Paste each person's feedback into an AI assistant with its sender and timecode, and ask for one table: timecode, requested change, who asked, any conflict with another comment, and whether the contract's revision round covers it. Fix the table by hand, send it to the client's one approver, and start editing only after written sign-off.
The assistant is good at the clerical part: merging duplicates, spotting contradictions and sorting by timecode. It is poor at two things you must keep for yourself. It will tidy or invent timecodes if you let it, and it cannot decide who is right. Three stakeholders will disagree somewhere, and the list's job is to show that disagreement to the one person who can settle it.
A worked consolidation: 23 comments, three stakeholders, one list
Take a 90-second brand film, version 2, delivered on Monday. The contract includes two rounds of revisions and this is round one. By Thursday the feedback has arrived in three ways: seven emails from the founder, eleven comments in the review tool from the marketing lead, and five more in a reply-all from the operations manager. That is 23 comments across two channels, several of them saying the same thing, and some saying opposite things. Done by hand, sorting them into a list takes about 50 minutes (an illustrative figure, as are the others in this example).
Give each raw comment an ID and a source before you paste. A selection of what goes in:
- F1, email: "Opening drags. Cut it down a bit."
- F2, email: "Love the testimonial. Could it run longer? Around 0:41."
- F5, email: "Music feels flat, can we try something more upbeat?"
- M1, review tool, 00:04: "Trim the opening shot, a few seconds."
- M4, review tool, 00:41: "Testimonial runs too long. Take 5 seconds out."
- M7, review tool, 01:12: "Logo needs to be bigger."
- O2, email: "The product at 00:55 still has the old label. Needs replacing."
- O3, email: "Music is fine, it's licensed and approved."
- O4, email: "Keep the end logo subtle, per brand guidelines."
Here is the result after one pass and one round of corrections, shortened to eight of the 16 rows:
| ID | Timecode | Requested change | Asked by | Flag |
|---|---|---|---|---|
| 1 | 00:04 | Shorten the opening shot by a few seconds (no length given) | F1, M1 | Merged duplicate. Ask how many seconds |
| 2 | 00:41 | Testimonial: longer (F2) or 5 seconds shorter (M4) | F2, M4 | DECISION |
| 3 | 00:55 | Replace the product shot that shows the old label | O2 | POSSIBLE EXTRA: needs new footage |
| 4 | 01:12 | Logo: bigger (M7) or subtle, per brand guidelines (O4) | M7, O4 | DECISION |
| 5 | 01:22 | Correct the spelling in the lower third | M10 | In scope |
| 6 | NO TIMECODE | Colour feels a bit cold | F6 | YOUR CALL: ask which scene |
| 7 | NO TIMECODE | Try a more upbeat music track (F5); Operations says keep the licensed track (O3) | F5, O3 | DECISION |
| 8 | NO TIMECODE | A 15-second cut-down for social | M9 | POSSIBLE EXTRA: new deliverable |
The full list has 16 rows. Of the 23 comments, four were duplicates and merged, which leaves 19. Six of those formed three contradictory pairs, and each pair became one decision row, so 19 minus 6 plus 3 gives 16: 11 changes within scope (one of them, the colour note, too vague to start until the client names a scene), two possible extras and three decisions. The assistant took about four minutes. The check took about 15, so the whole job was roughly 20 minutes instead of 50, and the list is better than a hand-typed one because every row names its sources.
The first draft needed four corrections, and all four are typical:
- An invented number. Row 1 first read "shorten the opening by 4 seconds". Nobody said 4, so it became "a few seconds" and a question.
- A decision made for you. Row 2 first read "compromise: trim by 2 seconds". That is a creative call and a political one, and it was deleted.
- A dropped voice. Row 7 first listed only F5, because O3 did not ask for a change. O3 is the other half of the conflict, so it was added back.
- A quiet fix. The assistant corrected "Seperate" to "Separate" inside the requested-change text, which is right here but would hide a misquote elsewhere. Keep the original wording in a quotation when the exact words matter.
Collect everything in one place before you prompt
Most failures start with the inputs. Before you open a chat, gather the review-tool comments, the emails and any call notes, and prepare them like this:
- Give every comment an ID, a sender and a source. F for founder, M for marketing, O for operations, then a running number. The ID is how every merged row stays traceable.
- State the version. Feedback written on version 1 but sent after version 2 can leave you re-doing changes that are already made. Write "all comments are on v2" at the top, and mark any that you know refer to an earlier cut.
- Note any timecode offset. If your timeline starts at 01:00:00:00 and the review player starts at zero, write the offset in the prompt and let the assistant copy times as written. Convert only when you build the final list.
- Keep timecodes as written. "0:41", "00:41" and "41s" can all be correct; frame numbers are not worth guessing at.
- Strip signatures and quoted replies. They add words and no information.
- Use roles instead of names if you prefer. "Founder" and "Marketing" are enough for the assistant.
On the tool side, Frame.io's plan comparison lists frame-accurate and range-based comments, with download and export of comments and copy-and-paste of comments among its features. Check which plan includes the export you need, since the Free plan is limited to two members, 2 GB and two projects. Other review tools will have a similar export or a copy button. If a client will not use a review tool at all, ask them to put timecodes in the email, because every NO TIMECODE row costs you a clarifying question.
The prompt that does the consolidating
You are helping a videographer consolidate client feedback on a 90-second
film (version 2). Below are numbered comments from three stakeholders:
Founder (F), Marketing (M) and Operations (O). Each has a source.
Task: produce one table with these columns:
ID | Timecode (exactly as written) | Requested change | Asked by (IDs) | Flag
Rules:
1. Copy timecodes exactly as written. If a comment has none, write
NO TIMECODE. Never estimate one.
2. Merge duplicates, but list every source ID.
3. If two comments contradict each other, do not choose. Make one row,
flag it DECISION and state both positions in the person's own words.
4. Do not add requests that nobody made.
5. If a comment gives no amount (a few seconds, a bit), say so and
write a question instead of a number.
6. After the table, list any comment you could not understand.
COMMENTS
F1 (email): ...Rule 3 is the one that keeps the assistant out of the creative decision. Rule 5 prevents the invented numbers. Rule 1 matters most of all, and you should check it independently: after the first run, open the review tool and compare five timecodes at random with the table. A single misread digit (00:14 for 00:41) sends the editor to the wrong scene, and it will not look wrong in a tidy table.
Handling conflicting feedback from three stakeholders
Conflicts come in four kinds, and it helps to know which you have. A direct contradiction asks for opposite things, such as a longer and a shorter testimonial. A stacked conflict has one person asking for something that another request would undo. An authority conflict pits taste against a constraint such as brand guidelines or a licence. And a vague conflict hides in words like "more punch" and "less busy", where the two people may agree and not know it. Expect the assistant to catch the first two reliably, and the last two only if you ask it to look for them.
Present conflicts at the top of the list in their own box: the two positions, what each would cost in time, and the deadline for a choice. Then send them to the single person who has final say. That person should be named in the contract or the kick-off email, and if nobody was named, that is the first thing to fix. A neutral message works better than a summary that quietly favours the loudest voice:
Hi [first name],
Three items in round one need a decision from you before we edit:
1. 00:41 testimonial. The founder would like it longer; marketing would
like it five seconds shorter. Either is a quick change. We can only
do one.
2. 01:12 logo. Marketing asks for a bigger logo; operations asks for it
to stay subtle per the brand guidelines.
3. Music. The founder would like a more upbeat track; operations notes
the current track is licensed and approved. A new track would need a
new licence, which we would quote separately.
Please reply with your choice for each by [date]. We will start the
edit when we have your reply to this message.
[your name]Notice what that message does not contain: a recommendation. If you have one, add it in your own words. The assistant drafted the neutral version in seconds, and you supplied the line about licensing because only you know how the track was sourced. For the general pattern behind this, ChatGPT for designers: turning client feedback into revision plans is the sibling for graphic work, and scope creep in AI projects covers the handling of change requests in general.
Checking each item against the contract's revision rounds
Paste only the relevant clause, not the whole contract. That keeps the assistant on task and keeps the rest of your agreement out of a chat window. The clause might read like this (illustrative):
The fee includes two rounds of consolidated revisions to the edit, graphics, colour and audio mix. A round is one written list from the Client's nominated approver. New footage, re-shoots, new scripts, additional deliverables and changes to elements approved at an earlier stage are extra and are charged at the hourly rate.
Add one instruction to the prompt: "Classify each row as IN SCOPE (edit, graphics, colour or audio mix on the current cut), POSSIBLE EXTRA (new footage, new deliverable, or reworking an approved element) or YOUR CALL (unclear). Quote the words of the clause you relied on. This is not legal advice." The result for the example:
- Row 5, spelling in the lower third: IN SCOPE. The change is to graphics on the current cut.
- Row 3, replace the product shot: POSSIBLE EXTRA. The clause says new footage is extra, and the replacement needs the updated label filmed.
- Row 8, the 15-second social cut: POSSIBLE EXTRA, as an additional deliverable.
- Row 6, "colour feels cold": YOUR CALL. Colour is in scope, but the request is too vague to cost until the client names a scene.
The classification is a first read, not a ruling. You decide whether to charge, discount or absorb each possible extra, and you may well absorb row 3 if the client is a good one. The words "POSSIBLE EXTRA" are chosen deliberately: the assistant has no way to know what was agreed on the phone. If you do charge, the sum is yours. At an illustrative $85 an hour, a three-hour pick-up shoot plus an hour of editing for row 3 is $340, and 2.5 hours for the social cut is $212.50, so $552.50 for the two. The assistant can lay that out in a clear quote paragraph; the numbers, and the choice, come from you. Video production pricing sheet: build a rate card with AI helps you set the hourly figure in the first place.
If a client disputes whether something is a revision at all, the wording of your contract decides that, and for a real disagreement a solicitor does. Spelling out the revision rules at the proposal stage prevents most of it; video production proposals with scope, budget and timeline shows how to draft that section.
The client approval step: what to send and what counts as a yes
The consolidated list is only useful if the client agrees it is the list. Send it to the nominated approver with the decision box on top, ask for a reply that says "Approved" or lists corrections, and make no changes until that reply arrives. Silence is not approval, so set a date, and say plainly that the approved list is round one. A note that works:
Hi [first name],
Attached is the consolidated list for round one of two on version 2
(16 items, three needing your decision, two that we believe are outside
the revision round and have priced separately).
Please reply "Approved" or send corrections by [date]. When we have your
reply we will deliver version 3 within [number] working days. Comments
sent after your approval will go into round two.
[your name]Keep that reply with the project files and name the next export so it carries the round, for example v3-r1. Then if anyone says "I asked for that months ago", there is a dated list and a reply that contradicts them. The assistant can update a running log after each round: round, item ID, request, status (done, declined, moved to extra), approved by and date. Three lines of it look like this:
| Round | Item | Request | Status | Approved by |
|---|---|---|---|---|
| 1 | 2 | Testimonial length at 00:41 | Done: client chose the shorter version | Approver, by email |
| 1 | 3 | Replace the product shot | Moved to extra, quoted | Approver, by email |
| 1 | 8 | 15-second social cut | Moved to extra, quoted | Approver, by email |
Once the list is approved, it also gives you a delivery estimate. Count the in-scope rows, give each a realistic number of minutes from your own history, and add export and quality-check time. The example has 11 in-scope rows, counting the colour note once the client has named a scene. At about 12 minutes each that is 132 minutes, plus 30 for the export and a full watch-through, so 162 minutes, just under three hours. Add the three decision rows once the client has chosen. A date promised from that sum is safer than "tomorrow" said before you have looked at the list. The assistant can total a column you give it, but check the total yourself, as with any arithmetic it does.
For tools that handle review, approvals and invoices in one place, videographer software: client review, approvals and invoices compared is the place to start. The principle matters more than the product: a written list, one approver, and a visible round count.
Feedback that arrives as calls, voice notes and screen recordings
Some clients will not write anything down. A call transcript or a voice note can go through the same process, with two cautions. Record or transcribe only with the client's agreement, and say so at the start. And expect the transcript to mishear product names and numbers, because "00:14" and "00:40" sound alike on a phone line. Ask the assistant to extract every request as a row, mark any time reference as UNCONFIRMED, and then confirm each against the cut yourself. Spoken feedback also leans on phrases such as "where the guy is talking", which only you can resolve to a timecode. Once you have, add the timecode in your own writing and send the list for approval as normal. Next time, ask for comments in the review tool, and explain that it speeds up their revisions.
A short transcript (illustrative) shows how it goes: "Yeah, so the bit where [first name] talks about the launch needs to come earlier. The lower third there has the wrong job title, I think it says Head of Sales and it should be Sales Director. Oh, and can we lose the shot of the car park?" Run through the same prompt, it becomes:
| Reference as spoken | Requested change | Flag |
|---|---|---|
| "The bit where [first name] talks about the launch" | Move it earlier in the edit | UNCONFIRMED timecode; you decide how much earlier |
| The lower third in that scene | Change the job title from Head of Sales to Sales Director | UNCONFIRMED timecode; get the correct title in writing first |
| "The shot of the car park" | Remove the shot | UNCONFIRMED; find which shot, and whether removing it breaks a cut |
The second row shows a rule worth keeping: a name or job title spoken on a call is not a source. Get it in writing before it goes on screen, because a wrong title on a lower third costs a re-export and a small embarrassment.
Setting the rules at kick-off so the list is shorter
Much of the mess in a feedback round is predictable in the first week. A short note in the kick-off email, drafted with the assistant from your contract, sets expectations while everyone is still friendly:
How feedback works on this project
1. [Name] is your approver. Their written reply is the only sign-off
we act on.
2. Please gather your team's comments into one list. Use the review link
and add a timecode to each comment.
3. Each round is one consolidated list. Your fee includes two rounds.
4. Anything outside the rounds, such as new footage or extra versions,
we will quote before starting.
5. Comments received after you approve a list go into the next round.The assistant is useful for matching that note to your contract's wording and keeping the tone plain. Check that every line still says what the contract says. Where they differ, the contract wins and the email changes.
Five ways a consolidated list goes wrong
| Failure | How it shows up | The check |
|---|---|---|
| Timecode slip | The editor opens the wrong scene; the client says "that is not what I meant" | Compare five random timecodes with the review tool before sending |
| Merge that loses nuance | "Bigger logo" and "keep subtle" collapse into "adjust logo size" | Read every merged row against its source IDs |
| Decision by assistant | A compromise appears in a DECISION row | Search the table for the words "compromise" and "suggest" |
| Old-version feedback | You re-do something already done | State the version on every batch |
| Out-of-scope work absorbed silently | Round two arrives with round one's extras still unbilled | Keep the change log and the POSSIBLE EXTRA flags visible |
A last point on effort. The whole routine, from gathering comments to sending the approval note, takes about 20 minutes for a 23-comment round once the habits are in place. The saving, though, is not mainly the half hour. It is the argument you do not have in week three. If you want a wider view of AI in a video business, how videographers use AI to edit, caption and quote faster covers the rest of the workflow.
Questions videographers ask about AI and client feedback
Can AI watch my video and check the changes were made?
Not reliably. ChatGPT can accept video files, but OpenAI says its analysis can be incomplete or inaccurate and may not cover the whole video or interpret its audio. Use the consolidated list as your checklist, and check each change yourself by watching the new cut at the listed timecode before it goes back to the client.
Is it safe to paste client emails into a chat assistant?
Client emails can carry names, product plans and details of unreleased work. Use a business plan, or switch off the model-training setting on a consumer plan. Replace names with roles such as Founder and Marketing, and leave out anything covered by a confidentiality agreement unless that agreement allows it.
What if a stakeholder sends new comments after the list is approved?
They belong to the next round, or to extra work, unless the approver amends the approved list in writing. Tell the approver that the new comment arrived after sign-off, log it with its date, and carry on with the approved list. Editing to a moving target is how a two-round contract becomes five rounds.
Can the assistant write the message about out-of-scope items?
It can draft a neutral message that quotes the clause and offers a price for each item. You decide the tone, the price and whether to waive an item as goodwill. Read the draft as if you were the client, and remove anything that sounds like a threat or a legal conclusion.
Further reads
- How to Price Videography Services: Day Rate, Project Fee or Package — Set day rates, project fees and packages before disputes start.
- Best AI Filmmaking Tools by Job for a Small Production Business — Filmmaking tools sorted by the job they do.
- Best AI for Filmmakers Selling Video Services to Local Businesses — A tool-by-stage guide for video producers with local clients: scripts, shot lists, rough cuts, monthly costs and the client-work checks that creator lists skip.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Frame.io pricing page and OpenAI's ChatGPT image inputs FAQ (video attachments), read in October 2026. Contract wording, comments, timecodes and prices in the examples are illustrations.