The single most repeated statistic in social video — that most people watch with the sound off — is repeated because it keeps being true. Which means for a large share of your audience the captions are not an accessibility feature. They are the audio track.
Two kinds of caption, often confused
| Burned in | Platform auto-captions | |
|---|---|---|
| Where they live | In the pixels of the video | A subtitle track the app renders |
| Viewer can turn them off | No | Yes, and many do |
| Typeface and colour | Yours | The platform's |
| Position | Exactly where you put it | Wherever the app decides |
| Accuracy | Whatever you typed | Good, not perfect, worse with accents |
| Survives cross-posting | Yes | No — regenerated per platform |
| Screen-reader accessible | No | Yes |
The specifics that make them readable
Captions fail for mechanical reasons far more often than stylistic ones. Every item here is a measurement rather than a taste.
- Three to four words a line, two lines maximum. A full sentence across the screen is a paragraph at phone size, and the viewer reads it instead of watching.
- Big. Around 60–80px at 1080 wide is a sensible floor. If it looks slightly too large on a laptop it is about right on a phone.
- A heavy weight. Semibold or bold. Thin typefaces disappear into busy footage and are the first thing compression destroys — why Instagram ruins your video quality.
- Contrast that does not depend on the shot. A dark scrim behind the text, or a solid outline. White text alone works until one clip is a bright sky.
- In the middle third, roughly 40–65% of the way down. Below that the caption and handle cover it; above it the interface does.
- Never centred over a face. Sit it under the subject, not across them.
Timing them to the cuts
On a beat-matched video there is an obvious answer that most people miss: change the caption when the shot changes. The cut is already a moment of attention, and a line of text that arrives with it reads as deliberate rather than as a subtitle track running underneath.
Which also sets the pace for you. Cutting on every second beat at 120 BPM gives a new shot every second, so a caption line per shot is roughly one line per second — close to comfortable reading speed. Cutting on every beat at 140 BPM does not leave time to read anything, so on those, caption sparingly or not at all. How many cuts a reel should have is the arithmetic.
The hook line is not a caption
Keep them separate. The hook is one fixed line at the top of the frame that states the promise and stays put for the first few seconds; captions are the running transcript underneath. Putting the hook in the caption stream means it scrolls away before anyone has decided to watch.
Two positions, two jobs, two styles — the hook larger and fixed, the captions smaller and moving. Reel hooks covers what the hook line has to say; reusable text overlays is how both stay identical across every video you make.
A workflow that does not take an hour per video
- Write the script first, not the captions. Captions are the script split into lines; transcribing after the fact is doing the work twice.
- Split on breath, not on grammar. Where you would pause speaking is where the line should break, even mid-clause.
- Style them once as a saved overlay — typeface, size, weight, scrim, position. Never restyled per video.
- Time them to the cut points you already have from the track.
- Proofread on a phone. Typos in burned-in captions cannot be fixed after posting without deleting the post.
Captions are also SEO
Text on screen is read by the platforms and used to work out what a video is about, which makes captions a ranking surface as well as a readability one. Two consequences worth acting on: say the phrase somebody would search at least once, and let the platform's automatic caption track run alongside your burned-in text so the words exist as data and not only as pixels.
In Easy Reels captions and the hook are both text layers on the template, styled once and reused, with the lines drawn from the copy pool so the words vary while the look does not — which is the same argument as everywhere else on this blog: the faults that make a reel look amateur are nearly all things that happen when a decision gets made again per video.