Captions are watched more than they are read. A large share of feed video plays muted by default, which means for many viewers the captions are the audio. They are also the part of the edit most often rushed, because they happen last and by then the interesting work is done.
A caption pass has a correct order, and most caption problems come from doing it out of order.
Lock the spoken edit first
Captions are timed to speech. Change the speech and every caption after the change is wrong. If you caption before finishing silence and filler cleanup, you are guaranteeing a retiming job, and retiming captions by hand is worse than making them in the first place.
So: silences, then fillers, then watch it through, then captions. If you make a substantive cut after captioning, expect to regenerate rather than patch.
Fix the words before you style anything
Transcription is very good and still wrong about the things that matter most: names, company names, product names, technical terms, numbers with units. These are exactly the words a viewer is relying on the caption for, and exactly the words a spellchecker will not catch because the wrong word is usually a real word.
Read the transcript through once before touching a font. Styling a transcription error just produces a beautifully set mistake, and it is demoralising to discover after you have placed everything.
Choose a look that survives your footage
Caption styling is a legibility problem first and a brand problem second. Text needs to hold against whatever is behind it, which on real footage means bright skies, white walls, and whatever your B-roll does. A background plate, an outline, or a shadow is what makes text readable over an unpredictable image; pure white type with none of them will vanish somewhere.
Keep lines short. Two lines of roughly forty characters is a reasonable ceiling for feed video: long enough to carry a phrase, short enough to read in the time it is on screen.
- Keep captions clear of the lower third if you also use lower thirds.
- Stay inside title-safe margins so nothing is clipped by a platform UI.
- Check contrast against your brightest shot, not your average one.
Time captions to phrases, not to word counts
A caption that breaks mid-phrase makes a viewer work. Break at natural boundaries (clause ends, before a conjunction, where the speaker pauses) rather than wherever the character limit happens to land.
Check that each caption appears with the words it represents and clears before the next phrase begins. Captions that lag are worse than captions that lead slightly, because a late caption is read after the moment it belongs to.
Watch it at full speed, muted
The final check is the one that matches how people will watch: play the sequence at full speed with the sound off and see whether it is followable. This catches timing problems, contrast failures and overlong lines that reading the timeline will not.
It also catches the case where captions are technically correct and the video still does not work muted, usually because something important was only ever in the audio.
Common mistakes
Captioning before the edit is locked
Guarantees retiming. This is the big one and it is entirely avoidable by ordering the passes correctly.
Styling before proofreading
You end up with well-set wrong names. Read the transcript first.
White text with no plate or outline
It will be invisible over something. Give the text a background it can rely on.
Breaking lines by character count
Break at phrase boundaries instead. A caption split mid-clause slows every viewer down slightly.
Questions
Should captions be burned in or a separate track?
Um Out creates timed caption items on your timeline, so they stay editable and you decide how to export. Separate keeps flexibility; burned in guarantees the look survives re-uploads. For feed video, burned in is the common choice.
How accurate is the transcription?
Accurate enough that proofreading is a pass rather than a rewrite, and not accurate enough to skip proofreading. Names and specialist terminology are where errors concentrate.
Do captions help with reach?
They make a video usable muted, which is how a lot of feed video is watched, and they make the spoken content available to people who need or prefer text. Both are real. We will not put a number on it.
Can I save a caption style and reuse it?
Yes. A look you have set up can be saved and applied again, which is what makes captioning a series practical rather than a per-video design exercise.