Subtitle Accessibility
compliance and better reach, same work

Burned-in captions cannot be turned off, resized, restyled or read by assistive technology, and they are frequently styled at contrast ratios that fail WCAG. They are excellent for sound-off feed viewing and insufficient for compliance.

Ship both: styled burned-in captions for retention, and an accurate SRT or VTT track for accessibility, caption toggling and platform indexing. The marginal cost is a few minutes per video.

Ninety-nine percent or better, including punctuation, speaker changes and non-speech sounds that carry meaning. Auto-captions at 92 to 97 percent do not meet this, and the errors cluster on names, jargon and numbers — the words that matter most.

Cap reading speed at 160 to 180 words per minute. Where speech is faster, condense rather than truncate, keeping the meaning and the speaker's voice intact.

Where meaning is carried visually — a chart, an on-screen figure, a silent demonstration — captions alone leave blind and low-vision users without the content. Either describe it in the narration, which benefits everyone, or provide an audio description track.

Narration-first design is usually cheaper and better than retrofitting description: writing scripts so that nothing important exists only on screen removes most of the problem at source.

Subtitles assume the viewer can hear and renders speech only. Captions assume the viewer cannot hear and include speaker identification and non-speech audio that carries meaning — a door closing, a laugh, music that signals a tonal shift. The distinction matters because accessibility standards ask for captions, not subtitles.

WCAG 2.1 Level AA requires captions for prerecorded audio content (1.2.2). Organisations subject to accessibility legislation — public sector bodies, many healthcare and education providers, and increasingly private companies operating in the EU under the European Accessibility Act — are held to that standard for published video, including social content.

Burned-in captions satisfy the readability need but not the machine-readable one. Supplying an SRT or VTT alongside the burned-in version covers both, and has the side benefit of being indexable.

Reading speed: aim for under about 17 characters per second. Above 20, comprehension drops sharply for most viewers and collapses for anyone reading in a second language.

Line length: 32 characters per line, maximum two lines on screen. Three-line captions on a vertical video cover the subject and are usually a sign the timing needs rebuilding rather than the font shrinking.

Contrast: caption text needs a 4.5:1 contrast ratio against what is behind it. Over video that means a semi-opaque plate or a consistent outline, not white text and hope.

Minimum duration: hold any caption for at least 1.2 seconds even if the phrase is short. Flashed single words are a common short-form styling choice and a genuine accessibility failure.

Avoid conveying information through colour alone — a red figure meaning 'bad' needs a word or symbol as well.

Keep flashing content below three flashes per second. Fast strobe transitions are a photosensitivity risk and are also disproportionately down-ranked by viewers reporting content as uncomfortable.

Provide a transcript for long-form. It serves screen-reader users, people who prefer reading, and search engines simultaneously, and it costs almost nothing once accurate captions exist.