Captions, Subtitles, and What WCAG Actually Requires for Video in 2026
WCAG 2.2 caption requirements decoded — the difference between captions and subtitles, Level A/AA/AA conformance thresholds, and the quality criteria beyond mere presence.
“Does it have captions?” is not the compliance question — it’s the floor. WCAG distinguishes captions from subtitles, prescribes which conformance level requires which, and judges caption quality on criteria most auto-generated tracks fail silently.
Captions vs. Subtitles — The Distinction That Matters
- Subtitles assume the viewer can hear: they translate spoken dialogue only.
- Captions assume the viewer cannot hear: they include dialogue plus non-speech audio —
[door closes],[music swells], speaker identification, and meaningful sound effects.
WCAG requires captions, not subtitles. A subtitle track that renders only dialogue fails conformance for prerecorded content regardless of how accurate the dialogue is.
The Conformance Ladder
| Level | Video Requirement | Who It Applies To |
|---|---|---|
| A | Captions on all prerecorded video with audio | Everyone, no exceptions |
| AA | + Captions on live video; audio description for prerecorded video | The de-facto legal standard in most jurisdictions |
| AAA | + Sign language interpretation, extended audio description | Rare; public-sector mandates |
For live streams, AA-level live captioning means real-time stenography or a high-accuracy ASR pipeline with human correction — not just turning on auto-captions and hoping.
Quality Criteria Beyond Presence
Auto-generated captions meet the letter of the law but often fail the spirit — and users. The measurable quality bar:
- Accuracy ≥ 99% for prerecorded content (industry convention, FCC-adjacent). Auto-captions typically land 85–95% — acceptable for drafts, not for published content.
- Synchronization within ~100ms of the corresponding speech.
- Speaker identification when the speaker isn’t visually obvious.
- Line length ≤ 42 characters and presentation rate ≤ 160–180 wpm so captions are readable in real time.
“A caption file that exists but renders dialogue without speaker labels or sound cues is the accessibility equivalent of a ramp that ends in stairs — technically present, functionally useless.”
Quick Audit Checklist
-
<track kind="captions">(notsubtitles) on every<video> - Captions survive fullscreen and quality switches
- Non-speech audio is described
- Captions are styled for contrast and user-configurable size
- Live events have a real-time captioning plan, not an afterthought
Our caption audit checklist and the auto-vs-human accuracy data behind the 99% bar are published in the WCAG caption requirements reference.