AI Twin says it can't find something it should know

When your AI Twin says "I don't know" about something you know it has been fed, one of three things is almost always happening. This article covers each cause with a specific fix.

Cause 1 — The source wasn't indexed

Open Knowledge → Sources and find the page or document.

• Status Indexed — the source is in the graph; skip to cause 2.
• Status Failed — see Website crawl failed or is incomplete.
• Missing entirely — re-crawl, re-upload, or add a sitemap.xml.

Cause 2 — The phrasing doesn't match

Reality Twin retrieves by semantic similarity, not keyword. If your source uses very different vocabulary from the visitor's question, retrieval can miss.

Fix

• Add a structured Q&A that uses the visitor's vocabulary — this bridges the gap without changing the source.
• Or edit the source to include the visitor's phrasing in a synonyms sentence.

Cause 3 — A boundary rule is silencing it

Open Twin → Personality → Boundaries. A rule may be intercepting the topic and forcing a refusal.

Fix

• Narrow the rule with more specific example phrasings.
• Or delete the rule if it was too broad.

Diagnostic tip: use the test panel with citations

The Twin → Test panel shows retrieval scores for every candidate chunk. If the right chunk has a low score, you have a phrasing problem (cause 2). If it isn't in the list at all, you have an indexing problem (cause 1).

Frequently asked questions

Why did an answer regress after re-crawling?

Usually a source page was updated with different wording. Add a Q&A to lock the answer in.

Does capitalisation matter?

No — retrieval is case-insensitive.

Can I force the twin to use a specific source?

Yes — mark the source High priority under Knowledge → Sources → source detail.