Skip to content

AI production records for AI producers

You sit between the people who generate and the people who sign. When a client, a lawyer, or a platform asks how an asset was made, the question lands on your desk, and your answer has to hold months after the project closed.

Published July 22, 2026.

The AI producer role exists because someone has to be accountable for how generative work is made, not only for whether it looks right. Clients now write that accountability into contracts, questionnaires, and delivery requirements, and the producer is the named person behind it. You did not run most of the generations yourself, yet you are the one who has to account for all of them.

The questions arrive on a schedule: procurement questionnaires before the contract, delivery forms at handoff, and legal follow-ups months later, usually about one shot, asking for records rather than assurances. A producer who has to reconstruct the answer from chat threads and personal tool accounts pays for that reconstruction in days of work and in credibility.

The structural problem is that generation happens in many places at once. Team members and freelancers work in their own accounts across several tools, and each tool keeps whatever it keeps for as long as it keeps it. If the record only exists inside those accounts, the producer is accountable for evidence that other people control. The fix is not more discipline after the fact, it is capture at the moment of generation into one system the production owns.

It also matters that you promise the right things. A production-side record can show which prompts, settings, references, and outputs existed, that they have not changed since, and when they were recorded. It cannot show that a specific reference caused a specific output, and it cannot show what a vendor trained a model on. The guide on what provenance can and cannot prove maps that boundary before a client call tests it.

Where documentation breaks down

  • Generation happens in team members' and freelancers' personal tool accounts, so the producer is accountable for records they never see.
  • Client questionnaires arrive before the project starts and ask about a documentation process the team has not agreed on yet.
  • Delivery paperwork asks questions per asset, while the team's notes, where they exist at all, are kept per project.
  • Legal review happens weeks or months after generation, when tool-side histories may be gone and the freelancer who ran the job has moved on.
  • Revisions overwrite their own trail, so nobody can say afterwards whether the final was an edit of the approved draft or a fresh generation.
  • Producers get asked to warrant things no record can support, such as what a model was trained on.

A documentation routine that holds up

  1. At kickoff, write down which tools and models the production will use, and obtain the client's written acknowledgment where the contract asks for it.
  2. Require everyone who generates to capture the prompt, the model and version, the settings, and any references at the moment of generation, in one shared system rather than personal notes.
  3. Review the record against the shot list once a week and chase gaps while the people involved still remember the session.
  4. When an asset is revised, record whether it was edited or regenerated, and keep the chain from first draft to final intact.
  5. At delivery, export the per-asset record alongside the files, so the answers travel with the work.
  6. Archive the full record with the project, because the hardest questions arrive after the invoice is paid.

Frequently asked questions

A client's legal team asks which model and version generated an asset we delivered four months ago. What should I be able to produce?

The per-generation record: prompt, model and version, settings, references, the date, and the delivered file it belongs to. If your only copy lives in a tool's own history, check the vendor's retention terms before you rely on it.

Can I certify that our outputs were not trained on a competitor's work?

No, and you should not sign language that implies you can. No production-side record can show what a vendor trained a model on. What you can warrant is your own process: which tools you used, what you put in, and what you changed.

Do I need records for pitch work that was never delivered?

Check the contract, but keep at least the tool, the date, and the references for anything shown to a client. Pitch material has a habit of becoming production material, and by then the pitch-week sessions are unrecoverable.

What belongs in the delivery package itself?

A per-asset summary: which parts are AI-generated, which tool and model, what humans changed, and who signed it off. The full generation record stays in your archive, ready for the follow-up questions the summary will provoke.

Our freelancers generate in their own accounts. How do I make records happen?

Make the record part of the deliverable in their contract, and give them one shared place to capture into at generation time. A producer cannot reconstruct someone else's account activity afterwards, so the record has to happen during the work.

Related guides

  • How to document AI-generated work for client delivery The delivery-side requirements a producer answers to, and the per-shot checklist that covers them.
  • Prompt and model documentation: a practical checklist for studios The exact fields to require from everyone who generates, so the producer's record is complete on day one.
  • AI provenance: what you can prove and what you cannot The boundary between defensible answers and overpromises, before a client call tests it.
  • EU AI Act Article 50 for creative teams What the transparency obligations actually say, for the producer who is asked whether the team is ready.
  • What is an AI production log? An AI production log is the per-asset record of prompts, models, settings, references, edits, rights decisions, and approvals behind AI-assisted work. Here is the minimum useful structure and the limits of what that record can prove.
  • C2PA vs. AI workflow documentation C2PA and AI workflow documentation solve different parts of the provenance problem. C2PA protects signed assertions attached to media; workflow records preserve prompts, sources, decisions, rights, versions, and approvals around the file.
  • AI shot log template for image and video production A practical AI shot log template for prompts, models, references, versions, human edits, disclosure, and approval. Download the CSV and adapt the field guide to your production workflow.
  • How to record client approval for AI-generated work A client approval record for AI-generated work should identify the exact version, disclosed AI use, review scope, requested changes, approver, and time. This guide turns sign-off into evidence instead of an ambiguous email.
  • Guide editorial and verification policy How Behind The Workflow researches, sources, dates, updates, and corrects its AI production guides. Product facts use primary sources, legal claims link to official text, and machine-readable markup mirrors visible content.
  • AI production records for creative directors Creative directors approve what ships, so questions about AI in the work come to them, in client meetings and in review rooms. This guide covers the records that let them answer precisely instead of vaguely.
  • AI production records for post-production teams In post, AI arrives as a hundred small fixes inside shots rather than one big generation, which is exactly what makes it hard to declare on a delivery form later. This guide covers records that map to shots, versions, and conforms.
  • AI production records for freelance AI artists Freelance AI artists warrant their own process in every contract they sign, with no legal department behind them. This guide covers the records that back those warranties, survive tool subscriptions, and win repeat work.
  • AI production records for agencies Agencies carry contractual responsibility for AI use across every team, freelancer, and production partner on an account. This guide covers records that hold at agency scale, where the person who generated and the person who answers are never the same.

Every requirement in this guide is answered from a record kept while the work happens. BTW keeps that record for you.

Start your own record