Human, agent or bot

human: a person in an ordinary browser. browser_agent: an AI agent driving a browser, known because the form was submitted through WebMCP, the browser is automated, or the agent named itself in its user agent. verified_agent: an AI agent that proved who it is with a cryptographic signature (Web Bot Auth), so the agent's domain, such as chatgpt.com, is shown with it. bot: a spam trap fired, or the sender is a script rather than a browser.

Only verified_agent is proof; the others are a careful reading of signals a sender controls, and the dashboard shows which signals decided each one. A submission from your own server (posting with curl or a backend library) is tagged bot, which is accurate: no browser sent it.

The inbox shows a tag on every submission not sent by a person, and the AI agents tab lists only those. The tag is in webhook payloads, the Submissions API (sender and agent, filter with ?sender=verified_agent) and the MCP tools.

{
  "event": "submission.created",
  "submission": {
    "id": "sub_...",
    "data": { "name": "Ada", "email": "ada@example.com" },
    "sender": "verified_agent",
    "agent": "chatgpt.com"
  }
}

In a routing rule, use the field _meta.sender (the dashboard suggests it and its values) or _meta.agent. Send agent bookings to a separate channel, or skip your CRM for anything that is not a person.

{
  "name": "AI agents to the bookings channel",
  "conditions": [{ "field": "_meta.sender", "op": "in", "value": ["browser_agent", "verified_agent"] }],
  "integrationIds": ["int_slack"]
}

An agent that supports Web Bot Auth signs each request with a key it publishes at its own domain, and says which domain in the Signature-Agent header. WebForms fetches that domain's public keys, checks the signature (HTTP Message Signatures, RFC 9421, Ed25519) and that it was made for this request and is current. A missing or failed signature never blocks a submission; it is simply not verified.