“Cut here” is incomplete when everyone is watching a different video
In this guide’s fictional example, one note asks to “remove the pause at 00:32” in V1 and another says “keep it at 00:32” in V2. Sorting them by time does not settle the job: we do not yet know whether they refer to the same shot. Before changing the edit, preserve each note’s version and request its location in the video being reviewed.
This guide proposes a text-only AI trial to prepare a review worksheet: what each person requests, which document supports it and what the editor needs to clarify. It includes invented inputs, a manual reference and a prompt. The trial has not been run or measured. The assistant receives no video: it cannot see shots, decide cuts or approve delivery.
First decide: do you need a review tool or better context?
Frame.io V4 documents timestamped comments, range-based comments and general comments without a timestamp. Its panel also supports comment export; the documentation warns that a filtered export may contain only the filtered results. Wipster describes comments associated with video, sorting by timecode and CSV or PDF export. These are provider descriptions, rather than operations tested here.
If you need clients to point to a moment while watching, investigate existing review tools first. This article’s AI proposal addresses a different task: comparing notes you already have with their version and instructions, then drafting questions about incomplete context. You can trial the fictional inputs without subscribing to a new platform.
Before building a list, record which export you are reviewing, whether filters were applied and which sources are missing. Do not label an extract “all comments”. Keep replies that clarify a request alongside the original note, with their own provenance.
The minimum record for a note you can find again
I propose these working fields: note ID, fictional or authorised author, source document, declared video file or version, exact text, literal time reference, declared reference type, request, question, responsible person’s decision and status. If a source lacks a field, write “not stated”. This proposal does not assume every application exports these fields.
Do not automatically turn “00:32” into four-part timecode. In this trial, only one note specifies minutes:seconds from the start of its export. Preserve the others as written and ask. Adobe documents that Premiere can show a clip’s source timecode even when its sequence has a different timebase. Record where a reference comes from before comparing or converting it.
Use only the fictional case below for the trial. Do not paste private review links, client videos, names, email addresses or unpublished comments without permission and appropriate confidentiality and processing conditions. The cover is a generated illustration, rather than a real commission.
Fictional case: two versions, a general note and an untraceable reference
All files, people, times and messages in PRUEBA-VIDEO are invented by the editor. No associated video exists. D1 · Working instruction: “Prepare questions for reviewing anuncio_V2.mp4. Do not apply changes, transfer markers between versions or approve delivery. Editorial decisions belong to the video editor.” D2 · Register: “anuncio_V1.mp4 and anuncio_V2.mp4 are different exports; no shot correspondence has been documented.”
N1 · Sheet A, person A: “anuncio_V1.mp4; 00:32; remove the pause.” Time units and origin are not stated. N2 · Sheet B, person B: “anuncio_V2.mp4; 00:32, minutes:seconds from the start of this export; keep the pause.” N3 · Sheet B, person B: “anuncio_V2.mp4; general comment without timestamp: the logo must remain in every closing section.”
N4 · Message C, person C: “00:01:12:08; replace this shot with the previous one.” It does not identify the version, timecode origin, timebase or which shot “the previous one” means. D1 does not authorise guessing. The worksheet below is a hand-written reference, rather than AI output.
Reference worksheet: useful questions before opening the timeline
These are proposed internal questions, rather than sent messages. No row establishes that a video has been watched or changed.
| Note | What the worksheet preserves | Missing context and question | Working status |
|---|---|---|---|
| N1 | V1; “00:32”; “remove the pause”; Sheet A | Which time reference was used? Where is that moment in V2, if it still exists? | V1 reference awaiting location in V2; unapplied |
| N2 | V2; “00:32”; minutes:seconds from start; “keep the pause”; Sheet B | Editor to review the request in V2. Do not declare conflict with N1 before locating both references. | Documented V2 request; editorial decision pending |
| N3 | V2; general comment without timestamp; “the logo must remain in every closing section” | Which closing sections are included? Review scope in V2 without inventing a timestamp. | General note retained; not reduced to one point |
| N4 | “00:01:12:08”; “replace this shot with the previous one”; Message C | Which version and reference were used? Which specific shot is requested? Confirm timebase if interpreting the code. | Location and replacement pending; no change authorised |
A prompt that prepares review without pretending to see shots
Copy D1, D2 and N1–N4 in full after this instruction. Save the response separately from the original sources.
“Work only with PRUEBA-VIDEO. Prepare one row per N1–N4 showing source document, declared version, literal time reference, declared reference type, exact request quotation, missing data, internal question and pending status. Cite the document supporting each field. Do not convert times or transfer markers between V1 and V2. Do not infer that equal times identify the same shot. Keep N3 as a general comment. Do not invent shots, timebases, approvals or completed edits. State which context is missing. Do not edit video or produce an approved delivery.”
Claude’s documentation recommends allowing the model to acknowledge uncertainty and requesting citations to check its claims. These inform the prompt, rather than guaranteeing that any assistant will comply. No particular tool has been tested here.
Three mistakes to check in the response
Check that all four notes appear. Reject assigning N1 to V2, merging N1 and N2 into a contradiction already located in the footage, or turning N3 into a comment at 00:00. Ensure “every closing section” remains scope to clarify, and that N4 receives no invented version or shot.
Try a separate variant: add D3, “The editor has located N1 in V2 at 00:35, minutes:seconds from the start; she has not decided the change.” The new worksheet should preserve N1’s original reference and add the location attributed to D3. It must not delete N2 or state that a cut has been applied. D3 resolves only N1’s location in this variant; it does not establish a relationship with N2’s pause.
The responsible person must open the right version, locate references and resolve editorial choices with authorised people. A tidy list establishes neither continuity, sound quality, usage rights, a correct edit nor client acceptance.
A pilot that measures the whole review
Prepare several fictional notes with a manual reference written first: missing version, equal times in different versions, a general comment and a later clarification. Compare the manual worksheet with the assistant’s proposal. Record omitted notes, invented data, unsupported reference transfers, inaccurate quotations and already resolved questions that keep recurring.
Measure source preparation, checking and correction time against the same manual task. Decide in advance which errors invalidate a worksheet, such as inventing authorisation. Keep a decision log and do not automate changes during the pilot. These are proposed measures; no savings, accuracy or customer results have been demonstrated.
When to investigate a custom integration
The manual trial means copying fictional text and checking a worksheet. Connecting automation would require investigating actual formats and permissions, preserving versions and export filters, controlling duplicates by identifier and recording who resolves each note. Each connection would need validation, with proposal, decision and action kept separate. Compatibility with Premiere, Frame.io, Wipster or any other editor or platform has not been verified.
Adaris develops custom software, CRM systems and AI automations case by case. A review inbox tailored to your workflow could be scoped with agreed acceptance criteria. This is a development proposal, rather than an existing proprietary video editing tool. Commercial contact is handled in Spanish.
Related services
Sources
Do notes get lost between versions?
Tell us how you receive feedback and who decides changes. We can assess a scoped development for your workflow. Commercial support is in Spanish.