01Set staff rules
Do You Have to Tell Clients You Used AI? When to Disclose, With Sample Wording
A five-question path for when to tell clients or colleagues that AI helped, what the rules say, and copyable wording for letters, emails and questions.
- Topic
- Set staff rules
- Written by
- AI Ground Rules
- Last reviewed
- Sources
- 7 checked sources
Usually there is no single law that says “tell every client whenever you use AI.” But you may still have to tell them. The duty can come from your client’s contract or engagement terms, your professional body, the law where you work, promises your firm has made, or the client simply asking. More than one can apply at once. If none applies, follow your firm’s own rule. Whatever you decide about telling the client, a person checks the work before it goes out.
This guide is not legal advice. Check your own obligations with a qualified adviser.
The five questions
Check all five. More than one can apply at the same time, so do not stop at the first “yes”. Do everything each “yes” asks. If two answers pull different ways, follow the stricter one and ask the person named in your policy.
- Does the client’s contract or engagement letter say anything about AI? Read the engagement letter, the contract and any written client instructions, such as supplier or outside counsel guidelines. Some clients ban AI on their work. Others require notice or permission first. If it says anything, follow it. A client’s terms beat your firm’s own rule.
- Does your professional body have a rule? Lawyers, accountants, tax agents, financial advisers and health professionals may have rules or guidance on AI, confidentiality and talking to clients. Look for your own body’s current guidance. A few examples are in the next section.
- Does a law apply to this use? For most everyday drafting, probably not. But some laws cover chatbots, published content and licensed services. See what the rules say.
- Has your firm promised something? Check your website, proposals, terms of business and privacy notice. If you have said “all work is done by our team” or “we do not use AI on client files”, your staff must keep that promise, or the firm must change it and tell clients.
- Has the client asked, or are they talking to an AI tool? If a client asks whether you used AI, give a straight answer. If a client is chatting with an AI tool rather than a person, tell them before the chat starts.
If the answer to all five is “no”, use your firm’s default rule. A sensible default for most small firms:
- No note needed when AI helped with spelling, tone, formatting, a first outline or internal notes, and a person wrote or fully checked the final version.
- Add a short note when AI drafted a substantial part of a deliverable, such as a report, summary or analysis, even after a person checked it.
- Always tell the client when they ask, when AI output will shape an important decision for them, or when they are talking to an AI tool rather than a person.
Review comes first, whatever you disclose
Disclosure is not a way round review. A person checks AI-assisted work before it goes to a client or is published. This is rule 4 in the policy template and check 03 in the AI Use Checklist. If no one has checked the work properly, do not send it, with or without a note. Hold it until it is reviewed.
Telling colleagues
Inside the firm, tell your reviewer when AI helped, and which parts, so they look harder at facts, figures, names, dates and references. The acceptable use policy makes this a rule.
What the rules say
These are the rules we checked against their primary sources on 14 November 2026. They are examples, not a full list. Laws and guidance change, so check the linked source before you rely on it.
| Rule | Who it covers | What it requires | What it does not require |
|---|---|---|---|
| EU AI Act, Article 50 | Firms that operate in the EU, or whose AI output is used there | From 2 August 2026, people must be told when they are talking to an AI system, unless it is obvious (this duty sits with the system’s provider). Deployers must label deepfakes, and label AI-generated text published to inform the public on matters of public interest. That text label is not needed if a person did a real review and someone holds editorial responsibility. | It does not require a label on every email or document a firm sends a client. The Commission says spell-checks and other surface checks do not count as human review. |
| ABA Formal Opinion 512 (US lawyers) | Lawyers, under the ABA Model Rules | Lawyers must disclose their AI practices if a client asks, or if the engagement terms or the client’s guidelines require it. In some cases they must raise it unprompted, for example when AI output will shape a significant decision in the matter, or when AI use affects the fee. Informed consent is needed before client information goes into a self-learning AI tool. General boilerplate in an engagement letter is not enough for that consent. | It does not require disclosure of every use. ABA opinions guide; your state’s own rules and bar guidance decide. |
| Utah Code 13-77-103 | Suppliers using generative AI with consumers in Utah. Also people in “regulated occupations”: jobs regulated by the Utah Department of Commerce that need a state license or state certification | A supplier must say the person is dealing with AI, not a human, if the person clearly asks. People in regulated occupations must prominently disclose that the client is dealing with AI when it is a “high-risk” interaction: collecting health, financial or biometric data, or giving personal advice someone could rely on for significant decisions, such as financial, legal, medical or mental health ones. In a spoken interaction, say it at the start. In a written one, put it in writing before the interaction starts. | It covers interacting with AI, not every piece of AI-assisted work. A separate Utah law, the AI Policy Act (Chapter 72), is due to be repealed on 1 July 2027. That repeal date is for Chapter 72; this disclosure rule is in Chapter 77. Check the current status before you rely on either. |
The EU AI Act dates
Article 50 has applied since 2 August 2026. The EU’s 2026 “Digital Omnibus” changes did not move that date. They gave one short extra period, until 2 December 2026, for providers whose AI systems were on the market before 2 August 2026 to meet the duty to mark AI-generated content in a machine-readable way. That duty falls on providers: usually the companies that build AI tools, not firms that simply use them. But if your firm has an AI system built and puts it into service under its own name or brand, such as an own-brand client chatbot, your firm can count as a provider too. Ask your adviser if that applies. Content made before 2 August 2026 does not have to be labelled after the fact.
Professional bodies
Many professional rules are about confidentiality more than disclosure. They ask for client permission before client information goes into an AI tool, which means you end up telling the client anyway. Our client data guide covers examples for Australian tax agents and US tax preparers. If you are in a regulated profession, find your own body’s current AI guidance before you write your rule.
Sample wording you can copy
Replace the square-bracket fields. Have your adviser check the engagement letter clause, especially in a regulated profession. If your professional rules need specific informed consent, a general clause will not be enough.
Some samples make promises about training, human review and opting out. Only use a promise if it is true. Check it for every tool, every plan and every connected provider your staff use, for example in the vendor’s terms and settings. If one tool does not meet it, change the wording or stop using that tool on client work. The client data guide shows what to check.
Engagement letter or terms of business clause
Use of AI tools. We may use AI tools approved by [Business name] to help with tasks such as drafting, summarising and checking our work. We use only work accounts that do not use your information to train AI models, and we put your information into them only where our internal rules allow it. A qualified member of our team reviews all AI-assisted work before it reaches you, and we remain fully responsible for it. We will not use AI on your work if you ask us not to, and we will follow any AI rules in your own policies that you share with us. Ask [name, role] if you have questions.
If the client says no, record it where staff will see it, for example in the client file or practice management system: “No AI on this client’s work. Instruction dated [date].”
A line on a deliverable
Use this when AI drafted a substantial part of a document, as in the policy template:
Parts of this document were drafted with AI assistance and checked by [name].
A short email line
For an email footer or a one-off note:
We sometimes use approved AI tools to help draft our work. A member of our team checks everything before it is sent.
Only add a standard footer if it is true for every email it appears on.
If you use an AI chat assistant or automated replies
You are chatting with an AI assistant. It can make mistakes. To speak to a person, [how to reach them].
Automated replies are for simple, approved messages only, such as “We received your message and [name] will reply by [day].” Write and approve the wording in advance. Do not let them give advice, figures or answers about a client’s matter. Anything else goes to a person, who checks it before it is sent.
When a client asks “Did you use AI on this?”
Give a straight answer. Say what the AI did and who checked it. Do not guess. If you do not know, find out and reply.
Yes. We used [tool] to [produce a first draft of the summary / check the figures against the source file]. [Name] reviewed it line by line against the original documents before we sent it. We only use approved work accounts[, and our provider’s terms say your information is not used to train its AI models]. If you would prefer we did not use AI on your work, let us know and we will note it on your file.
If AI was not used:
No, this was prepared by [name] without AI tools.
Who staff ask when unsure
Name one person. In most small firms that is the data decision owner, the same person who approves tools and data types. Name a backup too. Write both names, and how to reach them, into your AI policy and the client file notes.
Staff should ask before sending, not after, when:
- the client’s terms say anything about AI, or they cannot find the terms;
- a client asks about AI use and the staff member is not sure of the answer;
- AI did more than help with wording, and they are not sure whether to add a note;
- they are about to use different wording from the approved samples.
The default while waiting for an answer: hold the work. Do not send it with a guess.
Telling the client does not make it OK
Disclosure answers one question: does the client know AI helped? It does not answer the other one: was the firm allowed to put that information into that tool?
A note saying “we used AI” does not fix client information that went into a personal account, an unapproved tool, or a tool the client’s contract rules out. Restricted information stays restricted, whatever you tell the client. Check the client data rules first. Disclosure comes after.
The same goes for review. A disclosure line is not a substitute for checking the work. The person named in the line must actually have checked it.
Fictional examples
These scenarios use invented firms and people.
| Scenario | What to do |
|---|---|
| At Harbor Accounts, a bookkeeper uses the firm’s ChatGPT Business workspace to tidy the wording of a covering email. She wrote the content herself. | No note needed. A person wrote and checked the content. |
| At Northgate Advisory, a consultant uses Copilot to produce the first draft of a 12-page market summary for a client. A partner checks every figure against the sources. | Add the deliverable line, naming the partner. |
| Fernlea Legal’s engagement letter with a client says: “No generative AI on our matters without written consent.” A paralegal wants AI to summarise a long contract. | Stop. Ask the partner, who must get the client’s written consent first. A note afterwards is not enough. |
| A client of Lantern Talent asks whether a candidate shortlist was “picked by AI”. AI ranked the CVs; a recruiter made the final choice. | Tell the client exactly that: AI ranked, and a named recruiter decided and can explain the choice. |
| At Brightwater Planning, an adviser wants to paste a client’s bank statements into an AI tool and add a disclosure line to the report. | Telling the client does not make it allowed. Bank and payment-card details are “never, unless approved in writing”. Check the client data rules first. |
Put it in your policy
Your disclosure rule belongs in your AI policy, next to the rules on review and approved tools. Write down which client terms to check, when a note is needed, the approved wording, and who to ask. To see how your firm does on disclosure and the six other basics, try the free AI Use Checklist.
What this guide does not cover
This guide helps you set a practical business rule. It is not legal advice and does not decide your obligations under the law, your professional rules or your client contracts. Rules in your country, state or profession may differ from the examples here. Have a qualified adviser check your rule and wording, especially for regulated work.
02Evidence
Sources
- European Commission, Transparency obligations under Article 50 of the AI Act (FAQ, updated 24 July 2026)Checked
- EUR-Lex, Regulation (EU) 2024/1689 (the AI Act), Article 50Checked
- American Bar Association, Formal Opinion 512: Generative Artificial Intelligence Tools (29 July 2024)Checked
- Utah Code 13-77-103, Required disclosures (effective 7 May 2025)Checked
- Utah Code 13-77-101, Definitions (effective 6 May 2026)Checked
- Utah S.B. 332 (2025), Artificial Intelligence Revisions: repeal date of Title 13, Chapter 72 moved to 1 July 2027Checked
- European Commission, AI Omnibus enters into force (27 July 2026)Checked