Methodology · Article 50(1) checker
How the checker works
Article 50(1) asks that people are told they're talking to an AI, at the latest when they first interact with it. The checker opens your page and your chat the way a visitor would and reads what the chat says before the first message. It's an automated check that supports your compliance work, not legal advice.
What a scan does
- 01
Checks the address
Only public pages are scanned. Every request the scan makes, including the browser's, goes through a guard that refuses private and internal addresses.
- 02
Loads the page in a real browser
The page runs in Chromium the way a visitor's browser shows it, with its scripts and chat widgets. Images, video and fonts are skipped to keep the scan fast.
- 03
Doesn't flag itself as automated
Some chat widgets, such as Intercom and Crisp, don't load for browsers that announce they're automated. So the scan's browser uses Chromium's normal user agent and doesn't set the automation flag, and you see the chat your visitors see.
- 04
Behaves like a visitor arriving
It waits for the page to settle, moves the mouse and scrolls once, because some sites only load their chat after that.
- 05
Opens the chat
Intercom, HubSpot and Crisp chats are opened through the widget's own script, without clicking. Otherwise it clicks up to three buttons that look like chat launchers.
- 06
Reads what the chat shows before the first message
The chat's header, greeting, notices and input placeholder: what a user sees before typing anything.
What the scan never does
- Accept or reject cookies
- Sign in or create an account
- Send a message
- Fill in or submit forms
- Buy, download or book a demo
If the browser can't load the page, the scan checks only the HTML your server sent and says so in the report.
How the verdict is decided
The scan looks for a plain notice such as “You're chatting with an AI assistant” in the chat, on the page and in the chat widget's settings. Labels like “Powered by AI” or a bot named “AI Agent” count as vague: they hint at AI without telling users directly.
- Disclosure found
- The page tells users they're talking to an AI.
- Vague label only
- The chat is labelled as AI, but it never says plainly that users are talking to one.
- Check by hand
- There's a chat, but the scan couldn't confirm what it tells users. Open it and check.
- No disclosure
- The page has an AI chat, but nothing tells users it's an AI.
- No AI chat detected
- The scan found no AI chat or assistant on this page.
Support chats like Intercom or Crisp can be answered by people or by an AI agent. When one of them opens without saying who answers, the result is “Check by hand”: if an AI replies, it has to say so.
Chat widgets the scan recognises
Chatbase, Voiceflow, Botpress, Dify, Chatling, ElevenLabs Conversational AI, Intercom, Drift, Crisp, Tidio, Zendesk, HubSpot Chat, LiveChat, Freshchat, tawk.to. Chats built in-house are found by their input box and launcher button.
What the scan can't see
- The scan doesn't sign in or send messages, so chats behind a login, or text shown only after the first message, aren't seen.
- Chats that open only after longer scrolling, on timers or on exit intent may not be found.
- A found phrase doesn't prove it's shown at or before the first interaction, as Art. 50(1) requires.
- This is an automated check, not legal advice.