Guides

How to agree AI use with a client (before it gets awkward)

By Rory Wilmer, InkByte · 7 min read · 22 October 2026

Most agencies and freelancers now use AI somewhere in their work. Many clients do too. What's often missing is a conversation about it. The client assumes one thing, the agency assumes another, and it only comes up when someone spots a familiar AI phrase in a deliverable or asks where their brief has been pasted. By then it's a trust problem, not a process question.

The fix is simple and a bit unglamorous: agree how you'll use AI before the work starts, and write it down. This guide covers what to talk about and gives you a one-page note you can adapt.

This is general information, not legal advice. Check anything legal, HR or contractual with a qualified professional before you act on it.

Why this is worth doing now

Clients are asking sharper questions about AI. Some want to know whether they're paying for human work. Some have their own AI policies, written by legal or IT teams, that reach their suppliers. Some are using the same tools themselves and want to know what an agency adds. All of those are fair questions, and the agency that answers them clearly, before being asked, looks more trustworthy, not less.

It also protects you. If the client has agreed in writing which tools you use and for what, you're not left arguing about it after the fact.

1. Say which tools you use, and on which plan

Name the tools and the type of account, because the data terms differ. OpenAI's help centre says that by default it does not train on content from ChatGPT Business, Enterprise, Edu or ChatGPT for Healthcare workspaces, and that individual users can stop new conversations being used for training by turning off "Improve the model for everyone". Anthropic's privacy centre says it doesn't train on inputs or outputs from its commercial products (such as Claude for Work and the API) by default, and that on consumer plans (Free, Pro and Max) chats are used for training only if you allow it, or in a few other cases it lists, such as conversations flagged for safety review.

So "we use ChatGPT" is not enough. "We use ChatGPT Business, which by default isn't used for model training" is the kind of sentence a client's legal team can actually check.

2. Agree what client material can go in

Split the client's material into three groups and agree each one:

• Fine to use: published copy, public website text, approved brand guidelines.
• Ask first: unreleased campaigns, research, sales figures.
• Never: personal data about customers or staff, anything under a separate NDA, anything the client marks as restricted.

Personal data deserves extra care. If you'd be processing it on the client's behalf, data protection law applies, and that's a conversation for their data protection lead, not something to settle in a chat window.

3. Be clear about what AI does and what people do

This is the part clients care about most. Describe it in terms of the work, not the technology. For example: "We use AI to summarise research, draft first versions of social copy and generate headline options. A named writer edits every piece, and nothing goes to you that a person hasn't read and approved." If there are things you won't use AI for, such as final strategy recommendations or anything presented as a quote from a real person, say that too.

Don't overclaim in either direction. If AI drafts a lot of your first-round copy, say so. Clients usually mind being surprised far more than they mind the AI itself.

4. Talk about ownership early

Ownership of AI-assisted work is not as settled as ownership of work a person writes from scratch, so it's worth raising rather than assuming. In the UK, section 9(3) of the Copyright, Designs and Patents Act 1988 has a specific rule for computer-generated literary, dramatic, musical or artistic works: the author is taken to be "the person by whom the arrangements necessary for the creation of the work are undertaken". The government's 2024 to 2025 consultation on copyright and AI included questions on copyright protection for computer-generated works, so this area may change.

What you can control is the contract. Make sure it says who owns the deliverables, that it covers AI-assisted work as well as other work, and what happens to any working files. If the rules matter to the client, for example because they want to register a trade mark or license the work, that's the point to get a lawyer involved.

5. Decide who owns the voice files

If you build voice rules, prompts or a brand voice sheet for a client, agree up front whether they're part of the deliverables. A client who has paid for a set of prompts that make AI sound like their brand will reasonably expect to keep them. If you plan to reuse your general method, say which parts are yours and which parts are theirs.

6. Agree how you'll tell people

Decide together whether and how AI use is disclosed, both to the client's own teams and, where relevant, to the public. Some platforms and sectors have their own labelling rules for AI-generated or altered images and video, so check those for the channels you're working in.

A one-page AI note you can adapt

AI USE NOTE: [Agency/freelancer] for [Client]   Date: [  ]

TOOLS:          [Tool] on [plan]. Training on our content: [off / not by default].
FINE TO USE:    [published copy, guidelines, ...]
ASK FIRST:      [unreleased work, research, ...]
NEVER:          [personal data, NDA material, ...]
AI DOES:        [first drafts of X, summaries of Y, ...]
PEOPLE DO:      [edit and approve every piece; strategy; final copy]
OWNERSHIP:      Deliverables owned by [ ] per contract clause [ ].
VOICE FILES:    Prompts and voice rules built for [Client] belong to [ ].
DISCLOSURE:     [what we label, where, and how]
REVIEW:         We'll revisit this note every [6] months or when tools change.

Agreed by: [client contact]   [agency contact]

Keep it to a page. If it needs more than that, it probably belongs in the contract itself, with proper legal advice.

The short version

• Raise AI use before the work starts, not after someone notices it.
• Name your tools and plans, and check what each one does with your data.
• Agree what client material can go in, and what never does.
• Say plainly what AI does and what people do, and don't overclaim.
• Settle ownership and voice files in the contract, and take legal advice where it matters.

Done well, this note does more than avoid trouble. It shows the client you've thought about their brand and their data before they had to ask.