When a video comes back as a new version, the first question is always the same: did the editor do everything that was asked? AI Review answers it for you. On the new version it goes through the comments left on the earlier cut and gives each request its own verdict. If the earlier cut has no comments to check, there is nothing to verify and no verdicts appear.
Before you begin: this runs as part of AI Review, so the new version needs AI Review on — at upload, through Automatic AI review, or with Run AI Review afterwards. See When the AI review runs.
Which comments are checked
- The comments on the previous version. If the client has already seen an earlier cut — it was sent to them or they commented on it — the check goes back to that version, so a client request does not get lost behind an internal round.
- Replies count too. The final agreed wording often lives in a reply.
- Team only comments and hidden comments are left out.
- Short praise (“looks great!”) and plain questions are skipped: there is nothing to check.
What each verdict means
| Verdict | Meaning |
|---|---|
| Done | The change is there in the new version |
| Partial | Part of the request was done |
| Not done | The requested change was not found in the new version |
| Unclear | The check could not reach a confident answer |
| Check by eye | The request cannot be checked automatically — taste, pacing, a vague note |
A Not done without enough confidence is shown as Unclear, so red is kept for what is almost certainly still pending.
How a request is checked
The method depends on what the comment asks for:
- Text on screen — a title, a name, a lower third: the on-screen text of the new version is read from that point to a few seconds later and compared with the request.
- Cuts — “remove this part”: the speech at that point is compared in both versions.
- Audio — “lower the music”: the loudness of that section is measured in both versions.
- Visual changes — a logo, a color, a shot: frames from both versions are compared at that point.
Cuts between versions shift everything in time. The check lines the two cuts up first, so a request at 1:20 on v1 is checked where that moment now sits on v2.
Where to see the verdicts
- Open the new version. At the top of the comments, a line reads, for example, 4 requests from v1 · 1 not done.
- Click Compare with v1 on that line. The comparison opens, with the previous version’s comments on the side.
- Each old comment carries a verdict chip. Click a chip to see how it was checked, what was read or measured, and the confidence.
- Use See in this version to jump to that moment in the new cut. On a Done or Partial verdict, Mark as resolved closes the original comment.
Before sending to the client
If some requests are marked Not done with high confidence, the Send to client dialog warns: “1 request from the previous version does not seem to be addressed. Send anyway?” Check the Previous version requests list before you confirm.
The script check on later versions
When the task has a script, the first version is compared against the whole script. Later versions do not repeat the full check:
- On an internal round with the same script, only the lines of speech that changed are checked. If the script changed, the whole script is checked again.
- Once the client has seen a version and the script did not change, the script check is skipped — what the client saw and did not flag is taken as accepted. If the script changed, only the new lines are checked.
- Once the client has commented on an earlier version, the review focuses on those comments and on technical issues instead of the script.
This is also why new versions of the same video finish faster.
Limits
- It is a check, not a verdict on quality. A Done means the change is there, not that it is good.
- A comment with no clear request, or one that depends on taste, gets Check by eye.
- Nothing changes status on its own: the verdicts inform your decision.
- When Hide AI Review for clients is on, clients and visitors do not see the verdicts either.
What’s next
See the full list of what the review looks at in What AI Review checks, and how to work with cuts in Versions and comparison.