How-To Guide

How to Answer Client AI Due Diligence Questions Without Losing the Deal

Procurement teams now ask suppliers how they use AI — here is what they are really checking, and how to answer in a way that builds trust instead of raising flags.

B Biztrategy Published 6 November 2026 · 8 min read
A woman signing a document with a pen on a desk, emphasizing hands and writing.

Something changed quietly this year. If you sell services to anyone bigger than you — a mid-market client, a public body, a company with a real procurement function — the forms you get before contract signature have a new section in them. Sometimes two questions, sometimes a 40-row spreadsheet. The heading says “Artificial Intelligence and Automated Processing”, and it wants to know what you feed into which tools, and what happens to the client's data once you do.

Most small suppliers handle this badly, and not because they have anything to hide. They answer from instinct under time pressure, and both instincts are wrong. The first is to downplay — “we don't really use AI” — which is rarely true and falls apart the moment a colleague mentions a transcription tool on a call. The second is to overclaim: a paragraph about your “AI governance framework” invented while filling in the form, which you cannot evidence if asked.

Neither loses the deal on its own. What loses the deal is the follow-up — a procurement officer who now has to ask three clarifying questions, escalate to legal, and chase a document you do not have. You become the slow supplier. Below are the nine questions that actually turn up and how to answer each one. Prepare them once and the next questionnaire takes twenty minutes, not two days.

Why clients started asking

It helps to know who is on the other side of the form, because it changes your tone. The AI section is almost never someone trying to catch you out. It comes from one of three places: a compliance obligation of their own, where a client subject to the EU AI Act or to sector rules has to record whether automated processing touches their data; a requirement passed down from their own client, which means your answer must survive being copied into someone else's document without you there to explain it; or a bad experience, where somebody pasted a confidential document into a chatbot and now there is a policy. That third kind is blunter, and rewards specificity more than any other.

In all three cases the reader wants one signal above all others: does this supplier know what their own tools do? An imperfect practice described precisely beats a perfect practice described vaguely, every time.

The nine questions that actually get asked

Across the questionnaires circulating in 2026, the AI section reduces to these. Write your answers once, keep them in one file, paste from it.

1. Do you use AI or automated tools in delivering services to us? Answer yes if you use anything at all — a writing assistant, transcription, a coding assistant, the AI features already in your CRM. Then name the categories: “Yes. We use AI assistants for drafting and research, meeting transcription, and code completion. We do not use AI to make decisions about your customers, or to generate final deliverables without human review.” That second sentence is what relaxes them.

2. Which specific tools and providers? They want names and vendors, not categories. Keep a current list: tool, vendor, purpose, plan tier. If you cannot produce it in five minutes, that is the real finding.

3. Is our data used to train the provider's models? The most important question on the form. For most paid business plans the honest answer is no — but check per tool rather than assuming; free and consumer tiers are where this goes wrong. “No. All tools in use are on business plans where customer content is excluded from model training by contract. We review this at each renewal.”

4. Where is the data processed and stored, and for how long? Retention is usually the gap — most people have no idea how long their transcription tool keeps recordings. Go and look. “Processed in the EU; transcripts retained 30 days then deleted” is strong; “we will confirm retention per tool within five working days” is acceptable. A guess that turns out wrong is the only unacceptable option.

5. What human review happens before work reaches us? Describe the actual step, with a named role: “Every AI-assisted draft is reviewed and edited by the account lead before delivery; AI output is never sent to a client unread.” Buyers distrust “human-in-the-loop” now — it means anything from careful editing to a glance. Say what the human does.

6. Do you have a written AI policy, and can we see it? Many questionnaires now ask for the document, not just whether it exists. This is the question that most often stalls a deal for a week, because the answer is “yes” and then nobody can find the file. If you do not have one, our guide on how to write an AI policy for a small business gets you to a credible two-page document in an afternoon.

7. How do you stop confidential client information being entered into AI tools? They are asking about the gap between your policy and your team's behaviour. Strong answers describe a control, not an intention: an approved-tools list, business plans with training off, a rule about what never gets pasted anywhere, an onboarding step where new staff are told. For the practices behind the wording, see AI data security for small business.

8. Will you tell us when AI has been used on our work? Say yes and define the threshold, because a bare “yes” invites a contract line requiring you to flag every spell-check. “We disclose AI use at engagement level in our proposal, and will flag any deliverable where AI generated substantial content.” Our piece on telling clients you use AI works through where to draw that line.

9. Who is liable if AI-generated output causes a problem? The only correct answer is that your professional obligations are unchanged: you are responsible for your deliverables however they were produced. Do not carve out an AI exception in a questionnaire — it reads as shedding responsibility and will escalate the review. Liability belongs in the contract, and AI clauses for client contracts covers the clauses worth negotiating.

The one-page disclosure that saves you hours

Write all nine answers into one document and stop reinventing them. Call it an AI Use Disclosure, keep it to one page, and date it. The structure that works: two sentences on what you use AI for and what you never use it for; a table of tools with vendor, purpose, plan tier and training status; three bullets on data handling — where it goes, how long it stays, who can see it; one paragraph on human review naming the role that signs off; a line pointing to your full AI policy; a revision date and an owner's name.

Two things make this pay for itself. You can attach it instead of retyping, and a dated, owned document signals maturity in a way filled-in cells never do. And you can send it unprompted with a proposal — plenty of buyers have the AI question on their mind and no process for asking it, so answering first removes a silent objection. Review it quarterly: a disclosure dated fourteen months ago tells the reader your governance is theatre.

The answers that cost people deals

“We don't use AI.” Almost never true in 2026, and easy to disprove — your proposal document, your meeting notes, your developer's editor, something in your stack has AI in it. Being caught in a technically false “no” turns a routine form into a trust problem.

Listing a tool you cannot describe. Naming seven tools looks thorough until someone asks what the fourth does with recordings. A short list is not a weakness — auditing your AI tool stack and cutting what nobody uses improves both your answer and your costs.

Copying a large company's wording. A three-person consultancy describing its “AI Governance Board” invites exactly the evidence request you cannot satisfy. “The two directors approve any new tool before it touches client work” is credible governance at your size.

Promising a certification you do not have. ISO 42001, SOC 2 with an AI scope — if it is in progress, say so with a date; if not, say not currently held. Buyers accept both. They do not accept discovering a claim was aspirational. And answer fast: a questionnaire returned in two days with one “we will confirm within five working days” beats a complete one returned in three weeks.

When the honest answer is “not yet”

You will meet questions where your real answer is that you have not got there yet. Nobody has a complete picture of every vendor's sub-processors. The pattern that keeps deals moving: state the current position plainly, state what you do instead, give a date.

“We do not currently hold a formal AI risk assessment per tool. New tools are approved by the managing director against our approved-use list before client data is involved. A documented assessment is scheduled for Q1 and we can share it when complete.”

That answer rarely costs anyone a contract. What costs contracts is a blank cell, or a sentence that sounds like a yes and means nothing. Procurement teams are used to suppliers being partway through — their own employer almost certainly is. They are not used to suppliers being specific about it.

One caveat: if the work involves automated decision-making about individuals — screening candidates, scoring customers, triaging applications — the questions get far more serious, “not yet” stops working, and the EU AI Act puts you in a higher-risk category that deserves proper preparation.

Turning the questionnaire into an advantage

Here is the part most suppliers miss. The AI section exists because your buyer is uncertain, and uncertainty is usually where small suppliers lose to bigger ones. This is the rare case where it runs the other way: large agencies answer these forms through a compliance team that has never met your buyer, so it reads like boilerplate. You can answer in two paragraphs, name the person who signs off, and offer a fifteen-minute call.

So: attach your disclosure to proposals before anyone asks; offer the call, because almost nobody takes it and offering it is itself the reassurance; and when a questionnaire reveals a gap, fix it and tell them you did. Clients increasingly ask adjacent questions in the same breath — sustainability is the common one, and our piece on AI's carbon footprint for small businesses has the two-sentence answer for that part of the form.

The bottom line

AI due diligence questions are not a compliance tax. They are a cheap test of whether you know how your own business works, set by someone deciding whether to trust you with theirs. The suppliers who struggle answer from memory at 11pm the day before the deadline. The ones who sail through have a one-page document, nine prepared answers, an honest “not yet” where it applies, and a date against it. Write the nine answers this week, while no deal depends on them.

Where does your business stand on AI?

Take the free 3-minute AI Readiness Quiz and get a personalised score with your next steps.

Take the Free Quiz →