Captions and Accessibility: What Compliance Requires

Most video captions are written as though the viewer can hear. They contain the dialogue and nothing else, which is enough for someone watching in a quiet office with the sound off, and not enough for the person the accessibility standards were written for. If a door slams off screen and everyone turns to look, a dialogue-only caption leaves the viewer watching an unexplained reaction.

The published standards describe an outcome rather than a file: the viewer who cannot hear the audio should get the same information as the viewer who can. That framing changes what belongs in a caption and how accurate it has to be. This guide covers which criteria apply, what a caption has to contain to satisfy them, why automatic output usually does not, and how to check a video before it goes out.

Captions Are Not Subtitles

The two words are used interchangeably in everyday production talk, and the standards depend on the distinction.

A subtitle exists because of a language gap. The viewer can hear the tone, the music, the door, the second speaker interrupting — they just do not follow the words. Dialogue alone is therefore sufficient.

A caption exists because of an audio gap. Everything the soundtrack carries has to arrive some other way: who is speaking, that someone is speaking at all when they are off screen, and any sound that changes what the picture means.

This is why "we have subtitles" is not an answer to "is it captioned". The files may look identical, and the second one is missing information the first never needed to carry. It is also why the translation question and the accessibility question are separate decisions with separate answers, a distinction that runs through choosing between dubbing and subtitles for each market.

The Criteria That Apply

The Web Content Accessibility Guidelines are the reference that most public-sector procurement rules, and most organisational policies, point at. Several other frameworks — including the European EN 301 549 standard and the US federal accessibility requirements — adopt WCAG's criteria rather than writing their own, which is why learning the criteria is more useful than learning any single jurisdiction's rules.

Two things are worth noticing in that table. Captions and audio description are separate criteria, so a fully captioned video can still fail on the visual side. And the criteria that touch styling and resizing are not about captions specifically — they apply to text generally, and captions inherit them.

What Goes Into the Caption

Beyond the dialogue, three categories of information are expected.

Who is speaking. When the speaker is off screen, when several people are present, or when the picture does not make it obvious, the caption identifies them. A name in brackets or in capitals before the line is the common convention; consistency within the video matters more than which form you pick.

Sound that carries meaning. A phone ringing, a knock, laughter, an approaching vehicle, a sudden silence. The test is whether a hearing viewer would take information from it. Ambient sound that carries nothing does not need describing, and describing everything makes the meaningful entries harder to spot.

Music, where it matters. Song lyrics if they are audible and relevant, and an indication of mood or genre where the score is doing narrative work. A caption reading that ominous music is playing conveys something the picture may not.

There is also a requirement people underestimate: the captions must be accurate. A caption that says the opposite of what was said, or that renders a name or a figure incorrectly, does not give the viewer the audio content. Accuracy sits inside the criterion rather than beside it, which is why an unreviewed automatic pass is a compliance question and not only a quality one — the failure modes and the repair order are set out in a systematic way to fix bad auto-captions.

A Transcript Is Not a Caption

Publishing the transcript underneath the video is a good thing to do and it does not satisfy the captioning criterion. Captions are synchronised: they tell the viewer what is being said now, alongside the picture that goes with it. A transcript is the same words with the synchronisation removed, which makes it useful for reading, searching and referencing, and useless for following a video as it plays.

The exception is narrow and worth knowing. Where the media is audio-only, a text alternative genuinely can stand in for it, because there is no picture to stay in step with. For anything with both picture and sound, the transcript is a complement rather than a substitute — and it is a cheap one, since you already produced the text in order to caption.

Live Captions Are a Different Job

The live criterion looks like the prerecorded one with a tighter deadline, and it is not. There is no correction pass, no segmentation decision made at leisure, and no opportunity to check a name before it goes out. Whatever is produced is what the viewer gets.

Planning for it means deciding in advance what the failure mode will be. Supplying names, terms and running order to whoever or whatever is captioning removes the most common errors before they happen, which is the live equivalent of a vocabulary list. And a recorded stream should be re-captioned properly before it is republished, because the live pass was never intended to be the permanent version.

Timing, Styling and Reading

Captions can contain every required piece of information and still fail the person using them, and the standards touch this from two directions.

Contrast applies to the text as rendered. A white caption over a bright sky does not meet a contrast requirement in any meaningful reading of it, and since the background changes constantly the only reliable answer is a style that carries its own separation. The four techniques for that are in caption styling that stays legible on every background.

Resizing is the criterion that quietly rules on delivery. If the viewer must be able to enlarge text, captions rendered permanently into the picture cannot comply with that on their own, because the pixels are fixed. This is the strongest technical argument for shipping a closed caption track even when the picture already shows the words, and it sits alongside the other consequences in burned-in captions versus soft subtitles.

Reading pace matters too. A caption that a viewer cannot finish before it disappears has not conveyed its content, and the kinetic short-form style is particularly exposed here because the text cannot be re-read at all. Where captions serve accessibility rather than attention, a stable line is the defensible choice, for the reasons set out in when word-level and line-level captions each win.

A Pre-Publish Check

Next Steps

Pick your most-watched published video and run the sound-off pass on it. Most teams find the same two gaps: off-screen speakers who are never identified, and a sound effect that the picture reacts to without explanation. Both are quick to fix and both would have been missed by reading the caption file.

Then decide, once, what your house convention is for speaker labels and sound descriptions, and write it down where whoever produces the next video will see it. Consistency across a library is worth more than perfection on one video. The rest of the caption decisions — formats, timing, styling and delivery — sit under captions and subtitles.