Captions and Transcripts for Accessibility: A WCAG Guide
What WCAG asks of captions, transcripts and descriptive transcripts, at which level, and how to turn an automatic transcript into captions that pass.
Short answer
WCAG asks for captions on prerecorded video with audio at Level A and on live video at Level AA, and for a transcript of prerecorded audio-only content at Level A. W3C says automatic captions do not meet accessibility requirements unless they are confirmed fully accurate, so treat an AI transcript as a draft: correct it, add important sounds and speakers, then export captions.
Captions and transcripts are often treated as the same thing. The Web Content Accessibility Guidelines (WCAG), which many accessibility laws and procurement rules refer to, ask for them in different places and at different levels. This guide sets out what the guidelines ask, in W3C's own words, and how to get from a raw AI transcript to files that meet the bar.
What WCAG asks for
| Content | What it needs | Level |
|---|---|---|
| Prerecorded video with audio | Captions | A |
| Prerecorded video with audio | Audio description, or a descriptive transcript with all the visual and spoken information | A |
| Prerecorded video with audio | Audio description | AA |
| Prerecorded video with audio | A descriptive transcript | AAA |
| Live video with audio | Captions | AA |
| Prerecorded audio only (a podcast, a recorded talk) | A transcript | A |
| Live audio only | A transcript | AAA |
| Prerecorded video only, no speech | A descriptive transcript or audio description | A |
From W3C's Transcripts and Captions/Subtitles guidance (both updated 17 September 2024) and the WCAG 2.2 explanation of success criterion 1.2.3.
Three kinds of text, three jobs
- Captions are "a text version of the speech and non-speech audio information needed to understand the content", shown in time with the video for people who are Deaf or hard of hearing
- A transcript carries the same information as a document to read on its own, without the timing
- A descriptive transcript goes further: descriptive transcripts "also include visual information needed to understand the content", such as on-screen text and actions, which is what people who are both Deaf and blind need
Why an AI transcript is a draft, not the finished captions
W3C is clear: "Automatically-generated captions do not meet user needs or accessibility requirements" unless they are confirmed fully accurate, and "missing just one word such as 'not' can make the captions contradict the actual audio content." SubGrab's AI transcription is 95% accurate on clear English speech, which makes a strong first draft, but up to one word in twenty can still be wrong. Speech recognition also writes only the words: it does not add [applause] or [door slams], and it cannot see the screen.
From AI transcript to accessible captions
1. Transcribe the video on SubGrab. If the platform has captions they are fetched; otherwise AI writes them from the audio
2. Mark the speakers. For interviews and panels, switch on speaker labels before you confirm (1 credit per started hour on top, refunded if only one voice is heard), then rename them so the captions say who is talking
3. Export a caption file: WebVTT for a player on your own site, SRT for editors and platforms that ask for it
4. Correct it in a caption editor while watching the video, and check names, numbers, acronyms and technical terms closely. SubGrab does not edit caption text itself, so this step happens in the file
5. Add the sounds that matter, such as [laughter], [music] or [phone rings], wherever they change what a viewer understands
6. For a transcript, download Word, PDF or text with Include timestamps off, and add the visual information if you need a descriptive transcript
Where SubGrab helps and where you take over
| Step | SubGrab | You |
|---|---|---|
| Speech to timed text | Captions fetched, or AI transcription | Check it against the video |
| Who is speaking | Speaker labels, renamed once for every file | Confirm the switches between voices |
| Non-speech sounds | Not written | Add the ones that matter |
| Visual information | Not described | Add it for a descriptive transcript |
| Files | WebVTT, SRT, Word, PDF, text | Publish captions with your player |
Audio-only content needs a transcript first
For a podcast or a recorded talk with no video, the Level A requirement is a transcript, not captions. Upload the recording or paste the episode's link, correct the text, and publish it next to the audio. A Word or PDF download with timestamps off reads well as a page.
Frequently asked
Do automatic captions meet WCAG on their own?
Not unless they are confirmed to be fully accurate. W3C's caption guidance says automatically generated captions do not meet accessibility requirements otherwise, and that they usually need significant editing. They are a good starting point: correct names, numbers and terms, add the sounds that matter, and mark who is speaking.
What is the difference between a transcript and a descriptive transcript?
A basic transcript is the text of the speech and the non-speech audio needed to understand the content. A descriptive transcript adds the visual information too, such as on-screen text and actions, for people who can neither hear nor see the video. An AI transcript gives you the speech; the rest you add.
Which caption format should I upload, SRT or WebVTT?
Use WebVTT for a video on your own site: it is the format HTML video players read, and W3C calls it the most common caption format on the web. Use SRT for editors and platforms that ask for it. SubGrab exports both from the same transcript, with speaker names if you set them.