It usually arrives in a supplier form, between the insurance question and the bank details: "Please describe your use of artificial intelligence tools, including any processing of our data."
Or it comes up on a call. A client's new procurement person asks, quite nicely, "Which AI do you use on our files?" And there's a pause, because the honest answer is "a few people use ChatGPT, I think, on their own accounts."
If you run a small accounting practice, law office, agency or consultancy, expect this question this year, if you haven't had it already. The good news: the answer fits on one page, and most of it you can write this week.
The six questions that keep coming up
The UK magazine Business Matters listed the questions clients are starting to ask their suppliers in September 2026. In short:
- Which AI tools does your team use? Including the ones nobody approved.
- Does any of our data go into them? Names, documents, spreadsheets, emails.
- Is what we send used to train AI models?
- Where is the data processed, and under which country's law?
- Who approved it? A named person, not "the team".
- What happens when someone leaves? For example, a former employee's personal account still holding client work.
The magazine's own advice is blunt: "The mistake is treating it as a technology question. It is a commercial one." A client who can't get a clear answer from you can get one from the next firm on the list.
A weak answer and a good one
Here's what a weak answer tends to sound like:
"We take data security very seriously. We don't really use AI, and where we do, we use reputable tools."
It's vague, it names nothing, and "we don't really use AI" is often not true. A procurement team reads that as "they haven't looked".
Here's a good one:
"We use ChatGPT Business for drafting and summarising. Staff use it on our company workspace only, never personal accounts. Client names, bank details and ID documents never go in. OpenAI states it doesn't train on Business workspace data by default. Anything sent to you is checked by a person. Our office manager, Eleni, approved this in October and reviews it every six months. Leavers' access is removed on their last day."
Six sentences, six questions answered. That's the whole trick. (Eleni is an example; put your own person in.)
The one-page AI use statement
Copy this, fill in the brackets, and keep it where you can send it as a PDF. It's a starting point, not legal advice.
[Firm name]: how we use AI tools
Tools we use for client work: [e.g. ChatGPT Business / Claude Team / Gemini in Google Workspace], on our business account only. Staff don't use personal AI accounts for client work.
What we use them for: [e.g. first drafts of letters and reports, summarising long documents, tidying spreadsheets]. A person reads and approves anything that goes to a client.
What never goes in: passwords, bank and card details, ID documents, health information, and anything a client has asked us to keep out.
Training on your data: our provider states that it does not use our business account's data to train its models by default. [Link to the provider's page, and the date you checked it.]
Where it's processed: [the provider's stated region or data processing terms, if you know them; otherwise "under [provider]'s business data processing terms", with a link].
Who approved it: [name, role], on [date]. Reviewed every six months; next review [date].
Staff: everyone using these tools has been through our AI rules; last session on [date].
When someone leaves: their access to our AI accounts is removed on their last day.
Questions: [name, email].
If a client sends their own form, you can answer it straight from this page.
Check the training line before you write it
Question 3 is where small firms most often get caught out, because the answer depends on the plan, not the brand. Here's what each company says on its own pages, checked on 20 November 2026:
- ChatGPT: OpenAI says that by default it does not use data from ChatGPT Business, Enterprise or Edu for training (OpenAI business data page). On personal Free, Plus and Pro accounts, sharing for training is on by default; you can turn off "Improve the model for everyone" in Settings → Data controls (OpenAI Help Center).
- Claude: Anthropic says that by default it won't use inputs or outputs from its commercial products, such as Claude for Work and its API, to train its models, with an exception if you send feedback such as a thumbs up or down (Anthropic privacy centre).
- Gemini in Google Workspace: Google says Workspace does not use customer data to train models without the customer's permission or instruction (Google Workspace privacy hub).
So the honest sentence "our provider doesn't train on our data" is only true if the work happens in the business account. If half the team uses a personal ChatGPT Plus account, fix that first. That's usually the most useful thing this exercise turns up.
These pages change, so link to them and write down the date you read them.
Before you send it: a short checklist
- Ask the team, without blame, which AI tools they actually use, including on their phones. You can't describe what you don't know about.
- Move client work onto one business account. It changes the default data terms, and it lets you remove access when someone leaves.
- Name one person who approves new tools and owns the page.
- Write the staff rules (what never goes in, a person checks what goes out). Our one-page AI policy is a template for that.
- Hold a short team session and write down the date. In the EU, the AI Act's "AI literacy" rule already asks firms that use AI to help their staff use it knowingly, and a dated record is a sensible way to show you did.
- If personal data is involved, check what your data protection rules ask of you (in the UK and EU, that can mean a short risk check before you start). If it's high-stakes, ask a professional.
- Put a review date six months out.
Don't say "we're compliant"
Two words to keep off the page: "compliant" and "guaranteed". You can say what you do, what you don't do, and who's responsible. You can't promise that no mistake will ever happen, and a careful client will trust the plain version more.
If you'd rather not do it alone
This is the kind of job we do in a week. The AI Check-up ($149, credited if you go further) is a 60-minute call where we look at which tools your team uses and give you a one-page plan. If your clients are already asking, our AI Supplier Pack ($450, added to a Team Workshop or Team Set-up) gives you the client-ready statement, a staff policy, a dated training record and a staff quick-card, plus 30 days of follow-up questions by email. It's training and paperwork, not legal advice, and the tools stay in your own accounts.
Not sure where AI would save your business time?
A 60-minute call and a one-page plan naming the 3 jobs in your business worth automating first. Taken off the price of any build booked within 30 days.
Related Articles

A One-Page AI Policy for a Small Team (Copy It, Then Change It)
Your staff are probably already using ChatGPT, Gemini or Claude for work, often on personal accounts. A short, plain AI policy a small business can copy today, what each line is for, and what the EU AI Act's 'AI literacy' rule actually asks of a small firm.

Does Your Website Chatbot Have to Say It's AI? Yes. Here's the Wording.
Since 2 August 2026, EU rules say people must be told when they're talking to an AI. What that means for a small business with a chat widget or an AI phone line, who is responsible, and short wording you can paste in today.

The Invoice Looks Perfect. That's the Problem.
AI has removed the spelling mistakes and odd phrasing that used to give scam emails away. What the FBI's latest figures show, and the simple office habits that protect a small business when a fake invoice, email or voice message looks completely real.

