Most viewers meet your clip on mute. That makes captions the primary audio channel — and yet most clippers treat them as an afterthought toggle. Caption craft is one of the highest-leverage skills in clipping precisely because it's invisible when done right and fatal when done wrong.
The baseline: word-timed, not block-dumped
Word-by-word (or 2–3 word phrase) captions timed to speech outperform static blocks because they recreate the audio experience visually: pace, emphasis, comedic timing. A punchline that pops onto the screen at the exact moment it's spoken lands even on mute. A paragraph sitting under the video the whole time is subtitles, not captions — informational, not experiential.
Auto-captioning (CapCut and most editors) gets you 90% there. The paid skill is the pass afterward:
- Fix every error. One wrong word breaks trust in all of them. Names, game terms, and slang are where auto-transcription dies — exactly the words your niche audience will notice.
- Re-break the lines by meaning. Auto tools break by duration; humans break by phrase. "I have never / seen anything like that" reads wrong; "I have never seen / anything like that" reads right.
- Sync the beat words. The emphasized word should hit the screen on its spoken frame — that's the whole comedy-timing trick.
Sizing and placement
- Big enough for a phone held at arm's length — err larger than feels tasteful on your desktop preview.
- Upper-middle placement. Every platform's UI eats the bottom (progress bars, captions fields, action rails) and the right edge. The safe vertical band is roughly the middle 60% of the frame — platform-specific zones covered in each playbook.
- Two lines maximum on screen at once. Three lines is a wall; walls get scrolled.
Emphasis without noise
One accent treatment, used sparingly: a single color for the beat word, or a scale-pop on the payoff. The moment everything is highlighted, nothing is. Rainbow-word styles read as template spam — and feeds full of identical caption templates are exactly where distinct edit taste wins.
Emoji, profanity, and platform hygiene
- Emoji in captions: one per beat maximum, only when it adds a reading the words don't carry.
- Mask profanity in the caption text even when the audio carries it — visible profanity trips distribution filters more readily than spoken.
- Keep claims out of caption frames that the clip doesn't deliver; bait captions poison the account's rewatch data.
The ten-minute caption pass
For a 30-second clip: auto-generate (1 min) → error fix (3) → line re-breaks (2) → beat-word sync (2) → size/placement check on an actual phone (2). Ten minutes that routinely doubles a clip's completion rate — which, on pay-per-view campaigns, is arithmetic that pays you directly.
Captions are the score under the film. Learn to conduct.