Fahad Manzur
The advice circulating in freelancer groups is that Fiverr and Upwork both require you to tell clients when you use AI. Only half of that is true. Upwork’s help centre tells freelancers to disclose AI use to clients in every project. Fiverr’s published guideline puts the obligation the other way round: you must honour a client’s request for non-AI work when they make it, and you are encouraged to ask, but you are not required to volunteer your toolchain unprompted. Knowing which rule you are working under decides how you write a gig, how you answer a brief, and what happens if a dispute lands.
Both policies are live on the platforms’ own help centres as of this week, and both were written for exactly the situation a working freelancer in Lahore or Karachi is in: a ChatGPT draft, a Midjourney concept, a Claude Code scaffold, then hours of your own refinement before delivery.
What Fiverr’s guideline actually says
Fiverr’s Using AI on Fiverr page, filed under Trust and Safety policies, opens by permitting AI in every service category. The conditions attach to the output rather than the tool: work has to be original, meaningfully refined and customised to the client’s brief, and Fiverr states plainly that generic, unmodified or reused AI output falls short of its quality bar.
The disclosure rule is conditional. Freelancers have to honour a client’s explicit request for non-AI work, and when a client states that preference before or at the start of an order, the freelancer is expected to disclose the intended workflow and adjust to it. Elsewhere on the same page the burden on clients is spelled out: if they do not want AI used, they need to say so before ordering or immediately after, because sellers are not assumed to list every tool in their gig.
So on Fiverr the trigger is the client’s stated preference. No stated preference, no disclosure duty, although the page does encourage sellers to ask about preferences during the order discussion and to stay transparent when AI use could change the creative direction or the style of the output.
What Upwork’s guideline actually says
Upwork’s Ethics of AI on Upwork article, mirrored in its AI at work lesson, lists three guidelines. The first is headed Transparency and reads: “Always disclose to clients whether you use AI-generated content or tools during a project.” The second frames AI as a collaborator that supports, rather than substitutes for, the freelancer’s expertise. The third is about data bias. The page also reminds freelancers that proposals and job content must be original and comply with the terms of whatever AI platform produced them.
Upwork’s rule is unconditional in wording. It does not depend on the client asking. Whether Upwork actively enforces it on every contract is a different question, and the help article is a guideline rather than a Terms of Service clause, but it is the standard a client or a dispute reviewer can point to.
The two rules side by side
On Fiverr, AI is permitted in every category; disclosure is required when the client asks for non-AI work or states a preference; the client is expected to raise the subject early; and a dispute is judged on the whole order, with the client’s clarity about expectations as the first factor examined.
On Upwork, AI is permitted; disclosure is expected always, regardless of whether the client raised it; originality of proposal text is a stated requirement; and the platform frames non-disclosure as a trust issue rather than a quality issue.
The practical consequence: a seller who quietly uses AI on Fiverr and delivers polished, customised work is inside the rules unless the client had asked otherwise. The same behaviour on Upwork is outside the written guideline even if the client never complains.
How Fiverr handles a complaint after delivery
The dispute section is where Fiverr shows what it weighs. When a client raises AI concerns after delivery, Fiverr says it looks at the whole order rather than the AI question in isolation. Four factors are listed: whether the client clearly communicated expectations about AI before or at the start; whether the delivery is accurate, free of hallucinated content and tailored rather than templated; whether it reflects the freelancer’s own professional effort and meaningful input rather than raw output; and whether both sides communicated professionally.
The penalty if Fiverr finds a trust breach is severe: the order may be cancelled, the client refunded in full, and the freelancer’s account permanently suspended. What counts as a trust breach is defined narrowly: deceiving the client about how the work was produced, misrepresenting AI usage, or failing the quality standard. Under this wording, staying quiet about your tools does not by itself amount to deception, whereas advertising hand-written work in the gig title and then delivering raw model output does.
Where the real exposure sits: your gig copy
Because Fiverr’s rule keys off what the client was led to expect, the words in your gig description carry more weight than any private workflow. A gig that promises “100% human-written articles” or “hand-illustrated logos” has created an explicit expectation, and the client does not need to send a separate “no AI” message for it to bind you. If you use AI at any stage, remove those claims. Replace them with a truthful description of the outcome: research-backed articles, brand-matched illustrations, edited and fact-checked copy.
Proposals are the Upwork trap
Upwork’s guideline is unusual in that it mentions proposals specifically. AI-assisted proposal text is allowed provided it is original, so a template pasted into fifty job posts fails the guideline on its face, whatever tool produced it. If you draft with AI, the safe practice is to write the first paragraph about the client’s actual job post yourself and let AI help with the structure of the rest. Our guide to Upwork proposals that get replies is built around that ordering for a reason.
There is a second Upwork layer that many freelancers have not connected to disclosure. Since January 2026 Upwork’s terms allow it to use work product and messages to train AI unless either party opts out, a change we covered in Upwork’s AI training opt-out. A client who cares about AI in your workflow usually cares about that setting too, and mentioning both in the same conversation reads as competence.
A disclosure line that works on both platforms
Rather than maintaining two policies, adopt Upwork’s stricter standard everywhere. It costs one sentence in your order requirements or kickoff message:
“I use AI tools for research, first drafts and code scaffolding, and every deliverable is reviewed, edited and fact-checked by me before delivery. If you need a fully AI-free process, tell me before we start and I will quote for it separately.”
That sentence does three things. It satisfies Upwork’s always-disclose guideline. It triggers Fiverr’s “client states preference at the start” condition on your terms, so a later complaint has to be judged against a workflow you disclosed in writing. And it prices the no-AI option, which is slower, instead of absorbing that cost silently.
Put the same commitment into your contract for direct clients as well. The AI clause most freelance contracts are missing covers the ownership and confidentiality side that platform policies leave out, such as whether client data can be pasted into a third-party model at all.
Frequently asked questions
Do I have to label AI-generated images in my Fiverr gig gallery?
The AI usage guideline does not require a label on gallery images. The Community Standards do prohibit images that misrepresent a real person or fake an affiliation, and your gallery must show work you can actually deliver. If the gallery shows hand-drawn samples and you deliver AI renders, that is a misrepresentation regardless of labels.
Can a client cancel an order on Fiverr just because I used AI?
Not on that ground alone. Fiverr states it reviews the full context, starting with whether the client communicated a no-AI expectation before or at the start of the order. If they did not, and the work is customised and accurate, AI use by itself is not a breach.
Is there an AI detector running on Upwork or Fiverr deliveries?
Neither help centre article describes automated AI detection on deliverables. Both put the weight on originality and customisation, and Fiverr describes a human review of the full order when a dispute is raised. Plan for your work to be judged by a person reading it, not a detector scoring it.
