Guides
Caption Timing Best Practices: CPS, Line Breaks, and Duration Rules

Most creators obsess over caption accuracy and forget about caption timing — and timing is what viewers actually feel. A caption that appears half a second late makes the speaker look out of sync. A caption that vanishes too fast makes the viewer rewind. A caption that lingers after the speaker moved on makes the whole video feel broken. The words can be perfect and the video can still be hard to watch.
The good news: professional caption timing is not guesswork. Broadcasters and streaming platforms have settled on a small set of rules — characters per second, minimum and maximum durations, line-length limits, and where to break lines — and those same rules work just as well for a YouTube video, a course lesson, or a 30-second Reel.
The three numbers that control caption timing
Every caption is defined by three numbers: when it appears, how long it stays, and how much text it holds. Get those three right and almost everything else falls into place. The industry shorthand for the relationship between them is CPS — characters per second — which is simply the number of characters in a caption divided by its duration in seconds. A 42-character caption shown for 2 seconds runs at 21 CPS; the same caption shown for 3 seconds runs at 14 CPS. Same words, very different reading experience.
The other two numbers are guardrails around CPS. A caption should stay on screen for at least about one second and rarely more than about seven — longer than that and it starts to look frozen, especially if the speaker has moved on to a new thought. And a single caption should hold no more than about two lines of text, roughly 42 characters per line. These limits interact: if a speaker talks fast, you cannot just stretch the caption — you split it into two captions instead.
Characters per second: the speed limit your viewers can actually read
CPS is the single most useful quality check for captions, because it catches the problem viewers complain about most: text that disappears before they finish reading it. The widely used broadcast target is 17 CPS for adult programming, with 20 CPS as a practical ceiling for most online video. Children's programming aims lower, around 12–15 CPS. Fast-paced social video can push toward 20 CPS, but only when the audience is skimming familiar, simple language — never for technical explanations or accented speech.
Here is how to use the rule in practice:
- Check the worst captions, not the average. A video can average 14 CPS and still have five captions at 28 CPS. Scan for the spikes — those are the ones viewers rewind for.
- When CPS is too high, split — don't stretch. Extending a caption past the next line of speech creates overlap and confusion. Split the sentence across two captions at a natural pause instead.
- When CPS is too low, merge. A six-word caption held for six seconds looks dead on screen. Merge short fragments into one caption that covers the whole sentence.
- Count everything. Spaces and punctuation count toward CPS. A caption reading "Well, actually — no." is 21 characters, not 15.
Most caption tools and editors will flag CPS for you, but you can also check it by hand: count the characters, divide by the duration, and compare against 20. If you generate your SRT with Captionbench's free browser tool, the timestamps come from the actual speech, so CPS problems usually trace back to a caption that spans two sentences — split it and the numbers fix themselves.

Line breaks and line length: where to split, and where never to split
A caption should be at most two lines, and each line should be no longer than about 42 characters — shorter is better on mobile, where 32–37 characters per line keeps text comfortably inside the frame. Two short lines are easier to read than one long line, because the eye travels a shorter distance and the text block stays compact.
Where you break the line matters as much as how long it is. Break at natural phrase boundaries: before a conjunction, after a comma, or between clauses. A viewer should be able to read the first line, drop to the second line, and never lose the meaning.
- Good break: "We tested three tools / and only one kept up" — each line is a complete thought.
- Bad break: "We tested three / tools and only one kept up" — the eye has to reassemble the phrase.
- Never split these: a person's name, a number and its unit ("50 dollars", "3 minutes"), an article from its noun ("the / video"), or a verb from its object when the two lines would be read as separate ideas.
- Keep the top line shorter. When lines are uneven, put the shorter line on top. A pyramid shape (short line above a longer one) reads faster than an inverted pyramid.
One more habit from broadcast practice: balance the lines. If a caption has 60 characters, "30 / 30" beats "42 / 18" — two even lines are scanned faster than one long line and one stub.
Duration rules: minimum, maximum, and the gaps between captions
Duration is where most auto-generated captions go wrong, because speech recognition systems often produce one caption per breath or per sentence fragment, with no regard for how long each one stays on screen.
- Minimum duration: about 1 second. Anything shorter flashes by unread. If a word is spoken that fast — a quick "yes" or "right" — merge it into the caption before or after it.
- Maximum duration: about 7 seconds. Even a long, slow sentence should be split if it would need more than seven seconds on screen. Viewers assume a caption that old belongs to an earlier moment.
- Minimum gap: 2 frames (about 80 milliseconds). Two captions that touch exactly can flicker or merge visually in some players. Leave a tiny gap — most editors do this automatically, but check when you hand-edit timestamps.
- Match the shot, not just the speech. A caption should not cross a hard scene cut if it can be avoided. When the picture changes completely, the viewer re-reads the caption as if it were new — so split at the cut.
A useful rhythm to aim for: most captions in a normal talking-head video land between 1.5 and 4 seconds. If yours are consistently under a second, your sentences are being split too aggressively; if they are consistently over five seconds, you are probably captioning whole paragraphs at once.
Syncing captions to speech: lead-in, shot changes, and speaker changes
Timing is not only about duration — it is about alignment. A caption should appear at the moment the words begin, or a fraction before. Professional practice allows a caption to lead the audio by up to about 100 milliseconds, which feels instantaneous to the viewer, but it should never lag behind the speech by more than a beat. If you have ever watched a video where the caption arrives after you already heard the word, you know how distracting even a 300-millisecond lag feels.
Three alignment habits make the biggest difference:
- Start the caption with the speaker, not before. A caption that appears a full second early spoils the line — especially a punchline. Trust the speech timestamps and only nudge a caption earlier when it would otherwise collide with a shot change.
- Change captions when the speaker changes. In interviews, never let one caption span two speakers. Start a new caption for the new speaker, even if the previous caption was short.
Common timing mistakes (and how to fix them)
- The wall of text. One caption holding three sentences for eight seconds. Fix: split at sentence boundaries so each caption is one idea, 1–4 seconds long.
- The flash caption. A single word appearing for half a second. Fix: merge it into the caption before or after it — no caption should be under one second.
- The drifting caption. Auto captions that lag further and further behind as the video goes on. Fix: this is usually a frame-rate mismatch between the audio and the subtitle file. Re-export the SRT from the same timeline, or re-time from the audio.
- The caption that crosses a cut. Text from the previous scene hanging over a new shot. Fix: end the caption at the cut and restart it after, even if the sentence continues.
- Ignoring CPS on fast talkers. Interviews with rapid speakers can hit 30+ CPS. Fix: edit the wording — cut filler words ("you know", "like", "um") from the caption. Captions are a readable version of speech, not a court transcript.
Key takeaways
- Keep captions at or under 20 characters per second — split sentences instead of stretching them when a speaker talks fast.
- Cap captions at two lines and about 42 characters per line, break at phrase boundaries, and never split a name, a number, or an article from its noun.
- Hold each caption between about 1 and 7 seconds, leave a small gap between captions, and don't let a caption cross a hard scene cut.
- Align captions to the moment words are spoken, change captions when the speaker changes, and cut filler words to keep fast speech readable.
Ready to put these rules into practice? Generate free timed captions with Captionbench right in your browser — no account, no uploads, SRT/VTT/TXT export included.