A large majority of social video is watched without sound — in public, in bed, at work, in a queue. Uncaptioned video is therefore silent video for most of the people who see it.
Two audiences, one fix
Deaf and hard-of-hearing viewers need captions to access your work at all. Everyone watching on mute needs them to follow it. The same fifteen minutes of work serves both, which makes this one of the highest-return habits available.
Auto-captions need editing
Automatic captions are good and not good enough. They mangle names, technical terms, brand names and anything spoken quickly — which is usually the most important vocabulary in the video. Read them through before publishing and fix the errors.
Burned-in versus platform captions
Platform caption tracks are accessible to screen readers and can be turned off. Burned-in text is always visible and survives re-uploads. Doing both covers the widest range, and platform captions additionally feed the transcript that classification systems read.
Design considerations
Keep captions out of the bottom fifth of the frame, where interface elements sit. Use high contrast. Two lines at a time, maximum. Legibility beats styling.
The discovery side effect
Transcripts are text, and text is searchable. Accurate captions make your video findable for the words actually spoken in it — a real benefit, though it should be the second reason you do this rather than the first.