Accuracy
- Give it facts: prices, links, policies.
- Tell it to hand over anything not in the FAQ.
- Test real questions before switching it on.
Privacy
ReplyInbox uses Anthropic's Claude through the API; conversations aren't used to train the model. Only the recent messages of each chat are stored, so the AI can answer in context and you can read them.
Brand safety
- Hand-offs for refunds, complaints and collaborations.
- No promises it can't keep: it only states what's in your FAQ.
- You can take over any chat at any time.
Frequently asked questions
Should I tell people they're talking to AI?
Being open is good practice and required in some places; add a line like "Replies may be assisted by AI" to your About or welcome message.