Search troubleshooting

Start with the Project, environment and named index. A working preview index does not prove that a Production website uses the same content or connection.

A new page does not appear

Find the page in Content. If it is absent, run the relevant crawler or import an updated complete snapshot. Rebuilding the index does not fetch source changes.

For crawled pages, check ownership verification, include/exclude paths, page and depth limits, extraction selectors and the recorded skipped-page reason. The crawler needs public HTML; it cannot read content that appears only after client-side JavaScript or sign-in.

If the document is present, check Activity for successful publication and confirm the website uses that index. An accepted import or queued build is not yet the published result.

Two sources supply different content for one page

Open the content preview, inspect both candidates and choose the source that should supply the page. Review the final additions and removals again. A full-replacement snapshot must include the index’s other sources if you want to retain them.

A crawl waits for review

The first baseline requires review. Later automatic publication still stops for incomplete results or removals beyond the configured threshold. Inspect the page outcomes, narrow the scope or correct extraction, and explicitly review the resulting changes. Do not treat a truncated crawl as a complete site snapshot.

A page exists but the query misses it

Search the exact title in Playground. Inspect the stored body text and searchable fields, then compare field priority, synonyms and typo settings in Relevance. If a draft preview is stale after a content update, build a new comparison before applying it.

The website shows an unavailable state

Compare the website’s installed configuration with the intended index’s Settings. Browser delivery requires website search turned on, the website in Allowed websites and the current publishable key. After Rotate key…, the website needs a rebuild with the new key. A server proxy requires its private query credential. Indexing credentials cannot replace query credentials.

Check Production configuration separately from local development. A previous installation verification timestamp does not establish current delivery health.

Suggestions are missing

Confirm that Suggestions has a published selection and the website reads it for the correct index when the query is empty. Removed documents no longer appear. Publishing suggestions does not update an old interface that never implemented them.

Analytics stays empty

Check collection is enabled, the period and index are correct, and the website has event recording installed. Both modes respect opt-out, GPC and DNT. Only consent-required mode also waits for analytics consent.

A client written before 12 October 2026 still returns results, but SiteOS rejects its events because their receipts carry no collection mode. Update it: remove the analyticsMode option and call prepareAnalytics() before an interaction starts. See Website connection.

Waiting for events in Installation means no event has arrived yet. It does not mean search is disconnected: a passed installation check keeps the website connected while analytics waits.

Perform a completed search and click one of its results on the website. Playground, CLI/MCP queries, installation checks and empty-query suggestions are excluded. If searches arrive but clicks do not, inspect the result-click integration and the query response information it uses.

A removed document returns

An import or crawl can restore a record still present in its source. Exclude it from the crawler or remove it from the next complete source snapshot as well as the current index.

See Content and indexing, Crawler, Website connection and Search analytics for the corresponding workflows.