How far does 'obvious from context' go?

The chatbot disclosure duty has a built-in exemption: no notice is needed where it would already be obvious, to a "reasonably well-informed, observant and circumspect natural person," that they're dealing with AI, taking the circumstances and context of use into account. The exemption is real. It's also the most over-claimed sentence in Article 50, so here is how the draft guidelines actually read it.

First, the benchmark person is drawn from your actual audience, including children, the elderly, and low-literacy users if they're a foreseeable part of it. A notice-free setup that works for tech-savvy adults fails if your real user base is broader. Obviousness is judged for the people you actually serve, not an idealised user.

Second, the examples cut against the common hope. The draft guidelines treat developer-only code assistants and video-game NPCs as obvious: contexts where the artificiality is the premise. What they treat as not obvious by default is the embedded website helpdesk or support chatbot, the single most common SME configuration. The era of assuming "everyone knows it's a bot" is precisely what the duty was written against: support chats where a human and a bot are genuinely interchangeable from the user's side.

Third, "my bot has a robot avatar and a cute name" is weaker than it feels. Names and avatars are branding; the test is whether every reasonable member of your audience would immediately recognise the AI nature. Any honest doubt resolves toward disclosing. The disclosure costs a sentence, while a documented wrong bet on the exemption costs an argument with a regulator.

If you do rely on the exemption (say, for a developer tool), write the assessment down: audience, context, why it's unmistakable. A dated obviousness assessment converts a judgment call into evidence, and it's exactly what you'd want on file the day anyone asks why there's no notice.