BotNotice
Log In
How it worksPlansFAQsBlogWhich AI Laws Apply to Me?AI Law TrackerLog In

The AI disclosure index

A visitor can't tell from a chat window whether AI is answering unless the website says so. We looked for a public count of how often US websites do, and did not find one. So this index counts: it opens the chat on a list of websites, reads the first screen a visitor sees, and counts how many say that AI is answering.

The first edition is being collected

No numbers are published yet. An edition is published only after at least 100 chats have been opened and read, and after the results have been reviewed by a person. Until then, you can check any single website with the free scan below.

Check one website now

The method

  1. The list. Each edition says where its list of websites came from. The list is fixed before scanning starts, and no website is added or dropped because of what the scan found.
  2. Passive scans only. The page is read. If a chat tool is found, a real browser opens the chat the way a visitor would and reads its first screen. Nothing is typed. No button inside the chat is pressed. No message is sent.
  3. What counts as “said AI”. The page or the chat's first screen uses words that tell a visitor an AI, a bot or an automated assistant is answering, or that it is not a person.
  4. What counts as “AI answering, nothing said”. The chat opened, its first screen was read, the tool is one where a bot always answers or the first screen's own wording shows a bot, and nothing said so in the terms above.
  5. What is left out of the percentages. Websites that turned the scan away, could not be reached, had no chat, or whose chat could not be opened. They are counted and shown, never guessed.
  6. Totals only. No website is named. The point is the count, not a list of who to blame.

Limits

  • A notice can appear later than the first screen: after a visitor types, or picks an option. This index reads the first screen only, so it undercounts those.
  • A chat with people behind it needs no AI notice. Chats where the scan could not tell are reported as their own group.
  • BotNotice sells AI notices. That is why the method is published in full, the totals are frozen with the method version that produced them (currently 2026-11), and nothing here says any website broke a law. Whether a law applies to a website depends on facts a scan cannot see.

Writing about this? Email john@botnotice.app for the full table behind an edition. Related: the law tracker and the open dataset of AI disclosure laws.