Automatic captions now get most of the words right, and I think that makes it easier than ever to ship a video that still fails the people who depend on captions. A new WebAIM post argues that accurate text is only part of accessible captions.
The Web Content Accessibility Guidelines (WCAG) also expect captions to say who is speaking and to carry meaningful sounds. The author adds a warning I did not expect. Some captions sold as professional are worse than what the automatic tools produce.
What I took from it is that "human or machine" is the wrong question. It is the same question people ask about AI-written code, and it has the same answer. Judge what came back, not how it was made. For video that means someone watches it in the real player with the sound off and checks the speakers, the sounds, and the timing before it ships. A caption file being present is a checkbox, not a review.
In years of working on web teams, I have come to believe accessibility slips most at handoffs like this one, where content arrives from outside engineering and nobody's definition of done includes watching it.
Who on your team would notice if a video shipped with every word right and no sign of who was saying them?
Photo source: https://photos.robertstowe.com/victoria