Keyword search fails when a reader describes the problem in their own words and your page uses different ones; AI search answers by meaning, and the lasting fix is to write the page the failed searches were asking for.
Why does keyword search return nothing?#
Keyword search matches the words in the query against the words on the page. A reader who types "reset my password" gets nothing when the page is called "Account recovery".
Readers rarely type your headings. They type the symptom or the question:
- "why does my webhook keep failing"
- "can I use this without an API key"
- "what is the difference between the two plans"
When the words don't match the page that answers them, a keyword index comes back empty and the reader leaves or opens a ticket.
How does AI search work?#
AI search compares meaning instead of words. Most implementations follow the same three steps:
- Your pages are split into passages, and each passage is turned into an embedding: a vector that stands for what it means.
- The reader's question is embedded the same way.
- The closest passages are retrieved, and a language model writes the answer from them, citing the pages they came from.
Because retrieval works on passages, a section that names its own subject retrieves better than one that leans on the heading above it.
What does Docsbook do with search?#
Every Docsbook site gets a search box, and the misses become work for the agent:
- Search box — full-text search over every page, with titles weighted above body text.
- Ask AI — with the AI chat on (Pro), the search box offers to ask the question instead. The chat answers with the pages it used, cited; with Semantic Search on in Settings ▸ Agent, it finds those pages by meaning.
- Every miss is recorded — a search that returns nothing is logged once the reader stops typing; the agent reads those misses, and the Write the pages readers wanted trigger wakes on them.
- The agent writes the missing page — the Write the pages readers wanted trigger wakes on a search with no results; File the search gaps files the misses as a ranked issue every day.
A better search engine finds the page you have. Only a new page answers the question you don't cover yet, which is why Docsbook routes every miss to the agent (Find wins fast).
What should you measure?#
Three numbers show whether readers find their answers:
- Searches with no results — each one names a page or a section that is missing.
- Chat questions nobody answered — the Answer what the chat could not trigger works from these.
- Tickets the docs already answer — tag them for a month; if the count stays high, the answer exists but readers can't find it.
In Docsbook the first two are collected for you. You can also ask your agent in one sentence (Get discovered):
Read last month's failed searches and unanswered chat questions, and write the three pages that would answer most of them.FAQ#
Is AI search the same as an AI chat?
No. AI search finds the passages that mean what the reader asked; an AI chat also writes the answer from them, and Docsbook's cites the pages it used.
Does AI search replace good headings?
No. Retrieval works on passages, and a passage whose heading and first sentence name the subject is easier to find by keyword and by meaning.
Do I have to build a vector database?
Not on Docsbook: turn on Semantic Search in Settings ▸ Agent and the index is built from your repository and kept current, billed as AI usage. Building your own means owning the embeddings, the index updates and the answer model.