Search is the one feature static sites are supposed to give up. Pagefind removes that excuse, so I wired it into this site and measured what it cost.
What I Did
Pagefind runs after the build, against the generated HTML. It reads the DOM, extracts content from elements marked with data-pagefind-body, and writes a static index plus a small JavaScript bundle into the output directory.
Setup
- Build the site as usual.
- Run the indexer against the output directory.
- Copy the index into the published output.
- Load the index only when the user opens search.
The last step matters. The bundle is small, but there is no reason to ship it on first paint.
What Worked
- No service, no key, no quota. The index is files on the same CDN.
- Relevance is respectable. Titles and headings are weighted properly out of the box.
- Section and tag filtering works without extra configuration.
What Did Not
- Indexing must happen after every build. Forgetting it silently produces an empty search.
- Excerpts include template chrome unless the body is scoped carefully. Mark the content container, not the page.
- The fallback question. If indexing fails, search returns nothing rather than failing loudly.
Note to self
Because an empty index looks like “no results”, I kept a JSON fallback index built by Hugo itself. If the static index is missing, the client falls back to the Hugo-built one.Verdict
Good enough to ship, with one caveat: treat index generation as part of the build, not as an optional step. I now run it in the same command as the build so it cannot be forgotten.
