at $18 and wondering whether $24 buys you a better page, this data says no.

“Your expensive games are famous games.” 30 of the 154 carry a Metacritic score and none of them is in the cheap band. Drop all 30 and the correlation is +0.436 on the remaining 124, with the sub-$5 gap at -9.0 (p = 0.0005).

“You are just measuring a handful of great pages at the top.” Drop every page scoring 80 or better and the correlation is +0.443, with the sub-$5 gap at -10.5 (p less than 0.0001).

“The sale prices contaminated it.” Restrict to the 103 games carrying no discount at analysis time and the gap widens to -14.0 (56.0 against 70.0, p less than 0.0001).

“Old cheap games drag it down.” Here it gets weaker and you should know that. Restricted to the 98 games released in 2024 or later, the median gap falls to -5.0 at p = 0.12, which does not clear significance. The mean gap over the same subset is -8.0 at p = 0.0004. We report both. The pattern is present but softer in recent releases, on 20 cheap games.

By genre, with Holm correction across the six testable:

Simulation fails. The 11 cheap simulation pages in this set score 66.0 against 70.5, a difference that a permutation test cannot distinguish from noise. Strategy passes on 5 cheap games, which is thin enough that we would not defend it. Both are in the table because leaving them out would make the result look tidier than it is.

One more null worth publishing: the 11 free-to-play pages score a median 70.0 against 69.0 for all paid games (p = 0.90). Free is not the bottom of this gradient. Whatever is happening below $5 is not simply “cheaper means worse”, or free would be worst of all.

Three reasons the causal reading fails.

Resourcing explains all of it comfortably. A team charging $24.99 has usually shipped before, has usually raised or saved money, and has usually paid somebody to do key art. The same budget that justifies the price paid for the trailer, the screenshots and the copy. Price is a proxy for the budget, and the budget built the page.

The free-to-play null breaks the monotone story. If price caused page quality, $0 would sit below $4.99. It sits at 70, level with the $10 to $19.99 band. Free-to-play games are usually made by teams with a revenue model and a marketing plan, which is the resourcing story again.

The gradient dies exactly where you would expect if this were about resourcing rather than price. The sub-$5 band in this sample looks like small first-time work by every observable measure: none of its 30 games carries a Metacritic score, 8 of the 30 ship exactly Steam’s 5-screenshot minimum, and 23 of the 30 have a single trailer. Above $15 you are looking at teams that all cleared a similar production bar, and price stops sorting them.

There is a real second mechanism worth naming, and it runs the other way: the price is set after someone decides how much game there is. A developer who knows the game is 40 minutes long prices it at $2.99 and writes a short description because there is not much to describe. On that reading the thin page is honest, not lazy. This data cannot separate the two, and both should make you nervous about “just charge more”.

Treat the finding as diagnostic. It tells you where you sit in a distribution, not what to charge.

If your list price is under $5, you are in the band where pages are measurably thinnest, which means the page is probably the cheapest thing you can still improve. The measurable gaps in the table close without touching the game: the long description is roughly 1,900 characters short of the norm, the page is two or three screenshots short, and there is a 77% chance our rubric would flag something critical on it.

None of that requires a price change. A $2.99 game with 12 screenshots, a 5,000-character description and a capsule that says what the game is would sit near the top of its band, and nothing in this dataset says it must be an outlier.

If you are already above $15, this post has nothing for you on price. Go work on the capsule, which is where the strongest single defect in our corpus lives.

For the price number itself, our pricing strategy guide covers comparables, regional tiers and launch discounts, and pricing psychology covers anchoring and what a price signals to a shopper before they read a word.

The cheap bucket is 30 games. Everything about the sub-$5 band rests on 30 observations from a convenience sample of pages people chose to run through a free tool. That is enough for the effect sizes reported and not enough to slice further.

Two known rubric defects. These analyses were produced before we fixed two scoring bugs. Description subscores are systematically depressed by roughly 11 points because a formatting check read the wrong markup, and tag subscores are inflated by roughly 11 to 12 points because Steam feature flags were counted as tags. Both distortions apply equally to every page, so band-to-band comparison inside one category is valid, and cross-category comparison is not. The tags row in particular should be read as directional only.

The rubric sees only the first 10 screenshots. 121 of the 373 pages in the corpus (32.4%) carry more than 10. A page listing 12 screenshots was judged on 10 of them, so nothing here says anything about screenshots 11 and up.

No commercial outcome is joined to any of this. No sales, wishlists, revenue or conversion data. A 59.5 page may well outsell a 76 page. Our scores are one AI rubric’s opinion of a store page, not a Valve metric.

Regional and currency scope. All prices are US list prices. Regional pricing is a separate question covered in the regional pricing guide.

The full audit of all 373 pages has the corpus-wide distributions, the leaderboard shows public scores for pages already scored, and the analyzer will tell you which band your own page currently reads like.