Site search
What is indexed, how readers use it, and how to correct it.
Site search
Section titled “Site search”Every published site has search. There is nothing to switch on.
What is indexed
Section titled “What is indexed”- page titles;
- every heading, so a result can point at a section rather than a page;
- body text;
- tags;
- API operations, on generated reference pages.
Variables are resolved before indexing, so a reader searching for your API hostname finds the page that mentions it.
Drafts are not indexed. Neither are noindex pages.
What a reader gets
Section titled “What a reader gets”- Cmd+K or Ctrl+K opens it from anywhere.
- Results appear as they type.
- The matching words are highlighted in context.
- A result links to the exact heading anchor, not the top of the page.
- Typing tolerance covers a misspelling.
- The whole dialog is keyboard navigable.
Private sites
Section titled “Private sites”Search on a password or workspace site is authorised on the server. A reader
receives only results they are allowed to open.
Correcting it
Section titled “Correcting it”Search does not need tuning to work, and it usually needs some to work well. Two places to look:
Your no-result searches. Every query that found nothing is recorded. That list is the exact set of words your documentation is missing, in your readers’ own vocabulary. See Reader analytics.
search.yaml. Synonyms, aliases, pinned results and weights. See
Search tuning.
Tip
Work from the no-result list rather than from imagination. A synonym group you guessed at costs nothing and fixes nothing; a group taken from twenty real queries fixes twenty readers.
Where a query becomes a page
Section titled “Where a query becomes a page”A no-result query that keeps coming back is not a search problem; it is a missing page. Repeated failed searches are clustered and become items in the inbox, with the count attached, so the decision to write the page is made against a number.