The classic symptom: you can see the article in your dashboard, searching its exact title finds it, but asking the question in normal words doesn't.
That's a missing embedding.
Why it happens
Every article gets two kinds of index. Keyword indexing is immediate. Semantic indexing requires generating an embedding, which is a separate step that calls an AI model.
With an embedding, the article is found by meaning. Without one, only by literal word match — so a visitor phrasing the question differently from your title won't find it.
Check your coverage
Look at AI search coverage on your agent home. It's the share of your content with embeddings generated. Below 100% means some articles are only keyword-findable.
What to do
Re-save the article. Opening it, making a trivial edit, and saving re-triggers embedding generation. This fixes most single-article cases.
Wait a moment on bulk imports. A large import generates embeddings in batches. Coverage climbs over a few minutes rather than instantly.
Check for a service outage. Embedding generation depends on an external AI provider. If a large batch failed at once — coverage dropped sharply rather than gradually — the provider may have been unavailable. Re-saving after it recovers regenerates them.
Other reasons an article isn't found
It's unpublished. Unpublished content is excluded from retrieval entirely. This is the second most common cause.
Its collection is unpublished. The article stays individually searchable, but won't be browsable.
Something outranks it. A curated article on a near-identical topic can crowd out a standard or generated one. See Trust tiers.
The wording is genuinely too far off. If nothing in the article resembles how customers ask, add a sentence that does. Your knowledge gaps list shows their actual phrasing.
Confirming it's fixed
Search the phrase a customer would use, not the article title. The title will match either way, which is exactly what makes this problem easy to miss.