Skip to content

Search returns unexpected results

Symptoms

A search returns files that seem unrelated, or the "obviously right" file is not near the top.

What this usually means

Search matches the words in file names, paths, metadata, and extracted text, and ranks results by relevance — matches in more important fields (like the file name) count for more. Surprises usually come from words occurring where you did not expect them:

  • The word is in the path. A file inside a folder named Berlin 2019 matches "berlin" even if the file itself has nothing to do with it.
  • The word is deep in extracted text. A long document mentioning your term once is still a match, just ranked lower.
  • Common words match everywhere. Searching "photo" or "2020" matches half a catalogue.
  • The scan is still running. Ranking shifts as more files get their text extracted.
  • You expected meaning, not words. "Sunset at the beach" as a scene is a visual search query; regular search will instead find files containing those words.

Results from disconnected drives also appear by design; among equally relevant results, currently-usable files are listed first.

Safe steps to try

  1. Add a distinctive second word to narrow the match.
  2. Look at why a result matched — check its name, path, and details; the word is almost always genuinely there.
  3. For photos, switch to visual search when you are describing a picture rather than quoting text.
  4. Let indexing finish before judging ranking on a fresh repository.

What not to delete or change

  • Do not delete and re-add repositories to "fix" ranking.
  • Do not delete the catalogue database.

Information to collect for support

If ranking seems genuinely wrong (the exact-name match buried under weak matches, for example):

  • The query, the expected file, and where it ranked.
  • A screenshot of the first results.
  • Whether indexing had finished.