WCAG 3.1.2
Language of Parts
SeriousLevel AAWCAG 2.0
0.2%
of the 609 sites we scanned fail this
Pooled across five targeted cohorts, so treat it as a convenience sample rather than a rate for the web.
What it means and how to fix it
What 3.1.2 covers
- No engine can detect this: Language changes inside a PDF documentA PDF validator can only tell whether any language is declared at all, which is 3.1.1. Whether a passage is in another language, and marked as such, needs someone who can read it. We previously counted missing document languages here too, which double-counted them.
- No engine can detect this: Foreign-language passages on a web page with no language attributeFinding these means identifying the language of every passage on the page and comparing it with what was declared. Nothing in the markup flags an unmarked language change.
- No engine can detect this: Proper nouns and technical terms that need no markingThe criterion exempts them, which is another judgement rather than a rule.
Screen readers may mispronounce content in sections with invalid language attributes.
The fix. Use a valid language code for any section in a different language so it's read with the right pronunciation.
For what the criterion formally requires, read the W3C Understanding document.
How often it fails
US local government1/158
0.6%
Share of scanned sites in each cohort. These are five targeted studies, not a random sample of the web, so read the contrast between cohorts rather than any one figure as a web-wide rate.
What this does not measureLimits of these figures, and where the corpus came from
- These are five targeted cohorts, not a random sample of the web. Per-cohort rates are the figure to quote; the pooled rate is a convenience sample.
- Denominators count sites that produced a result. Sites that were blocked or errored are excluded rather than counted as passing.
- Occurrence counts are floors: the engine records at most 25 occurrences of a rule per page.
- The EU and Indian cohorts retained only the primary success criterion per finding, so a rule mapping to several criteria contributes to one of them in those two cohorts and to all of them in the other three.
| Cohort | Scanned | Attempted |
|---|---|---|
| Shopify DTC stores | 75 | 75 |
| US local government | 158 | 221 |
| US healthcare providers | 208 | 218 |
| EU e-commerce | 107 | 170 |
| Indian financial services | 61 | 72 |
Corpus generated 2026-09-18.
Check your own site
A free scan reports WCAG 3.1.2 alongside everything else, with the exact elements that fail.
Run a free scan