Video transcript example: what one looks like and how it differs from captions

A video transcript has to carry things an audio transcript does not: text that appears on screen, actions that are shown rather than said, and the moments where the narrator says "this one" and points. Here is the first minute of a two-person demo.

[00:00:00] [On screen: Northbank Tools — Spring Release]

[00:00:03] ROSA MWANGI: Right, so this is the bit everyone has been asking about. I am going to open a project that already has four takes in it.

[00:00:11] [Rosa opens a project; four audio tracks appear stacked in the timeline]

[00:00:14] ROSA MWANGI: Four takes, and the third one is the good one, which you cannot tell by looking at it. That is the whole problem.

[00:00:22] JONAH PRICE: And previously you would have had to listen to all four.

[00:00:25] ROSA MWANGI: All four, every time. Now — [clicks Compare] — you get them side by side with the transcript under each one.

[00:00:33] [On screen: four transcripts appear in parallel columns]

[00:00:36] JONAH PRICE: So you are reading rather than listening.

[00:00:38] ROSA MWANGI: Reading is about six times faster. That is the entire feature.

Captions are timed to the frame and broken into short cues that fit on screen for a couple of seconds each. They are a synchronisation format, delivered as SRT or VTT, and they are read while the video plays.

A transcript is a document. It is read instead of watching, paragraphed for reading rather than chopped into cues, and it can carry description that captions have no room for. The example above is a transcript: nobody would put [Rosa opens a project; four audio tracks appear stacked in the timeline] on screen as a caption.

You usually want both, and they come from the same generated text. Produce the transcript, check it, then export the caption file from the corrected version so you are not fixing the same errors twice.

Search engines read the transcript on the page, not the caption track inside the player. If the video is the main content of a page, publishing the transcript underneath it is the difference between a page with words and a page without.

Mark on-screen text in square brackets with an [On screen: ...] prefix. Titles, lower thirds, anything written that a listener would otherwise miss. Automatic transcription will not catch any of it, because nobody said it.

Describe an action only where the words stop making sense without it. Rosa's click at 00:00:25 earns its line because the sentence has a gap in the middle otherwise. Her hand moving to the mouse does not.

Watch for deixis — "this one", "here", "like that". These are the sentences that collapse in a transcript, because the meaning was on screen. Either describe what was pointed at, or accept that the transcript is less useful than the video at that moment and say so.

If the video is instructional and the transcript will be used as a written how-to, describe more. If it is a talking head, describe almost nothing. The amount of description is a decision about who is reading, not a rule.

A descriptive transcript carries everything a viewer would get from watching, including visual information that the audio never mentions. It is what someone who cannot see the video relies on, and it is the accessibility standard for video whose meaning is partly visual.

The test is simple: read the transcript with the video closed. If you have lost information that mattered, the transcript is not descriptive enough yet.

For a demo like the one above, that means describing what appeared on screen at each step, not just that something was clicked. For a lecture with slides, it means the content of the slides.

Publish the transcript as text on the page rather than as a downloadable file where you can. A PDF behind a link is worse for screen readers and worse for search.

Pull the audio out of the video, transcribe it, check the four error clusters, then add the on-screen text and action lines by watching it through once. Doing the description pass last means you are adding to a corrected document rather than a draft.

The video below covers generating and downloading the transcript itself.