reynforce

How do I prove to a customer what our AI agent actually said?

Last updated

Show them the entry written when it was said: the sentence, the time, the agent, and the line it relied on. What you can offer today is your own record, kept so that nobody can edit it — not yet something the customer can verify without taking your word for it.

The question usually arrives as a dispute: your agent told us this was included. The worst position is to answer from memory, or from a transcript somebody exported, edited and pasted into a reply.

What makes a record worth showing

What to show, and what not to

Show the entry for that promise: the sentence, when, which agent, and the line behind it. Do not hand over the rest of the record, and do not show internal approval limits that were never part of the conversation. The customer is owed what was said to them, not how your company decides.

What we cannot offer yet

A record a customer can check independently — signed, so that it proves itself without anybody trusting us — is planned and not built. The record is written in a form that makes it possible later: one format, stable field names, a version on every entry, and nothing that only makes sense inside our own systems. Until then, here is our record is a statement from you, backed by how the record is kept rather than by a signature.

When the agent said something the record does not show

If an agent sent something without asking first, there is no entry, and the honest answer is that you have no record of it. That is worth knowing before the dispute rather than during it.

/audit-what-ai-agents-told-customers.md covers reading the record yourself. /trust.md covers where it is held and for how long.