AI agents contacting your support
AI agents contacting your support
Spot AI assistants that contact you on a customer’s behalf, and choose how TR-AI treats them.
People increasingly ask AI assistants to contact companies for them. TR-AI recognises them from several signals and records each one, so you can see how often it happens.
- The browser or client: AI assistant browsers (for example ChatGPT agent, Claude, Perplexity) and automation tools.
- Agent-protocol headers sent by AI agents.
- The message itself: “I am an AI assistant acting on behalf of…”, machine-formatted messages (JSON, long forms).
- Typing speed in the chat widget that no person can reach.
| Policy | What happens |
|---|---|
| Answer normally (default) | The agent gets the same answers as anyone else. |
| Answer, but no account actions | Questions are answered, but procedures treat the requester as unverified: no order details, refunds or account changes. |
| Hand off to a person | TR-AI does not answer; the ticket goes to your team with an acknowledgement. |
Tips & tricks
- “Answer, but no account actions” is the safest default for most businesses: general help still works, account data stays protected.
- Check the detections list weekly at first — if you see real customers flagged, tell us the signal shown so we can tune it.
- Detection works best in the chat widget, where the browser and typing speed are visible; on e-mail only the message text is available.