Why is the gap bigger at the end than the start?

That pattern means picture and sound are running at slightly different rates, so the distance between them accumulates. A single offset would stay the same size from beginning to end.

Can I nudge the text without redoing the clip?

Here the text is placed against the clip itself, so the two are fixed together and there is nothing to slip. If you are working elsewhere, look for a global offset control before you touch individual lines.

Does trimming the front of a clip move the text?

In tools where text lives in a separate layer, yes, and that is a common cause. Trim first, add words after, and the problem does not arise.

Is half a second actually noticeable?

Yes. Viewers read faster than they listen, so early text spoils the line before it is spoken. Late text is more forgiving but still reads as sloppy.

Which is worse, early or late?

Early. A caption that arrives ahead of the voice removes the surprise from whatever is being said, which is exactly what a short clip is selling.

Does this happen more on long recordings?

Rate drift accumulates, so a two hour source shows it far more clearly than a ten minute one. An offset stays the same size regardless of length.

Can I check without watching the whole thing?

Watch ten seconds at the start and ten at the end. If the gap is the same in both, it is an offset. If it is worse at the end, it is drift.

Subtitles Out of Sync With Speech

Why does a caption drift out of time?

Three causes account for almost all of it. The source file has audio and picture recorded at slightly different rates. The clip was trimmed after the text was already placed. Or the text was written against a different copy of the video than the one posted. The first two show as steady drift, the third as a fixed offset.

Telling the three causes apart

Watch the last ten seconds of the clip with the sound on, then match what you see against the first column.

What you see	What it means	What fixes it
Gap grows towards the end	Picture and sound recorded at slightly different rates	Export the source again at a fixed rate before working with it
Gap the same size throughout	A single offset, usually from a trimmed start	Place the text against the trimmed file, not the original
Correct at first, wrong after a cut	The clip was trimmed after the text was placed	Trim first, add words afterwards
Only wrong in one section	That stretch has unclear or overlapping sound	Read and correct that section before producing
Wrong on one platform only	The upload was processed differently there	Compare against your own file before assuming the clip is at fault

Words that sit where the voice is

Watch the last ten seconds first. Drift grows over time, so the end of a clip shows it far more clearly than the start.
Decide: steady slide or fixed gap. A gap that stays the same size all the way through is an offset. A gap that widens is a rate problem.
Work from one copy of the video. Place text against the exact file you will post, not an earlier export of it.

Send one video through and watch the timing on the last few seconds. The first one costs nothing.

Questions people ask

Why is the gap bigger at the end than the start?: That pattern means picture and sound are running at slightly different rates, so the distance between them accumulates. A single offset would stay the same size from beginning to end.
Can I nudge the text without redoing the clip?: Here the text is placed against the clip itself, so the two are fixed together and there is nothing to slip. If you are working elsewhere, look for a global offset control before you touch individual lines.
Does trimming the front of a clip move the text?: In tools where text lives in a separate layer, yes, and that is a common cause. Trim first, add words after, and the problem does not arise.
Is half a second actually noticeable?: Yes. Viewers read faster than they listen, so early text spoils the line before it is spoken. Late text is more forgiving but still reads as sloppy.
Which is worse, early or late?: Early. A caption that arrives ahead of the voice removes the surprise from whatever is being said, which is exactly what a short clip is selling.
Does this happen more on long recordings?: Rate drift accumulates, so a two hour source shows it far more clearly than a ten minute one. An offset stays the same size regardless of length.
Can I check without watching the whole thing?: Watch ten seconds at the start and ten at the end. If the gap is the same in both, it is an offset. If it is worse at the end, it is drift.