Most people who hear fine use “captions” and “subtitles” interchangeably. The people who depend on them never do — and that difference tells you which one your video needs.
The one-line difference
Subtitles translate. They assume the viewer can hear the audio perfectly and just doesn’t speak the language. Captions transcribe. They assume the audio is unavailable — because the viewer is deaf or hard of hearing, or because the platform plays video muted until tapped.
A caption describes the sound as well as the speech:
- Caption:
[phone buzzes] Are you seeing this? [laughs] - Subtitle:
¿Viste eso?
That’s the whole distinction: captions carry non-speech information, subtitles carry another language.
Side by side
| Subtitles | Captions | |
|---|---|---|
| Audience | Hears the audio, doesn’t understand the language | Can’t hear the audio, or watches muted |
| Content | Translated dialogue, often condensed | Dialogue plus [sound cues] and speaker labels |
| Triggered by | International releases, dual-language audiences | Accessibility, autoplay-muted feeds |
| Convention | Follows speech, not word-for-word | Near-verbatim, plus everything audible |
Platforms blur the line. YouTube calls translated tracks “subtitles” and sound-inclusive tracks “captions,” both behind the CC button. Broadcasters say “open” for burned into the picture and “closed” for toggleable. The underlying files are usually SRT or WebVTT.
Which one you actually need
- You post to social feeds. Add captions. Feeds autoplay muted, and without captions your dialogue slides past unread.
- You publish for an international audience. Add subtitles in the languages your viewers actually speak. Machine translation is a starting point, not a finish line.
- You need both. Start from accurate captions, then derive subtitles from them. One trustworthy transcript feeds both.
Whatever you ship, the readability rules are the same: two lines maximum, roughly 37 characters per line, 160–180 words per minute, and never place text over faces or on-screen titles. Auto-captioning is decent at the first pass and unreliable at names, jargon, and homophones — “cache” versus “cash” changes the meaning of the video.
The test before publishing: mute the video and watch it. If you lose the story, you needed captions. If a non-speaker would lose it, you needed subtitles.