Adding a couple the other replies didn't cover, mostly from making stuff for people who read English as a second language:
- Read the captions out loud at normal speed while the video plays. If you can't finish a line before it disappears, a second-language viewer definitely can't. My rough rule is no line on screen for less than about 1.5 seconds.
- Break lines at natural phrase points, not wherever the width runs out. "I didn't think it / would work" is way harder to read than "I didn't think / it would work".
- Swap idioms and slang out of the captions even if you say them in the audio. Plain words in the text, personality in the voice.
- Check on the worst screen, not your best one. Cheap Android, brightness down, outside if you can. White text with no outline or shadow falls apart there.
If you want one repeatable step: export a 15 second chunk, send it to someone who isn't a native speaker, and ask them what was said. Takes two minutes and catches most problems.