Hreflang via HTTP header for 12,000 non-HTML files

Hreflang via HTTP header for 12,000 non-HTML files

Hypothesis: localized PDFs ranking against each other could be disambiguated with HTTP-header hreflang, the only method available for non-HTML resources. Methodology: one documentation-heavy property, ~12,000 localized PDFs across 5 languages, all ranking and competing. PDFs can't carry head tags, so we set Link: HTTP headers. 11-week window.

Findings:
— Pre-change, English PDFs outranked localized ones for ~55% of localized-query/file pairs — Google had no signal these were equivalents.
— After deploying Link: <url>; rel="alternate"; hreflang="xx" headers (plus reciprocal entries), localized PDFs began surfacing for their own languages: correct-language file served rose to ~79%.
— Localized PDF downloads (our proxy for clicks, tracked server-side) rose 24% aggregate; the German set rose most at +38%.

Caveats: header-based hreflang is harder to verify — most crawlers and the GSC UI don't surface it cleanly, so our coverage confidence is lower than for HTML. We confirmed deployment via curl on a sample, not across all 12,000. The download metric is a proxy and includes non-search traffic we couldn't fully strip out.

Conclusion: HTTP-header hreflang is the only tool for non-HTML files and it does work — but its invisibility to standard tooling makes monitoring the hard part. Budget for custom verification before you commit at this scale.