The fallback message appears when one of the chat's safety checks decided the answer wasn't good enough to show. Check, in order:
- No results at all? If the fallback appears on questions your content covers, that's the similarity threshold, not the accuracy checks — work through that checklist first.
- Citation Integrity is on and the model isn't citing. The answer is replaced whenever its citations don't resolve to the retrieved evidence. If this happens constantly, confirm the chat's answering LLM profile is OpenAI or Anthropic — other providers can't enforce citations (the switch is inactive for them, and Diagnostics records the mismatch).
- Strict Answer Verification rejected the claims. The answer contained statements the accuracy model couldn't support from the retrieved chunks — often a sign the retrieved evidence is too thin. Raise Top K slightly, or improve retrieval with Hybrid Search so better evidence reaches the model.
- The accuracy profile is failing. Verification uses the Accuracy LLM Profile, or the chat's main profile when that field is empty; a failing or rate-limited profile falls back defensively. Check the profile and your provider dashboard.
When a check fails, the visitor sees "I could not verify a supported answer from the supplied sources." The chat's No Results / Fallback Message setting is for the no-results case — make it useful by inviting the visitor to rephrase. Details on both checks: Citations & Answer Verification.
