Marketplace Listing Score¶
Free tool: vell.ai/listing-score · Sample report: vell.ai/listing-score/sample
No signup, no email capture. Paste a listing or give it an AWS Marketplace URL and it returns a scored report immediately.
What it measures¶
The score answers one question: can a buyer — or a buyer's AI agent — verify anything you claim?
It is not a completeness check. A listing with every field filled in, correct length and clean prose can still score in the thirties, because length is not evidence. The score is built from a 0.70 evidence / 0.30 structure blend per dimension, where evidence means things a reader can check:
| Signal | Counts as evidence | Does not |
|---|---|---|
| Outcomes | "cuts p99 latency 40%" | "dramatically faster" |
| Security | "SOC 2 Type II", "ISO 27001" | "enterprise-grade security" |
| Integrations | named systems — Datadog, Okta, ServiceNow | "integrates with your stack" |
| Pricing | a stated number, or a real free tier | "contact us for pricing" |
| Support | "4-business-hour response" | "world-class support" |
Superlatives are counted, not rewarded. A listing that leads with revolutionary, best-in-class and industry-leading, with no numbers anywhere, scores in the high thirties — and the report says exactly which phrases cost it.
Dimensions¶
Clarity · Buyer fit · Differentiation · SEO · Trust signals · Conversion · Rich media
Each returns a score, a short reason, and up to three concrete fixes. The overall score is a weighted mean over the dimensions that could actually be measured — see below.
Two ways to run it¶
By URL. Reads the listing straight from the AWS Marketplace page. Some fields are simply not
present on a public page, so those dimensions are marked partial rather than scored down.
By pasted copy. Paste your title, short and long description, highlights, categories, support text and keywords. This scores everything, including fields the public page never exposes. If a URL fails to read — AWS blocks automated reads on some pages — pasting gives the same report.
What it will not do¶
- It never penalises copy it could not read. An unreadable field is reported as unmeasured,
never as a zero. Look for the
partialflag on a dimension. - It never tells you to add adjectives. Every recommendation asks for something checkable.
- It does not measure backlinks, domain authority, or AI-search visibility. Those are real, and they are not part of this score.
- It does not publish a percentile it cannot support. Category percentiles appear only when the comparison corpus was scored by the same algorithm. The SEO percentile is always withheld: search keywords are backend-only and never appear on a public listing page, so a crawled corpus is keyword-blind while a pasting seller is not. Ranking one against the other would flatter the seller by construction, so it is left out rather than guessed at.
Reading your report¶
Look at the last two numbers together — checkable claims and structure score. A high structure score with zero checkable claims is the common failure: the listing is well-formed and says nothing a buyer can act on. That gap, not the headline number, is the finding.
Then take the three fixes in order. They are ranked by how much each would move the score.
Why agents make this urgent¶
Buyers increasingly arrive through an agent that reads the listing before a human does. An agent cannot be persuaded by tone — it extracts claims and looks for support. Copy that reads well to a person and carries nothing verifiable is invisible to that reader. The score approximates what survives that extraction.
FAQ¶
Is it free? Yes, and there is no email gate. The report renders in the page.
Do you store my copy? Runs are recorded so the scorer can be evaluated and improved. Do not paste anything you would not put on a public listing.
My URL returned nothing. AWS Marketplace is a client-rendered app and blocks automated reads on some pages. Paste the copy instead — identical scoring.
The score dropped since last time. The scorer was rebuilt to measure evidence rather than field length. Listings that scored in the nineties on completeness alone now score far lower, and that is the correction working, not a regression.
Why does a dimension have no percentile? Either the category corpus is not current, or the dimension is one whose corpus is not comparable to your input — see above.