Accurate Captions Alone Do Not Make Video Accessible

Captions that transcribe every word accurately can still fail WCAG 2.2 Success Criterion 1.2.2, because the criterion's Understanding document requires more than synchronized dialogue: captions must "not only include dialogue, but identify who is speaking and include non-speech information conveyed through sound, including meaningful sound effects."

That gap between accurate text and an equivalent experience is the core argument of a new WebAIM article on caption quality. WCAG 1.2.2 makes captions a legal requirement for many prerecorded videos, and captions remain the canonical example of the curb-cut effect — an accommodation built for Deaf and hard-of-hearing users that hearing viewers adopt widely to improve comprehension. The article restricts its scope to prerecorded video, though several principles carry over to live streams.

What changed in automatic captioning?

The author describes a framework that no longer holds. Captioning used to split cleanly into two categories: automatic, computer-generated captions of poor quality, and professional, human-generated captions of high quality. Demonstrating the problem of automatic output once required no effort — within the first 30 seconds of playback, a video would typically show multiple examples of incorrect or missing dialogue.

Today, a growing number of automated tools generate accurate text from spoken dialogue. But accuracy is not the same as accessibility. For a viewer who relies on captions, missing or poorly represented information can mean missing an instruction, a warning, or the entire point of a conversation.

Which questions do the guidelines leave open?

The WCAG Understanding document establishes three essentials — spoken dialogue, speaker identification, and meaningful non-speech information — but stops short of prescribing implementation. The article lists unresolved decisions that directly affect the viewer:

• Verbatim dialogue with filler words ("Start on uh slide number five. … uh, I mean slide 6") versus trimmed text that preserves meaning ("Start on slide six").
• Whether speaker identification is required at every speaker change, even when the change is visually apparent.
• The absence of a standard format for non-speech information in caption files.

Authors who create their own captions face further formatting questions: how many lines to display at once, how many characters per line, whether to use open captions (always on) or closed captions (toggleable), and how long each caption should remain on screen. The article points practitioners to reference material rather than answering each question inline.

How should teams evaluate caption quality?

The author's recommendation is to judge output against established quality guidelines rather than treat the production method as a guarantee. Two cautions follow. First, no automatic tool the author has seen matches a professional, trained captioner in delivering a quality captioning experience. Second, payment does not ensure quality either: the author has encountered vendors marketing professional captions whose end product was worse than what automatic tools produce.

The practical test is concrete. Before publishing, review the captions in the player your audience will actually use and verify:

• Dialogue is accurate
• Speakers are identified
• Meaningful sounds are represented
• Timing, readability, and placement hold up

Low-quality captions may inconvenience some viewers, but they can completely exclude users who depend on them. The measure of caption quality, the article concludes, is whether viewers can rely on the captions to understand the message of a video.

Three resources anchor the evaluation work: the Described and Captioned Media Program's Captioning Key, with worked examples; the BBC Subtitle Guidelines, covering content, timing, and presentation; and the Caption Quality Standards summary from Utah State University, where WebAIM is located. As automatic captioning improves, review workflows built on these references, rather than on assumptions about how captions were produced, will determine whether video experiences stay equivalent for every viewer.