Telling customers about AI

When to use it

Reach for this when someone says any of these — they are the phrases the skill itself triggers on:

• “do we tell customers it's a bot”
• “AI disclosure requirements”
• “should the bot say it's not human”
• “customer asked if they were talking to a person”

How it works

The method, in the order the skill runs it. The full procedure — tables, worked examples and the edge cases — is in the skill itself.

1. Design the interaction disclosure

   Up front, not on request. A bot that identifies itself only when asked has left every customer who did not think to ask uninformed. Disclose at the start of the interaction.

2. AI-assisted decisions

   Where AI contributed to a decision affecting the person — a claim, a refusal, a redress amount, an account action, or an agent's performance assessment: Say that it was involved, at the point the decision is communicated, Offer human review where that applies, and make it real, Record the provenance.

3. Auditing what actually happens

   Design is the easy half. Test the operation: Position in the conversation, Handover disclosure, both directions, Human-route completion, not availability.

Related skills

• PII redaction auditCheck an export or dataset isn’t leaking personal data
• Subject access requestsAnswer a data request without exposing other people
• Call recording rulesSet the rules for storing, sharing and deleting recordings
• Where support data goesMap where support data really goes, and who receives it
• Collecting less dataFind personal data you collect and never use, and stop
• Required disclosuresCheck customers got the required wording, at the right moment