AI Visibility
When your next customer asks an AI about you instead of Googling, the answer is decided by how legible and consistent you are across the web. Kirke Labs gives you both halves: a free, open-source scanner to measure how readable you are to AI, and a ten-step field guide to fix it.
Measure: the AI-Legibility Scan
ai-legibility-scan is a free, open-source CLI that scores how readable your site is to AI agents and language models: not how it looks to a human, but whether a model can parse who you are, what you do, and describe you correctly. One command, a 0–100 score, and a concrete fix list.
What it looks at
- Crawlability without JS. Does meaningful content render in raw HTML, or only after JavaScript an LLM crawler won’t run?
- Structured data. schema.org / JSON-LD that states who you are and what you offer, machine-readably.
- Semantic structure. Real headings, landmarks, and question-shaped sections an answer engine can lift.
- Canonical description. One consistent paragraph about you, repeated across surfaces: the ranking signal.
- Metadata & canonicals. Titles, descriptions, canonical URLs, Open Graph: clean per-page identity.
- Agent-facing files. llms.txt, robots, and sitemap that tell models what to read and how to treat it.
We hold our own site to it: we target ≥80 on the homepage and ≥70 on our identity pages, and ship the fixes in the open.
The Prove-It Protocol
Knowing your score is half the job. The Prove-It Protocol is a ten-step field guide to engineering the answer an AI gives about you: the visibility, trust, and verdict you’d be proud of. The fix isn’t to be louder; it’s to be legible, structured and corroborated enough that a machine can describe you without hedging.
The ten steps
- 1See what the machine already says about you, across every model.
- 2Pass the crawler test: AI bots don’t run JavaScript, so your story has to live in the raw HTML.
- 3Write one canonical paragraph and repeat it everywhere; consistency is the ranking signal.
- 4Speak the machine’s native language with schema.org JSON-LD.
- 5Become an entity the model can resolve (Wikidata).
- 6Tie every profile and mention back together with bidirectional links.
- 7Win your name back wherever it collides with someone, or something, else.
- 8Earn independent corroboration from sources that aren’t you.
- 9Have one sharp, defensible opinion worth citing.
- 10Measure it monthly, defend it, and let it compound.
Open-source (run, read, contribute): github.com/KirkeLabs/ai-legibility-scan · npm