Captions on Product Videos: The Barrier Hiding in the Demo Reel
Product demos and sizing videos often ship with auto captions full of errors or none at all. WCAG caption requirements, what accuracy means, and a practical workflow.
The sizing video on a product page explains, in forty seconds, the one thing the size chart could not: how the jacket actually fits a person who lifts their arms. A deaf shopper gets the chart. If the video has captions at all, they are the platform's automatic ones, which render the fabric name as a place and the measurements as a weather report.
Captions are among the oldest requirements in WCAG and among the most skipped on commerce sites, because video arrives from marketing, gets embedded by a deadline, and nobody's checklist owns the text track. This post covers what the standard requires for prerecorded video, why auto captions alone do not meet it, and a workflow sized for a store that publishes a few videos a month.
What WCAG requires, level by level
Success Criterion 1.2.2, Captions (Prerecorded), sits at Level A: all prerecorded audio content in synchronized media needs captions, meaning dialogue plus the non-speech information that carries meaning, a door buzzer, a zipper's sound used as a quality cue, the presenter's laugh at a fit problem. Level AA adds 1.2.4 for live captions and 1.2.5, audio description, where the visual track shows something the narration never says, such as a hand pointing at the seam being discussed.
Most store video is a person talking over a product. For that format, accurate captions plus a transcript usually cover the practical ground, and audio description rarely binds because the speech already describes the visuals. The exception is the silent-format video, common in ads and social cuts, where all meaning is visual and on-screen text; that needs its text alternatives thought through, not assumed.
Why automatic captions are a draft, not a deliverable
Speech recognition is good at ordinary words and bad at exactly the words a product video exists to say: model names, fabric terms, measurements, sizes spoken as letters. An error rate that would be a nuisance in a meeting recording becomes misinformation in a buying decision. 'Size up if between sizes' misheard is a return waiting to happen, for a hearing customer who trusted the caption too.
The fix is a human pass, and it is shorter than teams expect. A forty second demo has perhaps 120 words of speech. Correcting a generated caption file for a video that length is a five minute task once someone is assigned to it, and the corrected file, WebVTT in most players, is a permanent asset that travels with the video.
A workflow a small team can keep
Make the caption file a required field in the same ticket that embeds the video: no VTT, no publish. Generate automatically, correct against the script or by listening once at normal speed, check speaker identification where two people talk, and spot-check that captions are actually enabled and rendered in the embedded player on desktop and mobile, because a correct file attached to a player configured to hide it helps nobody.
Related reading: ADA Risk in Age Gates and Date of Birth Fields, Testing Contrast in Disabled Buttons, Sold Out Options, and Placeholder States, ADA Risk in Embedded Size Charts, Fit Guides, and Care PDFs.
Frequently asked questions
Do YouTube or Vimeo auto captions satisfy WCAG?
Not on their own. WCAG 1.2.2 requires captions that are accurate and complete, including meaningful non-speech sound. Auto captions are a starting draft with regular errors on names, numbers, and technical terms. A human-corrected caption file is what meets the criterion.
Does a transcript replace captions on a video?
No. A transcript is valuable and worth publishing, but for synchronized media the requirement is captions presented with the video. A transcript alone does not satisfy 1.2.2, though it supports users who prefer reading and helps search indexing.
What about short silent product clips?
If a clip is purely visual with no audio, it is not a caption case; it needs a text alternative under the non-text content rules if it conveys information. Decorative loops marked as decorative are different from a silent clip demonstrating how a mechanism works.
Who should own captions in a small company?
Whoever publishes the video. Caption correction belongs in the publishing checklist, not in a later accessibility project. Five minutes at publish time is the whole cost; retrofitting a back catalogue is the expensive version everyone is trying to avoid.