TL;DR
On this page
The fix for the AI question in a client contract is a clause, not a confession. An AI clause is the contract language that settles, in advance, how AI fits the work: whether you use it, how client data is protected, who owns AI-assisted deliverables, and where your liability stops. Put it in your own contract and the gray area becomes a normal term you control.
This walkthrough is part of the complete guide to freelancing in the AI era.
The question usually arrives one of two ways. A client's contract bans AI tools on the project. Or the client emails to ask, point blank, whether you use AI. The instinct is to freeze, because a normal tool suddenly feels like something to hide. It is not. It is a term to write down.
Why this is worth getting in writing
Most people handle AI by saying nothing and hoping it does not come up. The data suggests that is the higher-risk path, not the lower one. In a survey of 157 freelance writers by coach Ed Gandia, 49.7% had not told their clients they use AI. Yet among those who did disclose, reactions skewed positive or neutral over negative by roughly four to one.
The fear is overpriced. Half stay silent, while the ones who speak up mostly get a shrug or a nod. Silence does not protect you, it just leaves the question unanswered until a dispute forces it open. A short clause answers it on your terms, while everyone is still friendly.
There is a second reason to write it down, and it has teeth in some markets. Under the EU AI Act, general-purpose AI transparency rules took effect in August 2025, with full enforcement arriving in August 2026, and violations can draw fines of up to €35 million or 7% of global turnover. The Act notes that AI-generated content may need to be labelled as such depending on the context. If you serve EU clients and sell AI-assisted content, transparency stops being optional and becomes a compliance item your contract should reflect.
The four components of a complete AI clause
Whatever prompted the conversation, a good AI clause covers the same four areas. Think of them as the parts you add to your standard contract, not a separate document.
A useful taxonomy comes from Tascon Legal's practical guide, which organizes AI contract terms into disclosure and consent, data use and training limits, IP ownership, accuracy safeguards, security and liability, and exit procedures. Collapsed to what a solo operator actually needs, that is four working parts.
| Component | What it settles | Sample language to adapt |
|---|---|---|
| Disclosure and consent | Whether and how you use AI | "Supplier may use AI tools to assist in producing Deliverables and will review all output before delivery." |
| Data handling | Whether client data touches a model | "Supplier will not enter Client confidential data into public AI tools, and will not allow Client Data to train any model without prior written approval." |
| IP ownership | Who owns the AI-assisted result | "Supplier assigns all rights it holds in the Deliverables, including AI-assisted components, to Client on full payment." |
| Liability scope | Where your responsibility ends | "Supplier is responsible for reviewing Deliverables and for its own conduct, not for the underlying training data of third-party AI models." |
Tascon's own sample language for the data-training point is blunt and worth borrowing nearly verbatim:
Supplier will not train or fine tune models on Client Data without Client's prior written approval.
Source: Tascon Legal, "AI Clauses in Contracts: The Practical Guide"
That single sentence covers the concern behind most client anxiety about AI, which is rarely the tool itself and almost always the data.
The data-handling part is the one clients actually care about
When a client worries about AI, they are usually picturing their confidential brief, customer list, or unreleased product pasted into a chatbot that learns from it. That fear is well founded, because the default settings are not on your side. As strategist Sarah Noel Block notes, the consumer plans most people use do not protect inputs automatically:
If you didn't actively opt out, your conversations may already be included.
Source: Sarah Noel Block, Brand Strategy Sarah
On personal paid plans like ChatGPT Plus, training on your inputs is on until you manually disable it; only business or team plans exclude it contractually by default. The safe, contract-friendly position is simpler than tracking every provider's settings: keep confidential client data out of consumer AI tools entirely, and say so in the clause. If you need help drawing that line in practice, the deeper walkthrough is in is it safe to paste client info into ChatGPT.
The IP wrinkle worth naming
Ownership is where AI genuinely changes the contract, if only slightly. The US Copyright Office has stated that work generated solely by AI cannot be copyrighted by a human, because human authorship is a requirement, a position reinforced when copyright was denied in the Thaler case for an image made without human authorship.
In practice this rarely bites you, because your deliverables are not solely AI-generated. They are your prompts, selection, editing, and arrangement, and that human contribution is protectable. The clean move is to keep your normal IP-transfer-on-payment clause and add a line confirming you assign whatever rights you hold, AI-assisted parts included, so the client is never left guessing. The broader mechanics of assignment, work-for-hire, and transfer-on-payment are in IP ownership clauses.
What to do when the client raises it first
Half the time you will not be adding a clause to a blank contract; you will be reacting to one the client wrote. There are three versions, and each has a clean response.
The client requires disclosure. This is the easy one. A disclosure requirement asks you to state that you use AI and, sometimes, that you review the output. You almost certainly do both already. Agree, and make sure your own clause language matches what you actually do, so you are never out of step with a term you signed.
The client bans AI outright. Read the clause literally before you react, because "no AI" means wildly different things. A strict reading bans any tool that ever touched the work, which would also rule out spell-check and most modern design software. A sensible reading bans undisclosed, unreviewed, autonomous AI output. Ask the client, in writing, which they mean. Usually they are protecting against data leakage and low quality, in which case you can keep the clause and add a carve-out: AI permitted where no client data is entered and all output is reviewed. If they truly want zero AI, decide whether you can deliver profitably under that limit and price it in, or decline. The one thing not to do is sign a blanket ban you cannot honor, because breaking it later is a contract breach, not a tooling slip.
The platform adds its own layer. If you work through Upwork or Fiverr, their policies stack on top of your contract. Per a summary of Upwork's January 2026 update, AI-training permissions there run on a dual opt-in, both you and the client must agree before a contract's data is used for training, and accounts are opted in by default as of January 5, 2026, so you configure your AI preferences in account settings rather than the contract. Check the platform's current setting before you rely on the contract alone.
pro tip
If a client's restriction is tied to confidentiality rather than AI specifically, the obligations may already live in your NDA. Cross-check the NDA guide so your AI clause and your confidentiality terms do not contradict each other.
Keep your liability proportional
Unlimited indemnity for "any issue arising from AI" means carrying risk for decisions you did not make, like how a model vendor trained its system. Scope it down. Your contract should make you responsible for two things: reviewing the deliverable, and your own conduct. You signed off on the work, so you stand behind the work. You did not build the model, so you do not insure its training data. That single distinction keeps your exposure matched to what you actually did, which is the whole point of a liability clause.
Put it in the contract once, then stop worrying
An AI clause is not a niche legal flourish. It is becoming a standard part of a service agreement, the way a late-fee clause or a revision cap is. Add the four components to your template now, while no client is asking, and the next time the topic comes up you have a clean answer instead of a frozen pause.
Your AI clause is complete when...
FreelanceDesk builds contracts with the ownership, payment, and confidentiality terms already in place, so adding an AI-use clause is a small edit, not a from-scratch draft. It is free, and the document never leaves your browser. For the foundations the AI clause sits on top of, start with contract essentials, and if you are drafting with AI in the first place, check whether an AI-generated contract is even legally binding before you send it.