Opens in a new tabSkip to content
Agent LighthouseAgent Lighthouse

    Searches the text of every published page. The evidence sources themselves are not in this index — search all of them on the trusted sources page.

    GitHub ↗
    Browse checks and page contents
    content-extraction/language-attribute

    Language attribute

    What it checks

    AI agents use the lang attribute to select the correct language model and tokenizer when processing your content. Without it, agents may misinterpret content language, leading to poor translations or incorrect answers in multilingual AI systems.

    Why it matters

    User agents read the lang attribute on <html> to programmatically determine the page’s natural language. Screen readers select pronunciation rules from it, visual browsers select scripts and characters, and media players select captions. A page without it cannot have its language programmatically determined, and fails WCAG 2.2 SC 3.1.1.

    Evidence

    • WCAG 2.2 SC 3.1.1 (Level A), W3C Recommendation of 12 December 2024: “The natural language of each web page can be programmatically determined.” — w3.org/…/WCAG22 (verified 2026-08-21)
    • Documented consumer behavior: “Screen readers can load the correct pronunciation rules”, “Visual browsers can display characters and scripts correctly”, “Media players can show captions correctly” — w3.org/…/language-of-page.html (verified 2026-08-21)
    • W3C i18n guidance names <html lang> as the required declaration mechanism: “you should always declare the language of the text in a page using a language attribute on the html tag” — w3.org/…/qa-html-language-declarations (verified 2026-08-21)
    • Same source rules out the fallback the audit does not currently consult: “You should never use a meta element with the http-equiv attribute set to Content-Language to indicate the language of a page” — w3.org/…/qa-html-language-declarations (verified 2026-08-21)

    Limits

    Google Search explicitly ignores this attribute for language determination: “We don’t use any code-level language information such as lang attributes, or the URL… Google uses the visible content of your page to determine its language” — developers.google.com/…/managing-multi-regional-sit… (verified 2026-08-21). No LLM vendor documents the audit’s stated mechanism (selecting a “language model and tokenizer” from lang); OpenAI’s crawler documentation mentions no HTML metadata — developers.openai.com/…/bots (verified 2026-08-21). The grade rests on the accessibility/i18n consumer path, not on an AI-specific one; the audit’s description and guidance.impact should be restated accordingly.

    Note also that the standard is satisfied by a valid BCP 47 tag, which the current presence-only check does not verify — an invalid or wrong lang still fails SC 3.1.1 while scoring 1.0 here.

    How it scores

    Ratified W3C Recommendation (WCAG 2.2, SC 3.1.1 Language of Page, Level A) plus WHATWG HTML, with named consumer classes whose behavior W3C documents explicitly.

    Sources