Publishing workflow · free review worksheet

How to check video captions before you publish

Automatic captions save transcription time, but a plausible-looking line can still change the meaning of a product claim. This review catches the errors that matter in short ads, demos, podcasts, and Shorts.

Published September 29, 2026 · Updated October 9, 2026 · 9 min read

Two different jobs: a transcript records what was said; captions synchronize speech and meaningful audio with the video. A text summary or shot report is useful for planning, but it is not a checked caption track.

Why review automatic captions?

YouTube explicitly asks creators to review and edit automatic captions. It notes that accents, pronunciation, background noise, and overlapping speakers can produce mistakes. For a demo, “can” becoming “can't” is a materially different promise; for an ad, a brand or price error can reach customers before anyone spots it.

Accessibility is another reason to check more than words. W3C's transcription guidance includes relevant non-speech sounds, and its prerecorded-caption guidance explains that captions convey audio needed to understand the content. A caption track that omits an off-screen speaker or a meaningful sound can leave a viewer missing the point.

Five-pass caption QA

1. Verify the exact words. Play with sound and follow every caption line. Check names, acronyms, numbers, prices, negations, technical terms, and calls to action against the source. Mark uncertainty; do not invent a word to fill a gap.
2. Check the timing. Watch each cut. Does a line appear when its speaker begins and disappear when that thought ends? In a rapid montage, delayed captions can attribute the previous speaker's words to the next shot. Pause at each cut rather than trusting the waveform alone.
3. Identify speakers and meaningful sounds. When the speaker is ambiguous, label them consistently. Add sound cues only when they carry meaning, such as [door slams] before a reveal or [applause] after a claim. Avoid flooding the screen with decorative cues.
4. Read on a phone-sized preview. Check line breaks, contrast, text size, and whether captions collide with a product UI, lower-third, or platform controls. The transcript can be accurate while the published video remains hard to read. Preview the actual placement, not just the editor canvas.
5. Compare sound-on, sound-off, and transcript-only viewing. A mute pass tests whether the essential message survives without audio. A transcript-only pass catches missing context from visuals, such as a result shown on screen but never said aloud. Add an accompanying text description or edit the video if that context is essential.

A concrete 30-second example

Illustrative example, not a real customer report. Imagine a product demo in which a founder says, “Export the report in under ten seconds.” Auto-captions show “Export the report in ten seconds.” At 0:12, a second speaker asks “Without a card?” but the line stays under the founder's shot. At 0:20, a notification chime confirms the export, yet the sound is omitted and the result is off screen.

The correction list is specific: restore “under,” move the question to the correct speaker and timestamp, and make the successful export understandable both visually and in text. None of those edits requires a speculative engagement claim.

Download a caption QA worksheet

Download the free caption QA CSV and open it in Excel, Google Sheets or your preferred spreadsheet. It is a blank manual review sheet, not an automatic caption checker or a timed subtitle file. No payment is needed to download this worksheet.

Keep one row per issue and identify the exact video export and caption track. For a short ad, review the complete clip. A few spot checks can miss the one line containing a price, promise or negation.

Source filename and version
Caption track and language
Issue start | Issue end | Speaker
Current caption | Verified wording or sound
Issue type | Required correction
Decision | Reviewer | Reviewed date
Revised export | Placement or published reference
Recheck date | Recheck decision

Make the review decision explicit

These labels are an editorial workflow for your team. They are not generated confidence scores or an accessibility certification.

A corrected caption still needs a final-version check

Track text, timing and delivery separately. Fixing a word does not move a late caption. Moving a caption does not correct its speaker. Exporting a new video does not prove that the uploaded version uses the revised track.

  1. Replay the affected range with sound in the caption editor. Check the line before and after it so the correction keeps its context.
  2. Apply the wording and timing changes to the intended language track.
  3. Replay the revised export or published caption track. Confirm the filename or published reference in your worksheet.
  4. Check a phone-sized placement for collisions, clipped lines and unreadable text. Log this separately from whether the words are correct.
  5. Record the recheck decision. Leave it unresolved when you cannot access the final version.

Illustrative worksheet decisions. In the hypothetical demo above, restoring “under” can be marked Corrected after the text edit. It becomes Verified only after replaying that revised line. Moving the question at 00:12 requires a separate timing check. If the final upload is not available, its recheck stays Unresolved even when the local edit looks right.

A teammate can now revisit the affected ranges and see which version was checked. This issue log does not replace the initial full caption review, but it avoids repeating that entire review just to resolve a documented correction.

Fix caption text and timing in YouTube Studio

YouTube's caption-editing instructions distinguish text edits from timestamp edits. In Studio, open Subtitles, select the video and language track, then choose the appropriate edit option. Automatic tracks use Duplicate and edit. Review the wording, adjust the start and end timestamps where needed, and publish the revision. Check the resulting track during playback.

YouTube also lists reasons automatic captions may be missing or delayed, including audio quality and overlapping speech. A missing track is not evidence that the video contains no dialogue. If you cannot understand a line after replaying the original, keep it unresolved and request a clearer recording or a source transcript.

Where AdScanVideo fits

Upload a video you have the right to use and choose Podcast notes for spoken content, Shot breakdown for the visual sequence, or Ad intelligence for an ad review. A generated report can help you locate candidate ranges and understand their context. Replay the original audio before treating a line as an exact quote. A report time reference is not a verified caption cue.

AdScanVideo does not export a timed SRT or VTT caption track and does not replace a caption editor. The free worksheet above organizes your manual corrections; it does not automatically compare audio, change captions or certify accessibility. For a fuller wording check, use our transcript verification workflow.

For YouTube, make the final text and timing changes in YouTube Studio's caption editor or another editor that produces a timed caption track. Our dialogue-extraction guide explains what to do when a transcript is unavailable; our demo audit helps check the rest of the creative. For vertical placements, use the safe-zone checklist after the text is correct.

Common questions

Are subtitles and captions the same thing?

People often use the terms interchangeably. For this checklist, captions include speech and the meaningful non-speech audio needed to understand the clip; translated subtitles may primarily translate speech. Check the requirements of the platform and audience you serve.

Does a Word or PDF report include timed captions?

No. A report is a document, not a caption track. Use a caption editor to create and check timed subtitles. Downloading a report does not verify its dialogue against the audio.

Can I publish automatic captions unchanged?

You can, but do a human review first. If the audio is noisy or speakers overlap, the error risk rises. YouTube's own guidance recommends reviewing and editing its automatic output.

Does a transcript make a video accessible by itself?

A transcript helps, but timed captions serve a different purpose during playback. A transcript may also need visual information added to communicate the full video, as W3C explains.

Sources and further reading

Find the moments worth checking.

Analyze a video you have the right to use, then compare its visual timeline and available dialogue with your caption draft. One daily on-screen analysis is free without signup, up to 10 minutes and 200 MB. Built-in Copy, Word and PDF for an eligible existing free report unlock with one credit, without rerunning the analysis. The blank caption QA worksheet is free to download.

Analyze a video free