Do You Have to Tell Clients You Use AI? What Is Required, and What Is Just Wise
Whether you have to disclose AI use to clients: what the law actually requires, why your contract matters more, and how to word it without losing the work.
- Updated
- Read
- ~8 min

For most client work there is no general legal duty to announce that AI helped you produce something. The thing that does bind you is usually your own contract — and it rarely mentions AI at all.
This question gets answered badly in both directions. One half of the internet tells freelancers the EU AI Act now forces them to label everything; the other half says disclosure is purely optional and a bad idea commercially. Both are reading one narrow rule as if it were a general one, and both miss the obligation that actually catches people.
There are two disclosures, not one
Almost every argument about this collapses once you separate the two questions hiding inside it.
| The question | What it is about | What usually governs it |
|---|---|---|
| Did a tool help make this? | Credit, process and expectations — how the work was produced | Ethics, client expectations, and occasionally a professional body or platform rule |
| Did you send my information to a third party? | Data — whose records went where, and who else now holds them | The confidentiality and data-handling clauses in the contract you already signed |
The first row is the one everybody debates. The second is the one that creates actual liability, and it does so whether or not anyone ever raises the subject of AI, because the clause does not need to mention AI to cover it. “You will not disclose confidential information to third parties without written consent” is a sentence that has been in contracts for decades, and pasting a client’s customer list into an AI tool is disclosing it to a third party.
That is the reframe worth carrying away. You may well have no obligation to tell a client that AI helped you write their landing page, and simultaneously have a binding obligation not to have put their unpublished financials into the tool that helped you.
What the law actually requires
The EU AI Act, read properly
Article 50’s transparency obligations have applied since 2 August 2026, and they are the source of nearly every overstated claim on this topic. What they say, in the parts that could touch a small business:
- Providers — the companies that build AI systems — must mark synthetic audio, image, video and text so it is machine-readable and detectable as AI-generated. That duty is OpenAI’s and Anthropic’s, not yours.
- Providers of systems that interact directly with people must make sure people know they are talking to an AI, unless that is already obvious.
- Deployers — which can include you — must disclose deepfakes, and must disclose AI-generated or manipulated text published on matters of public interest where it has not had human editorial review with a person taking responsibility for publishing it.
- Deployers must also tell people when they are using emotion recognition or biometric categorisation on them.
A grace period runs to 2 December 2026 for the machine-readable marking of generative systems already on the market before August, and content generated before that date does not have to be labelled retroactively. The Commission issued non-binding guidance on Article 50 in July 2026, and the EU AI Office published a voluntary code of practice on transparency of AI-generated content at the end of that month. Penalties at the top of the scale reach €15 million or 3% of worldwide annual turnover.
Now read the third bullet against what you probably do. A freelancer who drafts a client’s marketing email with AI help, edits it, and sends it for the client’s approval is not publishing unreviewed text on a matter of public interest. There is human editorial review, there is a person responsible, and it is not public-interest commentary. The duty that people quote at freelancers mostly does not reach them.
Outside the EU
There is no equivalent across-the-board duty in most jurisdictions to say that AI helped you produce a piece of work. What exists instead is a patchwork that is easy to miss:
- Regulated professions — legal, accounting, medical, financial advice — increasingly have rules or guidance from their own bodies, and those can be stricter than any statute.
- Platforms and marketplaces sometimes have their own terms about AI-produced deliverables.
- Public sector and enterprise procurement routinely asks the question directly now, in the questionnaire rather than the contract.
- Consumer protection law already covers misleading claims, which is the route by which “written by our expert team” becomes a problem if it was not.
Your contract is the part that binds you
Before worrying about legislation, read the agreement you have actually signed. Four clause types do the work, and none of them will contain the word “AI”:
- Confidentiality. Restricts disclosing the client’s information to third parties. An AI tool is a third party.
- Data processing. May require you to name your sub-processors, keep data in certain regions, or get consent before adding a new one.
- Warranties about the work. Promises of originality, accuracy, or that the deliverable does not infringe anyone’s rights. AI output can put pressure on all three, and the promise is yours regardless of how the draft was produced.
- Subcontracting. Sometimes requires consent before anyone other than you works on the deliverable. Whether a tool counts is arguable; the argument is one you would rather not be having.
The practical test
- Best for
- Checking three things before the next project: does the contract restrict third parties or name sub-processors, does it warrant originality or accuracy, and has the client issued an AI policy of their own
- Not for
- Assuming silence means permission. A contract that does not mention AI is not a contract that permits putting client records into it
How to word it
Disclosure fails when it reads like a confession and when it arrives late. Put it in the proposal, in plain sentences, before anyone has paid you anything.
What a client is actually asking, underneath the question, is three things: did a competent person check this, who is accountable if it is wrong, and where did my information go. Answer those and the tool becomes uninteresting.
A form that holds up in a proposal:
I use AI assistance for research and first drafts. Every deliverable is reviewed and edited by me before it reaches you, and I am responsible for its accuracy. I will not put your confidential material into any third-party tool without asking you first.
Three sentences, and each is doing a job. The first is honest about the process without overstating or understating it. The second names a responsible human — the part that reassures. The third answers the data question before it is asked, which is the one that could otherwise become a contractual problem.
If you want a version for a client who has asked after the fact, be more specific rather than more general: name which parts of the work the tool touched, what you did to the output, and what you verified. Specificity reads as control. Vagueness reads as evasion, even when it is not.
The one thing not to do
Do not deny it.
A client who asks whether you used AI has usually already decided the answer and is checking what you will say. The cost of a vague or false denial is not that they learn you used a tool. It is that they stop trusting the rest of the work, including the parts you did entirely yourself, and there is no version of the conversation afterwards that recovers that.
This is the whole commercial case for putting the line in the proposal. It is not that disclosure is free — with some clients it genuinely is not. It is that disclosure is cheap and discovery is not, and you only control which of the two happens while you are still the one choosing.
Write down your own rule
If anyone besides you touches client work, your answer to this question has to be the same whoever gets asked. One page is enough, and the useful version names only five things:
- Which tools people may use, and on which accounts — a company workspace rather than personal logins, so the setting is yours to set. See ChatGPT Business vs Plus for why that distinction matters more than it looks.
- Which data may go in, and which never may. Write the second list first; it is shorter and it is the one that matters.
- Who reviews output before it leaves the business, and that it is always a named person.
- What you tell clients, in the exact words, so nobody has to improvise.
- Who decides the grey cases. One name.
That document is also the thing you hand over when a client, insurer or tender asks how you handle AI — a question that is now routine and that nobody answers well off the cuff.
Where to go next
- Where the work should live. ChatGPT Business vs Plus covers company-owned accounts, training defaults and why a personal subscription makes the data answer unverifiable.
- If you built a client workflow on a custom GPT. Custom GPTs are retiring explains the December deadline and where to move it.
- Keeping the human review real. How to use AI tools covers the checking habits that make the accountability claim true rather than decorative.
- Judging the output you are putting your name to. Best AI writing tools covers what these tools are and are not good at.
- The wider category. AI business and productivity tools is the hub for admin, meetings, bookkeeping and support.
Share this guide
Keep reading
Category hub
AI Business & Productivity
How solopreneurs and small business owners can use AI tools for admin, meetings, scheduling, bookkeeping, and support - and how to choose the right ones.
In this category
Best AI Tools for Small Business in 2026: Picks by Use Case
Eleven AI tools for small business, picked by job - writing, design, CRM, meetings, bookkeeping, support, automation - with current pricing and real limits.
In this category
AI Tools vs Virtual Assistant
AI tools vs hiring a virtual assistant - compare costs, strengths, and risks, with worked examples and a framework to decide what to automate or delegate.
In this category
How Much Does AI Cost for a Small Business?
What AI costs a small business at three budget levels - free, $20 to $50, and $100+ per month - with named tools, current list prices, and hidden costs.
In this category
Claude for Small Business: 8 Practical Workflows
Explore 8 practical Claude for Small Business workflows, the integrations they depend on, their limits, and where human approval still matters.
In this category
Claude vs ChatGPT for Small Business: Which Is Better?
Claude vs ChatGPT for small business, compared across real daily workflows: documents, support, spreadsheets, integrations, team plans and current pricing.
In this category
AI Chatbot for Small Business: What It Can Actually Do
What an AI chatbot can and cannot do for a small business, what it needs from you first, how to add one to your website, and how the pricing really works.
In this category
Custom GPTs Are Retiring: What to Do Before 11 December
OpenAI is retiring custom GPTs on 11 December 2026. What a plugin is, what migrates and what does not, and how to pick between a plugin, a Project and a Skill.
In this category
ChatGPT Business vs Plus: Which Plan Do You Need?
ChatGPT Business vs Plus for a small business: what each tier buys you, how the training default differs, and the one rule that settles which to pay for.
In this category
AI Use Policy Template for Small Business
A one-page AI use policy for a small business, with wording to copy: approved tools, what data may go in, who reviews output, and who owns the grey cases.
In this category
AI Notetaker Consent: When You Have to Ask
When you need consent to use an AI notetaker, which states require everyone on the call to agree, and the line to say at the start to stay on the safe side.
In this category
Is ChatGPT Safe for Business Data?
What really happens to company data you paste into ChatGPT - training, retention, deleted chats, share links, personal accounts - and what should never go in.
In this category
Best AI Browser for Small Business 2026: Honest Comparison
Perplexity Comet, Dia, and ChatGPT desktop compared for small business. Pricing, agent features, privacy, and which one actually saves you time.
In this category
Small Language Models for Solopreneurs: Save on AI Costs
You are overpaying for GPT-4o on tasks a smaller model handles fine. The 4 cheaper models, where they win, and how to start using one this week.
Questions
Frequently asked questions
Do I legally have to tell clients I use AI?
In most ordinary client work, no — there is no general legal duty to announce that you used an AI tool to help produce something. The duties that do exist are narrower than the internet suggests and they attach to specific situations rather than to AI use in general. What binds most freelancers and agencies is not a statute at all. It is the contract they already signed, which may well restrict what they can put into a third-party tool even though it says nothing about AI.
Does the EU AI Act require me to disclose AI-generated content?
Only in a narrow set of cases, and probably not the one you are in. The marking obligations in Article 50 fall mainly on providers — the companies that build the AI systems — rather than on you for using one. The duty that applies to users of a system covers things like deepfakes and AI-generated text published on matters of public interest without human editorial review. Client marketing copy that you wrote with AI help and then edited yourself is not that. Article 50's obligations have applied since 2 August 2026, with a grace period running to 2 December 2026 for generative systems already on the market before that date.
What is the difference between disclosing that you used AI and disclosing what you put into it?
They are two separate questions and only one of them is usually binding. The first is about credit — whether a tool helped you produce the work. The second is about data — whether you sent your client's information to a third party. The first is mostly an ethics and expectations question. The second is frequently governed by the confidentiality and data clauses in your existing contract, which means it can be a breach whether or not anybody ever asks about AI.
Will telling a client I use AI cost me the work?
It can, with a client who equates AI with corners being cut. Being found out later is the worse outcome by a distance, because it reframes everything else you have delivered. The framing that holds up is not a confession that you used a tool — it is a statement that a named, accountable human checked the output and stands behind it. Most buyers are not objecting to the tool; they are checking that somebody is still responsible for the result.
How should I word an AI disclosure?
Put one or two plain sentences in the proposal rather than in the small print, say what the tool does and what you do, and name who is accountable. A workable form is to state that you use AI assistance for drafting and research, that every deliverable is reviewed and edited by you before it is sent, that you remain responsible for accuracy, and that you will not put their confidential material into any third-party tool without asking first. It fits in a proposal and answers the three things a client actually wants to know.
What should I do if a client asks me directly whether I used AI?
Answer honestly, in specific terms, and do it in the same message. Say which parts the tool touched, what you did to the output, and what you checked. A vague denial is the single most expensive move available to you here, because it converts a question about process into a question about your honesty, and the second one is not recoverable.
Can a client ban me from using AI on their work?
Yes, and some do, usually through a clause about third-party tools and confidential information rather than a clause naming AI. If the restriction is in the contract, it binds you regardless of what you think of it. If you cannot deliver profitably without AI assistance, that is a conversation to have before signing rather than a rule to quietly work around — and the quiet route is also the one most likely to end in a dispute you lose.
Do I need a written AI policy for my business?
If anyone other than you does client work, yes, and it can be one page. Name the tools people may use and on which accounts, sort your data into what may and may not go into them, require that a human reviews anything before it leaves the business, and name one person to ask about the grey cases. The value is not the document; it is that you can answer a client's question in one consistent sentence rather than guessing on the spot.