ai-legibility-scan
Open-source · MIT · A gift to the Algorand ecosystem

Is your site legible
to an AI agent?

Buyers don't Google you anymore — they ask an AI that reads the web for them. One command tells you what that AI can actually see, and hands you the structured data to fix it.

$ npx @kirkelabs/ai-legibility-scan https://your-site.com

No install. No account. No data leaves your machine. Node ≥ 20.

Why it exists

Monitoring tools tell you you're invisible. This one fixes it.

Most "AI visibility" products are $300-a-month dashboards that watch and report. ai-legibility-scan is a free CLI that fetches your page the way GPTBot and ClaudeBot do — one plain request, no JavaScript — scores eight dimensions, and writes the schema.org JSON-LD and draft llms.txt you need to close the gaps. It's also the only scanner we know of that checks on-chain legibility: whether your Algorand proof is server-rendered and explorer-linked, or trapped invisibly inside a JS dashboard.

How it scores

Eight weighted dimensions

01 Server-rendered content w10

Is the meaningful text in raw HTML, or hidden behind JavaScript an AI never runs?

02 Schema.org data w10

Valid JSON-LD; Organization/Person/Product; the sameAs identity graph.

03 Entity identity w10

Links to Wikidata, Crunchbase, LinkedIn, GitHub — corroboration that you exist.

04 Answer-shaped w8

Opening answer, question-headings, tables — the shapes engines lift.

05 Fact density w7

Statistics and explicit sourcing per 200 words (GEO research).

06 Crawler access w7

Does robots.txt block the AI bots and zero out everything else?

07 llms.txt w4

Present and well-formed — a low-cost, unproven hedge, weighted honestly.

08 On-chain legibility w5

Algorand proof server-rendered and explorer-linked, not JS-only.

Full rubric, thresholds and the evidence behind every weight: METHODOLOGY.md. These are documented judgement calls — disagree in an issue.

What you get back

A score, and the files to fix it

Honest limits

What it does not claim

This measures heuristic indicators of AI legibility. A high score makes a page easier for a model to read, trust and cite — it is not a guarantee of citation, and no tool can promise that. It does not render JavaScript (by design — if your content needs JS to appear, that is the finding) and it does not query ChatGPT to check if you're mentioned. It's a static-legibility auditor, and it says so.