5 of 5 AI vendors caught, 0 false positives, and a 29 field row per domain at $0.003

The drawing

The problem

A homepage that says AI tells you little. Navigation menus repeat the word on every page, so a raw count rewards the biggest menu. A chargeback fee written as a price per resolution reads like an AI charge. What a buyer wants is the level: does the company only mention AI, run a known AI vendor on its site, or put a price on AI in its plans.

What we built

AI Tooling Detector, an HTTP only actor at 256 MB. It fetches the homepage, the robots file, llms.txt, ai.txt, and the pricing page. It strips navigation, header, footer, and SVG text before scoring, and it counts a pricing phrase only when it carries AI context inside a 90 character window. It returns ai_maturity on four levels: none, declared, deployed, and commercialized.

Each row carries a confidence, the tools found, the evidence, the signals that fired, and the signals that didn't. A site that blocks the fetch comes back with blocked set to true and a confidence of 0.2 or lower, so a weak answer is visible rather than silent.

What it produced

• A gate before publish on 16 domains with known answers: 5 of 5 commercialized companies found, 0 false positives, and 9 of 9 negatives correct. Two domains blocked the fetch and were left out of the score.
• A second sample of 15 AI forward domains: 14 of 15 reached at least declared, 9 reached deployed or commercialized, and 5 of 5 negative controls read none.
• A live run on 7 domains finished in 5.7 seconds, and all 7 rows carried the same 29 fields.
• $0.003 per domain, $3.00 per 1,000, with an MCP server so an agent can call it.
• The README states the limit: a company that loads its own AI in the browser, with no vendor script in the served page, stops at declared.

Your scenario

Run AI Tooling Detector on Apify against 20 accounts you already know, and check the four levels against what you expect. For domains that come back blocked, a proxy API such as Scrape.do can fetch the page for your own review. Tracking AI adoption across a whole market? Send us the list and we'll tell you what share of it the four levels can read.