Deaf and Hard of Hearing Users
Every piece of information you deliver by sound needs an equivalent in text. Auto-captions are a starting point, not a deliverable.
Auto-generated captions get the names, the numbers, and the jargon wrong — which is most of what matters.
A training video with no captions is a training video that was not published. An auto-captioned one is worse in a specific way: it looks captioned, so nobody checks, and it renders the CEO's name as three wrong words, the dollar figure as a different figure, and the technical term as nonsense. For users whose first language is ASL, English text is a second language, which is an argument for plain language and for sign interpretation on critical content, not just captions.
Scale: About 15% of American adults report some trouble hearing, and hearing loss rises sharply with age. (Source: NIDCD. Survey-based disability figures undercount — treat this as a floor.)
What they're using
- Captions and transcripts, which are the accommodation rather than a device
- Hearing aids and cochlear implants, which help with speech but not with poor audio quality
- Sign language interpretation, where the user's first language may be ASL rather than English
- Live CART captioning for meetings and webinars
5 barriers we find most often, and the criteria behind them.
Video published without captions
Prerecorded video with audio needs accurate synchronized captions, including speaker identification and meaningful non-speech sound.
Success criteria: 1.2.2 Captions (Prerecorded) A
Audio-only content with no transcript
Podcasts, recorded calls, audio announcements, and voice messages all need a text alternative.
Success criteria: 1.2.1 Audio-only and Video-only (Prerecorded) A
Live events with no real-time captioning
Webinars, council meetings, and streamed classes need live captions. This is a Level AA requirement, and it is one of the most commonly ignored.
Success criteria: 1.2.4 Captions (Live) AA
Information delivered only as sound
An alert tone with no visual counterpart, a video where the narration explains something the visuals do not show, an interface that confirms success with a chime.
Success criteria: 1.2.3 Audio Description or Media Alternative (Prerecorded) A · 1.2.5 Audio Description (Prerecorded) AA · 1.1.1 Non-text Content A
Audio that starts on its own
Autoplaying audio interferes with hearing aids and with screen reader output, and there is often no obvious way to stop it.
Success criteria: 1.4.2 Audio Control A
The changes that actually remove the barrier.
Budget caption correction, not caption generation
Auto-captions are a first draft. A human pass to fix names, numbers, and terminology is the actual deliverable, and it is cheap relative to producing the video.
Publish the transcript as a page, not an attachment
A transcript on the page is searchable, linkable, indexable, and readable on a phone. It is also the single best SEO argument for accessibility work.
Contract live captioning for live events
Book CART for anything a member of the public or a student is expected to attend. Automatic captioning in the meeting platform is not equivalent.
Give every sound a visual counterpart
If a tone means something, show it too.
Three tests you can run before you call anyone.
- Mute your computer and watch your own most important video. Write down what you missed.
- Read the auto-captions on a video containing a person's name and a dollar amount. Count the errors.
- Check whether your last three webinars have transcripts posted.
If any of those go badly, you have your answer and you did not need a vendor to get it. What a vendor is for is the part after: knowing which of the 55 criteria that maps to, how deep it goes, and what it costs to fix properly rather than patch.
Run the Free CheckerWhere this shows up in practice
Other people the standard was written for.
Keyboard and Switch Users
If it works with a mouse but not with a keyboard, it does not work. This is the single highest-yield thing to test, and it takes ten minutes.
Read the profileVoice Control Users
Voice control users say what they see. When the visible label and the accessible name disagree, the command fails silently.
Read the profileAll 55 Success Criteria
Every Level A and AA criterion in plain language, with the failure we see most often for each.
Reference