Your Steam page, read the way a stranger skims it.
Paste a store URL for a free structure check. The full pack adds a blind vision + copy review of your actual capsules, screenshots and text on a fixed, published rubric - with every observation tied to the image it came from.
How pages are scored
70 points come from a multimodal model reviewing the actual images and copy on anchored 1-5 bands; 30 points are deterministic structure. Reviews, price, release date and popularity are never scored. If the review cannot be verified, the page is marked unreviewed instead of guessed. The free check runs the structural 30% only.
| Weight | Area | How | What is checked |
|---|---|---|---|
| 20 | Capsule & header | Vision review (full pack) | The real images, judged at the size players see them: is the title readable at 231x87, does the art say genre and tone, contrast, consistent branding. |
| 15 | First 5 screenshots | Vision review (full pack) | Does each read as a thumbnail, do they show actual play, do they show different situations. |
| 15 | Short description | Copy review (full pack) | Read for meaning, not length: what the player does, a specific hook, genre and tone. |
| 10 | About this game | Copy review (full pack) | Leads with the strongest information, skimmable, concrete features instead of promises. |
| 10 | Copy-visual coherence | Vision + copy (full pack) | Do the words promise what the pictures show; would a browser get the game in five seconds. |
| 30 | Structure | Deterministic (free) | Tags, screenshot count, trailer, resolution, About structure, languages, OS; price and date only once released. |
What the $29 pilot pack contains
Everything is grounded in your page and in real, live-verified Steam games. Nothing is padded or made up.
Visual + copy review
Asset-by-asset observations on your real capsules and first screenshots, rated on a published rubric, with confidence levels and at least three concrete edits.
Rewritten short + About
Rewritten from facts already on your page. Unsourced numbers and names are flagged for you to check.
Suggested tags
Tags that verified peers share and you lack, plus generic tags worth moving down.
HTML + PDF report
Observed issues separated from optional experiments, the evidence that the model saw your pixels, and measured model cost.
About 5 real peers
Same-genre games from live Steam search, re-fetched and checked, with the tag overlap shown.
3 capsule concepts (not final art)
Concept comps, not finished custom art. Three directions - brand preservation (small 462x174), editorial (main 1232x706), detail-forward crop (header 920x430) - from your own logo, key art and screenshots.
Pricing
One-time payment through Stripe's hosted checkout. You get a private link to your pack; generation usually finishes within minutes.
$29 per game
- Blind vision + copy review on a fixed rubric
- HTML + PDF report with 3+ concrete edits
- Rewritten short description and About
- Suggested tags, about 5 verified peers
- 3 capsule concepts (concepts, not finished art)
$15 / month
Weekly re-audit with change tracking.
Coming soon. Not built yet: there is no subscription and no recurring billing.
Straight answers
Will a higher score sell more copies?
We do not claim that. The score is an expert-style judgement of how clearly the page presents the game. It has not been validated against wishlists or sales, and model ratings vary by a few points between runs.
Do you actually look at my art?
In the paid pack, yes: a multimodal model receives your small capsule at list size, header, main capsule and first five screenshots. Each review includes a control image with a random code and must describe every image; if that check fails, the pack is not delivered as reviewed - the job retries and otherwise reports a failure. The free check does not look at images.
Where does the data come from?
Steam's public appdetails endpoint, the public store page (for user tags and the review summary) and the appreviews summary when it responds. Fetches are allowlisted, time-limited and cached.
Is AI used?
Not for the free check. The paid pack uses Anthropic's Claude API (no tools, structured output) for the vision + copy review and for a copy rewrite grounded in your own facts. Token usage and an estimated model cost are recorded in every pack.