A Checklist for Disclosing AI-Assisted Writing to Clients

Written by

in

Clients usually care less about whether you opened an AI tool than about what it did, what information went into it, and who stands behind the finished work. A writer who uses a tool to suggest headings is doing something different from one who submits an almost untouched draft. Treating both cases with the same vague disclaimer is unlikely to build confidence.

If you are wondering how to disclose AI assisted writing to clients, the practical answer is to agree on the boundaries before work begins, describe your use in plain language, and make clear what you personally checked. This checklist is for freelancers, agencies and other professionals who create written work for clients. It is not legal advice: contracts, confidentiality duties and local rules may require more specific steps.

Start with the client’s expectations, not the tool

There is no single disclosure sentence that fits every project. A client commissioning a sensitive policy document may want a strict no-AI process. A small business asking for ten social captions may be comfortable with AI-assisted drafting, provided a human edits and checks them. Your first job is to find out which situation you are in.

Ask about the client’s rules before using a tool on their material. Check the contract, statement of work, editorial policy, security requirements and any instructions from the person approving the work. If the documents are silent, ask directly rather than treating silence as permission.

Questions to settle before drafting

  • Does the client allow AI tools for this kind of work?
  • Are there prohibited tools, accounts or categories of data?
  • Is AI use allowed for brainstorming and editing, but not for producing draft copy?
  • Does the client expect disclosure in the deliverable, in a project note, or only on request?
  • Will the work be published under the client’s name, used in a regulated setting, or submitted under rules that restrict assistance?
  • Who will approve factual claims and final wording?

These questions matter because “AI-assisted writing” can describe very different workflows. For a plain-language explanation of where writing assistants differ from conventional grammar checkers, see our guide to AI writing tools versus grammar checkers.

Use this disclosure checklist

Before you begin, make sure you can answer each point. You do not need to send clients a long technical report. You do need enough detail to prevent a reasonable person from misunderstanding how the work was made.

1. Say what kind of AI assistance you mean

Name the task, not just the technology. “I may use AI” is too broad to tell a client much. Explain whether a tool might help organise a brief, suggest an outline, rewrite a sentence, draft options, summarise source material or check tone.

For example: “I may use an AI writing tool to suggest an initial outline and alternative headlines. I write and edit the article, and I do not use AI-generated text as final copy without review.” That tells the client more than “AI may be used in production.”

Be accurate about the actual workflow. If you expect to use AI for first drafts, do not describe it as a spellchecker. If you only use a grammar tool to flag possible errors, do not imply the whole piece was machine-written.

2. Draw a boundary around client information

Explain what you will and will not enter into the tool. A client’s unpublished product plans, customer details, internal reports, personal information and login credentials should not be pasted into an AI service unless the client has approved the tool and the handling is appropriate for that data.

“I won’t enter confidential client material into public AI tools” is a useful starting point, but be precise about what you mean by “public” and how you will work instead. You might use a sanitised brief, remove identifying details, work only in an approved account, or avoid AI on that project altogether. Do not promise that data is never stored or used for training unless you have checked the current terms and settings for the specific service and account.

Privacy obligations depend on the information, the tool and the relevant jurisdiction. The UK Information Commissioner’s Office explains that organisations using AI still need to consider data protection duties, including how personal data is used. Its AI and data protection guidance is a useful reference point when a project involves personal information. If you are a freelancer processing data for a client, clarify roles and requirements with them rather than assuming your usual workflow applies.

3. State who is responsible for the finished work

A client should know that a human is accountable for the delivered copy. Say what you will review: facts, names, quotations, tone, structure, links, brand terminology, and any claims that could affect legal, financial or health decisions. “Human-reviewed” is not a magic quality guarantee, so describe the checks that are relevant to the assignment.

For instance, a blog post based on supplied research may need source checks and a review of quoted figures. A customer email may need a check that the reply answers the customer’s actual question and does not promise something the business cannot provide. Our guide to fact-checking AI-assisted content covers a practical verification workflow.

4. Explain what the client will receive

Decide whether disclosure belongs in the proposal, contract, delivery email or the work itself. A client may want to know about your process without wanting a public note attached to every blog post. Another client may require an internal record or a visible statement. Confirm the destination and wording in advance.

Do not assume that a disclosure to your client automatically satisfies the rules of a publisher, employer, regulator or academic institution. If the copy will be submitted under a separate policy, ask who is responsible for checking that policy and what evidence is needed.

5. Keep a proportionate record

For a small, low-risk assignment, a short note in your project file may be enough: tool category, task, whether client data was used, and the checks completed. For a larger or sensitive project, the client may need a more formal record, including approved services, reviewers and sign-off. Agree the level of documentation rather than creating paperwork no one will use.

Choose the right level of disclosure

Disclosure should be proportionate to the work and the client’s expectations. The aim is not to list every feature you tried. It is to give the client a fair picture of the parts of the process that could affect trust, confidentiality, originality or quality.

Use case What to tell the client What to check
Grammar and spelling suggestions Whether an editing tool is used, if required by the client’s policy Meaning, tone and unintended changes
Outline or idea suggestions That AI may support early planning, while you develop the argument and final structure Whether the outline reflects the brief and original sources
AI-assisted first draft Which sections may be drafted with assistance and how you will rewrite and verify them Facts, originality, brand voice, citations and unsupported claims
Confidential or regulated material Whether AI is permitted at all and which approved process applies Contract terms, data handling, specialist review and sign-off

Use the table as a conversation starter, not a universal policy. Some clients draw a hard line at generated prose. Others permit it but want the final writer to own the accuracy and style. If a client asks for “no AI,” clarify whether that includes ordinary predictive text, grammar suggestions or transcription tools. Agree what the phrase means for that project.

How to write a disclosure that sounds clear and credible

A useful disclosure has four parts: the task the tool may support, the limits on its use, the human checks, and the client’s choice or approval. Keep it short enough to understand and specific enough to be meaningful.

Proposal or contract wording

“With your approval, I may use an AI writing tool for early outlining and to generate alternative phrasing. I will not submit unreviewed AI-generated copy as final work. I will check the finished text for accuracy, fit with your brief and brand voice. I will not enter confidential or personal client information into an AI service unless we agree on an approved tool and process in writing.”

Adjust this to match your actual work. If you use AI for drafting, say so. If you do not use it at all, do not include a generic clause that suggests otherwise. The clause should also be consistent with your contract and any promises about confidentiality.

Short note in a delivery email

“For transparency: I used an AI tool to suggest headline options and organise the initial outline. I wrote and edited the article, checked the claims against the sources listed, and reviewed the final version against your brief. No confidential client material was entered into the tool.”

Only include the last sentence if it is true. If you used a sanitised extract or an approved workspace, describe that accurately instead. A disclosure that overstates your safeguards can damage trust more than a simple, honest explanation.

When you need to ask permission first

“Your brief includes internal customer examples. Before I use any AI assistance, can you confirm whether it is permitted and, if so, which tool or account is approved? I can also complete the work without AI if that is your preference.”

This is not an awkward confession. It is a normal project question, like checking image rights or confirming who can approve a draft. Giving the client a genuine alternative makes the decision easier.

Three realistic client scenarios

A freelancer writing routine website copy

You are preparing service-page copy for a local bookkeeping firm. The client has supplied a public list of services and a tone guide. You use an AI assistant to brainstorm section headings, then write the copy yourself. A concise proposal note can disclose the outlining use and explain that you will check service descriptions with the client. There is little reason to make a dramatic claim about AI; the useful information is what it helped with and who verifies the details.

An agency producing a customer email sequence

The client wants a sequence for people who have downloaded a guide. You use a tool to draft subject-line variations and suggest a sequence structure. The final emails include product claims and a call to action. Tell the client about the drafting assistance, avoid entering customer data unless approved, and ask the client’s product or legal contact to confirm claims and terms. AI can help generate options; it cannot approve the business’s promises.

A consultant summarising internal interviews

The source material contains names, job titles and candid comments. This is not a case for quietly pasting transcripts into whichever tool is convenient. Check the contract, data-handling requirements and approved services first. If permission is unclear, ask. If the client says no, use a conventional method. The time saved by a tool is not worth creating an avoidable confidentiality problem.

What not to say

  • “AI was only used for inspiration.” This is vague if the tool produced paragraphs that remain in the final work. Explain the actual contribution.
  • “The output is guaranteed accurate because a human checked it.” Review reduces risk; it does not guarantee that every claim is correct. Name the checks and flag anything needing client confirmation.
  • “No one can tell.” Detection tools cannot reliably establish how a particular text was made. Do not use a detector score as proof of authorship or as a substitute for an honest process note. Our guide explains what AI content detectors can and cannot tell you.
  • “The tool owns nothing, so the copy is automatically yours.” Rights and ownership can depend on contract terms and applicable law. In the United States, the Copyright Office explains that copyright protection depends on human authorship, with AI-assisted works considered in context. Its AI and copyright information is a useful starting point, not a substitute for advice about a particular contract or jurisdiction.
  • “I use AI responsibly.” This is a conclusion, not an explanation. Tell the client what you do, what you avoid and what you check.

A simple workflow for each assignment

  1. Read the brief and contract. Look for restrictions on tools, data, authorship, confidentiality and disclosure.
  2. Ask about ambiguity. Get written approval if the brief is unclear or the work is sensitive.
  3. Choose the minimum necessary use. If AI can help with a low-risk outline, do not feed it an entire confidential archive.
  4. Keep notes as you work. Record the assistance used and any relevant data-handling choices.
  5. Review against the task. Check facts, sources, voice, completeness and client-specific details. For high-impact claims, ask the appropriate client contact to verify them.
  6. Disclose in the agreed place. Use the wording and level of detail the client approved, and update them if your process changes materially.

This workflow also gives you a clear answer if a client later asks how a passage was produced. You can describe the process without relying on memory or making broad claims about what the tool did.

When disclosure is not enough

Some projects should not use AI assistance unless the client explicitly approves a defined process. That may include material containing sensitive personal information, confidential commercial plans, unpublished research, legal or medical content, or work governed by a strict publication policy. Even with approval, you may need a secure environment, limited access, a qualified reviewer or a client-controlled system.

There are also cases where the better choice is not to use AI. A tool may add little value to a short, straightforward task, while introducing extra review and explanation. Or it may struggle to preserve a writer’s distinctive voice. If you use AI for editing, it helps to compare suggested changes with the original rather than accepting them by default; our guide to editing with AI without losing your voice covers that trade-off.

FAQ

Do I have to tell clients that I use AI to write?

There is no one answer for every job or jurisdiction. Your contract, client policy, publication rules or data-protection duties may require disclosure or restrict tool use. Even when no specific rule applies, agree expectations before work begins if AI could affect confidentiality, authorship or the client’s approval of the process.

Should I disclose grammar-checking tools?

Usually, the key question is whether the client’s policy includes those tools. A spelling suggestion is different from generating a draft, but some organisations define AI broadly. If the policy is unclear, ask what it covers rather than deciding on the client’s behalf.

Can I put an AI disclosure on every piece of client content?

You can, but a blanket note may be unhelpful if it does not describe the actual work. Agree whether the client wants a process note, a disclosure in the document or no public label. Match the wording to the assistance used and the intended audience.

What if the client asks me not to use AI?

Clarify the scope of the request, then follow it. Ask whether it covers drafting only or also outlining, editing, transcription and other tools. If your usual workflow relies on a restricted tool, explain the alternative and any effect on timing or cost before accepting the work.

Is a detector score proof that I did or did not use AI?

No. Detector results should not be treated as reliable proof of a text’s origin. Keep an accurate process record and explain your workflow directly if the client asks.

Comments

Leave a Reply

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