You updated the help article, but Fin still gives customers the old answer. Don't start by rewriting the guidance. First find out whether Fin saw the new article at all.
Check the answer and the source separately
- Repeat the question in a test conversation for the affected brand, language, and audience. Record the wrong statement and the source used, if available.
- Open the cited article. If the live page is wrong, fix it at the source. If the live page is right but the answer is old, check the version Fin can use.
- Look for another article, snippet, or source that still contains the old rule. Decide which page should win for that audience.
- Run the same test after the change, then test a different customer role and a case that should go to a person.
If the source is an external website
In Fin AI Agent → Train → Content, open the website source and check its sync history. Intercom's website sync guide explains how to resync a source or retry a failed page. Check that the exact article is included, reachable without signing in, and enabled for the intended audience. A successful sync does not prove that the important part of the article was captured.
Don't assume a publish immediately changes Fin's answers. Scheduled website sync can be delayed or paused. For a time-sensitive update, manually resync and verify the actual answer. Content that must stay private belongs in an appropriately restricted source, not a public copy created for the crawler.
If Fin is still wrong
| What you see | Where to look next |
|---|---|
| Fin cites an old article | Look for duplicate or legacy sources still enabled for that audience. |
| Fin cites the right article but misses a rule | Put the exception next to the relevant instruction, then refresh and retest. |
| Fin uses different sources for different customers | Check brand, language, audience targeting, and what each customer can open. |
| Fin gives no useful source | Check whether the page synced, is enabled, and has enough text to answer the question. |
Save recurring cases in Fin Evals, including the expected source and whether Fin should hand off. Use the chatbot QA sheet for the human review. If the source problem started during a site move, check the Intercom help center migration guide for the cutover sequence.