Why a chat message is not disclosure you can defend
A line in a thread, "by the way, I used AI for parts of this", detaches from the work the moment it is sent. It gets buried, it can be screenshotted out of context, and it says nothing checkable: which parts, which AI, what was verified? When the dispute opens, the platform asks for documented evidence, and a vague message is the weakest form of it. Evidence created at delivery time, specific enough for the client to check without you, outweighs any explanation written after the complaint.
Disclosure with the record attached
Hand the work over through Zambo, and the delivery itself carries the disclosure:
- Every step of the job, each with a verifiable receipt the client can open and re-check in a browser, no account needed.
- The AI named for each step, as the producer stated it. Producer-stated is the honest standard: the record carries what the producer declared, and says so.
- The not_checked list: the steps that did not complete or were not checked, printed on the delivery, not discovered later.
- The client's own answer: accept or flag, recorded with a receipt of its own, so the disclosure and the response live in one record.
One more guard, for the clients who bar AI outright in places. When a handover is created with client_prefs.no_ai_for naming areas (say, pricing), Zambo checks the steps against it at creation: the creator is warned in the create result itself ("Client asked for no AI here: pricing"), and on the delivery the matching step carries that same line in its header, before the client opens the step. A barred area cannot slip through as a labeling accident; the record flags it at both ends.
The mechanics are documented at zambo.dev/docs/handover. Ask your AI to hand the work over to the client when the job is done, and the page exists.
The statement writes itself now
Since the v5 publish, you no longer draft the statement. Every handover page carries a Copy AI-use statement button, and the handover tool result carries the same statement, generated from the receipts' own AI labels, exactly as recorded. A delivery where two AIs split the work reads like the real one generated during acceptance testing: "AI use: Claude helped with 2 of 9 steps. Grok helped with 2 of 9 steps. See the listed receipts." Where the producer labeled nothing, the statement invents nothing: "AI use: Unspecified AI helped with 4 of 4 steps. See the listed receipts." The generated statement is short, under 280 characters plus the link, so it drops into a delivery message, a bio, or a dispute thread as-is.
Writing your own long-form version? It should read like this
"AI use on this delivery: this work was produced with AI assistance and delivered through Zambo. The full record is at [your handover link]. It shows every step of the job, the tool that ran, the AI named for each step as I stated it, and the steps that were not checked. You can verify each step in your browser on that page, and accept or flag the delivery there. Your answer is recorded with a receipt of its own."
Every clause in that long form is a field the record actually has: steps, tools, producer-stated AI labels, the not_checked list, and the recorded answer. Send it where the client will actually read it, in the platform's delivery message or alongside the files. If the platform has its own AI disclosure toggle, use that as well. The toggle is the platform's checkbox; this statement is the evidence layer underneath it.
What the record says when the producer says nothing
The honest unknown
Open the live public handover bd7e5325 and look at its steps. The producer stated no AI for them, so the record labels those steps "Unspecified AI" rather than inventing a name, and the generated AI-use statement uses the same label. That behavior is the point of a disclosure record: it distinguishes "the producer stated this AI" from "not stated", in the open. Name your AI when you hand work over, and the record carries your statement per step. Stay silent, and the record carries that too, which is exactly why clients can trust the labels that are there. The same delivery lists two steps in its not_checked list that did not complete, because a disclosure that only lists the good parts is marketing, not disclosure.
What this does not do
It does not certify the quality of the work, and a verifiable receipt shows the saved result has not changed since it was recorded; it does not prove the result was correct. It does not override a platform's rules or a contract that bars AI work: if a client ordered human-only work, a receipt does not launder AI work into compliance with that order. It does not disclose for you inside the platform; you paste the statement. And it does not decide disputes. What it does is narrower and, in a dispute, decisive-shaped: it makes the disclosure specific, fixes it at delivery time, and lets the client check every claim in it without trusting you. Arguments about what was used and what was said stop being arguments.
Questions, answered
How do I disclose AI use to a client without starting a dispute?
Disclose at delivery, specifically, with the record attached. Send the handover link with a short statement: the work was AI-assisted, the record shows every step, the AI named for each step as you stated it, and the steps that were not checked, and the client can verify it all in a browser and accept or flag it there. Clients dispute surprises and vagueness. A disclosure that arrives with checkable evidence, before anyone has a reason to argue, reads as professionalism, because that is what it is.
A client says I hid AI use in my freelance work, what evidence settles it?
Evidence created at delivery time. Platform disputes run on documented evidence, not opinions, and a handover is a timestamped delivery record: the steps, the receipts, the AI labels as the producer stated them, and the unchecked list, all fixed when the work shipped. If your delivery message carried the disclosure statement and the link, the record shows the client received the means to see exactly what an AI did, and their accept or flag is recorded with a receipt of its own. That does not decide the outcome; platforms decide by their own rules. It gives you the strongest form of the one thing they ask for.
How do I show a client which parts of the work an AI did?
On the handover, step by step. Each step names the tool that ran and the AI the producer stated for it, and each step carries a verifiable receipt the client can open. Where the producer stated no AI, the record labels the step 'Unspecified AI' instead of guessing, and the AI-use statement Zambo generates from the receipts uses that same label. State your AI when you produce the work and the delivery carries your statement per step; the client never has to rely on a summary you wrote afterward.
What should I send a client when I deliver AI-assisted work?
Two things: the work, and the handover link with the disclosure statement. The statement is generated for you: the handover page has a Copy AI-use statement button, and the handover tool result carries the same text, built from the receipts' AI labels exactly as recorded, with 'Unspecified AI' wherever the producer labeled nothing. Paste it into the platform delivery message, and use the platform's own AI disclosure toggle as well where one exists. The link does the heavy lifting, because the client can check every claim in it without an account and without trusting you.
Does disclosing with a record protect me from refunds or suspension?
No record can promise that. Fiverr's guidelines make misrepresentation the offence: cancellation, a full refund, and suspension attach to misrepresenting AI use, and a truthful, evidenced disclosure is the opposite of misrepresentation. But platforms judge their own cases, contracts can bar AI work outright, and a receipt does not prove the work is good. What disclosure with a record does is remove the factual argument, what was used and what was said, from the dispute, and replace it with a record both sides can open.