A Freelance AI Deliverable Provenance Log: What to Record Before Handoff

earnkitlab 1

AI Tools

A Freelance AI Deliverable Provenance Log: What to Record Before Handoff

By Earn Kit Lab TeamUpdated November 20269 min read

Freelancer and client reviewing an AI-assisted delivery checklist

An AI deliverable provenance log is a small record of how AI-assisted client work was made, checked and handed over. It is not a confession, a legal guarantee or a dump of private prompts. It is a practical way to answer the questions that arrive after delivery: which file is final, what was machine-assisted, what did a person verify, and which source or approval supports it?

For a solo designer, writer or developer, the useful version can fit in one row per deliverable. The goal is not to document every keystroke. Record just enough to make the work reproducible, reviewable and safe to hand off.

☰ Table of Contents

Why client handoffs need provenance

AI-assisted work often passes through several versions: a client brief, a model output, a human edit, a fact check and an exported file. If the chain is invisible, ordinary production questions can sound like accusations. A provenance log makes the chain legible without implying that AI use is automatically a problem.

It also keeps three separate claims from getting blurred: what tool was used, what the tool produced, and what the freelancer personally verified. That distinction matters more than a broad label such as “AI-generated.” For an existing overview of comparing tools on evidence rather than marketing, see our source-first AI vendor comparison workflow.

The six fields worth recording

Capture a stable deliverable name and version, the tool and model (if known), the task it assisted with, the source material supplied, the human review performed, and the final approval or handoff location. A date helps when a vendor changes its model or retention settings. Avoid copying sensitive client material into the log itself.

Field Record this Keep out
Deliverable Filename, version, date and client project code Unnecessary personal or confidential details
AI assistance Tool/vendor, model/version if shown, and the specific task “AI made it” with no context
Inputs Type of source used, permission/status, and a secure reference Raw prompts, credentials, private records
Human checks What was checked, by whom, and the result A blanket “reviewed” with no scope
Approval Client sign-off link or timestamp and final file location Unapproved claims that imply client endorsement

A freelancer recording tool, version, source and review status in a project log

A practical provenance log template

Use a spreadsheet, project tracker or a short handoff note. One row per final asset is usually enough. Keep the log in the same access-controlled project space as the deliverable, not in a public portfolio or shared prompt archive.

Deliverable / version AI-assisted step Human verification Client-facing note
Landing-page copy v3 Outline alternatives; tool/model and date recorded Checked every product claim against client source; edited voice and structure “AI assisted with early outline options; final copy reviewed and edited by me.”
Campaign illustration final Concept exploration only; final asset produced/edited in named design tools Checked brief fit, text, visual artifacts and rights/asset provenance State the workflow accurately if requested or required by the brief
CSV cleanup script v2 Generated a first-pass transformation suggestion Ran on a synthetic sample, compared row counts and edge cases, reviewed diff Link to versioned code and tests; do not imply the model validated the data
Quick rule: record evidence of the check, not just the fact that a check happened. “Verified 12 cited prices against the approved source sheet” is more useful than “fact-checked.”

What to record for images, text and code

Text: note whether AI generated an outline, draft, rewrite or translation. Keep the human edit and fact-check scope clear. If the work contains citations, link each important factual claim to the source actually checked.

Images: note whether the asset is generated, edited or a composite, plus the source of any supplied reference material and any client-approved restrictions. Do not describe a synthetic scene as a documentary photograph.

Code or data: identify the suggested change, repository commit or file version, test environment, and test results. Never put secrets or production customer data in a prompt just to make the log look complete. Our client-data privacy checklist covers the separate decision about what can safely be shared with a tool.

A light review process that catches errors

Before handoff, open the final exported file rather than trusting the source document preview. Compare it with the brief, check factual claims against client-approved sources, test links and formulas, and inspect for invented details or inconsistent names. For code, run the relevant tests and inspect the diff. For images, check text, hands, logos, licensing and whether the scene could mislead an audience.

Then mark the row complete with a link to the evidence: a test report, annotated proof, source list or client approval. This is a quality-control record, not a claim that no error is possible. If a check was not done, leave it visible as not checked instead of implying assurance.

Creative team comparing the delivered file with an approved client brief

What to tell the client, and when

Follow the contract, brief, platform rules and any applicable disclosure requirement. If the client asks whether AI was used, answer specifically and accurately: name the kind of assistance, describe the human review, and identify any remaining limitation. Do not promise exclusivity, factual accuracy or rights clearance merely because a person reviewed the output.

If disclosure is relevant to the audience or the nature of the work, agree on wording and placement before delivery. A short, plain note is usually clearer than burying a broad disclaimer in a footer. Keep the decision and approval in the project record.

Keep the log useful, not bureaucratic

Use a threshold: record AI involvement when it materially shaped the deliverable, introduced a factual or rights-sensitive claim, or affects how the client will use the asset. For trivial spelling suggestions, a per-keystroke record adds noise. Reuse a template, restrict access, set a retention period with the client, and delete sensitive working material according to the agreed process.

A good provenance log does not make AI work risk-free. It makes the workflow understandable, the checks visible and the handoff easier to defend. Start with one deliverable and six fields; add complexity only when the project needs it.

Frequently asked questions

Does every AI-assisted deliverable need a provenance log?

No. Use a proportional record for work where AI materially shaped the result, where the client asks for traceability, or where factual, privacy or rights issues make the production path important.

Should I save the full prompt?

Usually not by default. Prompts can contain confidential details, and a prompt dump is not proof that output is correct. Record the task and safe source reference instead, unless the contract specifically requires a prompt record and the storage is approved.

Can I say a human reviewed the entire deliverable?

Only if that is true. State what was checked, such as claims, calculations, links or visual artifacts. A focused review does not establish that every possible error was eliminated.

What if a client bans AI use?

Do not use it on the restricted work. Clarify whether the ban covers all tools or only specified tasks, and keep the approved workflow documented before processing client material.

Should I disclose AI use to the client’s audience?

It depends on the contract, audience expectations, platform rules, the content and applicable law. Agree on wording with the client rather than assuming either that disclosure is always required or never required.

How long should I keep the record?

Follow the contract and your documented retention policy. Keep only what is useful for the project and remove sensitive source material when it is no longer needed.

E
Earn Kit Lab Team

Practical workflows for freelancers and small teams, built around clear checks and evidence.

Leave a Comment

Your email address will not be published. Required fields are marked *