A Checklist for Disclosing AI-Assisted Writing to Clients

Written by

in

Clients usually care less about whether a writer opened an AI tool than about what it did to the work they commissioned. Did it suggest an outline, rewrite paragraphs, generate a first draft or supply facts? Was confidential material entered into it? Who checked the result?

If you are deciding how to disclose AI assisted writing to clients, the useful answer is not a blanket disclaimer pasted onto every project. It is a clear agreement about the task, the tool’s role, the checks you will make and any limits the client wants respected. That conversation is easier before work begins than after a client spots an unexpected tone or asks how a draft was produced.

This checklist is for freelancers, agencies and small teams using AI as part of writing work. It is practical guidance, not legal advice: contractual, regulatory and professional obligations can vary by location and assignment.

Why disclose AI assistance at all?

Disclosure helps set expectations. A client may be comfortable with AI helping to organise notes but not with it producing customer-facing copy. Another may welcome a first-pass draft but require a named editor to verify every claim. You cannot infer those preferences reliably from the brief.

Being open also gives you a chance to explain what you remain responsible for. AI output can be fluent and still contain invented details, miss important context or flatten a brand’s voice. A responsible writer does not treat tool use as a substitute for editorial judgement. For a practical review process, see our guide to fact-checking AI-assisted content before publication.

There is no single disclosure rule that applies to every writing task. The European Union’s AI Act, for example, sets transparency obligations for particular uses and types of AI-generated content; that is not the same as a universal requirement to label every sentence that received AI assistance. Check the rules that apply to your location, sector and deliverable rather than treating a general statement as legal clearance.

A checklist for disclosing AI-assisted writing

1. Describe the actual task, not just the tool

“I use AI” is too vague to help a client make a decision. Explain what the system will do on this assignment. There is a meaningful difference between asking for headline options and asking a chatbot to draft a complete article from a short prompt.

Be specific about the stage of work and the material involved. For example, you might use a tool to sort interview notes into themes, suggest an outline from a client-approved brief, or flag awkward sentences in a draft you wrote. If the tool will generate substantial prose, say so plainly.

It is also useful to distinguish assistance from delegation. If you will make the final decisions, revise the material and check its claims, say that. Avoid implying that every sentence is wholly human-written if that would give the client a misleading impression.

2. Check the brief, contract and client policy

Before starting, look for specific instructions in the statement of work, procurement terms, editorial policy or confidentiality agreement. A client may prohibit certain tools, require approval before use, or permit use only for non-sensitive material. A general agreement to use AI does not necessarily override a project-specific restriction.

If the documents are silent, ask. This matters particularly in regulated or high-stakes work, such as financial services, healthcare, legal information, public communications or materials written on behalf of a government body. The client may need an internal approval trail even where the writing itself seems routine.

Do not assume that a client’s use of AI means they permit you to use it on their data. The tools, accounts and data-handling arrangements may differ. The UK Information Commissioner’s Office guidance on AI and data protection is a useful starting point for considering personal data. If a brief contains identifiable customer, employee or patient information, clarify the approved process before entering anything into an external service.

3. Say what information will go into the tool

Clients are often more concerned about their information than about the mechanics of drafting. Tell them what you plan to provide: a public brief, anonymised notes, internal research, unpublished product details or customer examples. If you can complete the task without uploading sensitive information, that may be the simpler option.

Do not promise that information is “never stored” or “completely private” unless you have checked the specific service, account type and settings and can support that claim. Data use and retention terms can differ by provider and plan, and may change. Where privacy terms are material to the assignment, verify them directly and document the arrangement rather than relying on a general impression.

A useful rule is to share only the minimum information needed for the task. Replace names with roles, remove contact details, and paraphrase sensitive examples where that still allows useful work. If the client requires a particular platform or no external processing, follow that requirement.

4. Explain the human review you will provide

“I’ll review it” can mean anything from a quick read to a full editorial check. Specify the checks that matter for the deliverable. For a blog post, that might include comparing factual claims with cited sources, removing unsupported statements, checking links, revising tone and confirming that the final structure answers the brief.

For copy based on client materials, state that you will check the output against those materials rather than treating the tool’s summary as authoritative. For specialist content, identify whether a subject-matter expert or the client must approve technical claims. You should not imply that your review guarantees perfection; describe the process and its limits.

Keep the boundary clear: you remain accountable for the work you deliver, but some claims may require client confirmation or expert sign-off. A review process is only as good as the evidence available to the writer.

5. Agree what “AI-assisted” means for this project

Clients may use the phrase to mean anything from spelling suggestions to generated paragraphs. Agree on the practical boundary. You can define permitted uses in plain language, such as:

  • Permitted: brainstorming titles, organising a client-approved outline and checking grammar in a human-written draft.
  • Permitted with approval: generating draft paragraphs from an approved brief, provided you substantially edit and verify them.
  • Not permitted: uploading confidential source files, generating quotations, or presenting unverified AI-supplied facts as established information.

These are examples, not universal categories. Adapt them to the assignment. A client commissioning a personal essay may reasonably want no generative drafting at all; a business commissioning routine product descriptions may have a different preference.

6. Choose a disclosure level that matches the work

Not every use needs a long explanation. The right amount of detail depends on how much the tool contributed, the sensitivity of the material and the audience for the finished work. A small grammar suggestion may not need the same treatment as a generated first draft.

Consider three practical levels:

  • Brief note: for limited support such as outlining or copy-editing, when permitted by the client and relevant to the agreement.
  • Specific explanation: for substantial drafting, summarising or transformation of supplied material. State which stages involved AI and what review followed.
  • Formal approval or no use: for sensitive, regulated or explicitly restricted assignments. Get written permission before using a tool, or complete the work without it.

Some organisations also have rules about labelling published material. If the client will republish your work, ask whether their own disclosure policy applies and who is responsible for the final label. The NIST Generative AI Profile treats transparency and risk management as considerations for organisations, but it is a voluntary framework, not a substitute for a client’s policy or applicable law.

7. Put the agreement in writing

A short email or contract note is often enough for a low-risk assignment. Record what uses are allowed, which information may be entered, what review you will provide and whether the finished work needs a disclosure. If the client says no AI, confirm what that means in practice—for example, whether ordinary spelling and grammar features are included in the restriction.

For ongoing work, save the agreed wording with the project records. If the scope changes—for example, the client later sends confidential interviews and asks for a fast summary—pause and confirm that the original approval still covers the new material and task.

Disclosure examples you can adapt

Light assistance: “I may use an AI writing tool for outlining and language checks. I will not enter confidential information, and I will review and edit all copy before delivery. Please let me know if you would prefer no AI assistance on this project.”

Substantial drafting: “With your approval, I plan to use an AI tool to create an initial draft from the brief you provide. I will verify factual claims against reliable sources, rewrite the copy to fit your brand and take responsibility for the final version. I will not upload customer or confidential material.”

Client approval required: “Your policy is not clear on generative drafting. Before I use any AI tool beyond standard spelling correction, could you confirm what is permitted and whether the tool must be approved? I will wait for your answer before processing project materials.”

No AI use: “Understood. I will not use generative AI for this assignment. Please confirm whether your restriction also covers automated editing features built into writing software, so I can follow the policy consistently.”

These are starting points, not magic clauses. Adjust them to describe what you actually do. If your workflow includes an AI-based grammar checker, be precise rather than calling it a generative writing tool—or dismissing it as harmless by default. Our comparison of AI writing tools and grammar checkers explains why the distinction can matter.

Common disclosure mistakes to avoid

  • Disclosing after delivery. A client who would have chosen a different process should have the chance to decide before the work is done.
  • Using vague labels. “AI-enhanced” does not tell the client whether a tool checked punctuation or drafted the article.
  • Overstating verification. Say what you checked and against which sources; do not imply that a tool or a quick review makes every claim certain.
  • Promising perfect privacy. Describe the actual information-handling arrangement and check the relevant service terms.
  • Relying on an AI detector as proof. Detection tools can make mistakes and cannot reliably establish how a particular passage was created. A detector score is not a substitute for a transparent process or a clear agreement.
  • Assuming disclosure transfers responsibility. Telling a client that AI was involved does not remove your responsibility to meet the agreed brief, protect information and correct errors in your work.

A quick pre-project checklist

  1. Read the brief, contract and any client AI policy.
  2. Identify the exact writing tasks for which you want to use AI.
  3. Decide what information, if any, would be entered into the tool.
  4. Check whether the service and account are approved for that information.
  5. Agree the human review, fact-checking and client sign-off required.
  6. Choose a disclosure level that reflects the tool’s contribution and the project’s risk.
  7. Get approval in writing where the policy is unclear or the work is sensitive.
  8. Revisit the agreement if the scope, source material or tool changes.

The best disclosure is specific, proportionate and early. It gives the client a real choice, protects the working relationship and makes your own process easier to explain. If you cannot describe what the tool will do or how you will check its output, pause before using it on paid client work.

Frequently asked questions

Do I have to tell every client that I used AI?

There is no single answer for every country, industry and assignment. Check the contract, client policy and applicable rules. Even when disclosure is not expressly required, tell the client when AI use is material to the agreed process, involves their information or could affect their expectations about authorship and review.

Should I disclose AI use if it only corrected grammar?

That depends on the tool, the client’s definition of AI assistance and the project terms. Some editing features use automated processing; others may include generative rewriting. If a client has a strict no-AI policy, ask whether it covers those features. For ordinary work, a brief explanation of the feature may be enough when the distinction matters.

Can I put client documents into an AI writing tool?

Only after checking the client’s permission, the information’s sensitivity and the tool’s applicable data terms. Avoid entering personal or confidential information unless the client has approved the process and you have confirmed it is appropriate. Anonymising or summarising the material may reduce exposure, but does not automatically make every use acceptable.

Does disclosing AI use mean the client owns or approves the output?

No. Disclosure explains the process; it does not settle ownership, licensing, acceptance or liability. Those issues depend on the contract and applicable law. Put expectations about rights, revisions, approvals and responsibility in the project agreement where they matter.

What if the client asks whether a draft was written by AI?

Answer directly and describe the tool’s role without exaggerating either its contribution or your review. For example, say whether it helped outline, draft or edit, and explain what you checked yourself. If you are unsure whether a particular feature counts, describe the feature rather than relying on a broad label.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *