Declared language matches the content
<html lang> matches the language the visible content is actually written in, per a conservative deterministic language profile. The classic failure is a template default nobody edited: a German shop declaring lang="en". A mismatch WARNs, never fails - the detector is probabilistic and stays silent when unsure.
Adoption of what this check tests is still climbing, so failing it is an opportunity rather than a defect. Remediation lists rank these after established fixes, and the weight moves only when measurement earns it.
No rate published. Too few labelled cases to publish a rate: 1 flagged case (30 needed to publish), 1 defective case (30 needed to publish). The counts are the honest answer here. The matrix behind coherence.language_declared_matches_content is 2 labelled cases — reported here rather than turned into a percentage that would read as more certain than the evidence is.
Corpus v1, measured under rubric r2026.11.0. Full method and every other check at /rubric/accuracy.
Set `<html lang>` to the language the page is actually written in. Agents pick tokenizers, target markets, and which variant to quote from this attribute; a wrong declaration routes the wrong readers to the page.
Free scan, 30 seconds, no signup — this check and the other 35, with the exact evidence for each verdict.
Scan my site →