The governance wedge · Article 7 of 7
“How do you use AI?” is now a normal client question
The firms that dread the question are the ones without an answer. A prepared, consistent disclosure turns AI from something you hope nobody raises into evidence that the firm runs on decisions.
1 · The question is coming either way
It arrives in different costumes. A security questionnaire with a section that was not there last year. An RFP line asking you to describe any use of generative AI in service delivery. A general counsel who read something over the weekend and wants twenty minutes. A longtime client, slightly embarrassed to be asking, wondering whether the AI thing touches their account.
Different costumes, one question underneath: do you know what your own firm is doing, and will you say it plainly?
Most firms experience this as a threat, and their answers show it. A defensive line about taking data security very seriously. A hedge. Or the outright disastrous claim that the firm does not use AI, delivered while half the staff quietly does. This article starts from the opposite premise, which is that the question is an opening. You are being invited to demonstrate, at the exact moment your competitors are flinching, that your firm runs on decisions rather than drift. Very few marketing opportunities arrive hand-delivered by the customer. This one does.
2 · Candor is a trust strategy
The instinct to downplay AI use rests on an assumption that clients punish it. What the evidence we can see suggests is something more interesting: people punish unaccountability, and they reward a named human standing behind the work.
One data point, offered with its caveats attached. A survey of AI-influenced shoppers released in August 2026 found that people who trust AI recommendations lean harder on human trust signals rather than less, with 44% wanting reviews before acting on an AI recommendation, and disclosure of sponsorship increasing trust among a majority of the heaviest AI users. It is a directional consumer study, not quota-balanced, and not about professional services at all. We cite it because its shape matches what we hear in client conversations: AI can get a firm considered, and only proof, process, and people get it trusted.
Disclosure done well is not an admission that a machine did the work. It is the demonstration that humans run the machine.
The logic runs the other way too, and harder. Concealment has a failure mode, and the failure mode is total. A client who discovers undisclosed AI use, however careful that use was, will wonder from then on what else went unmentioned. You cannot un-ring that bell. A firm that said it first, in writing, before being asked, has removed the entire category of discovery from its relationships.
3 · One answer, not twelve
The operational core of this domain is consistency. Your firm needs one prepared answer: short, true, and identical from every partner. Improvised answers diverge, and divergence reads to a client as concealment even when it is only disorganization.
The prepared answer carries five load-bearing parts: what you use AI for, what you never use it for, whose accounts it runs on, who reviews the output, and who holds the accountability when something goes wrong. Here is the shape, in words you are welcome to take and adapt:
We use AI-assisted tools for defined categories of work, such as research synthesis and first drafts, under a written, risk-tiered policy. Client work runs only on business-grade accounts with model training disabled and contractual data protections; certain categories of sensitive work are excluded from AI use entirely. Every AI-assisted deliverable is reviewed by a qualified professional before it reaches you, and accountability for the work rests with the named people who sign it, exactly as it did before these tools existed. Our policy summary is available on request.
Notice what makes that paragraph possible. Every sentence in it is a receipt from an earlier article in this series: the inventory gives you the first, the policy the second, the data terms the third, the review gates the fourth. Disclosure is not a writing problem. It is the visible surface of governance you either did or did not do, and a firm that tries to write the paragraph without the system underneath produces something lawyers call aspirational and the rest of us call fiction. It gets tested by the first follow-up question.
The last piece of consistency is repetition infrastructure. The answer lives in one place, staff know where that is, and the client-facing people have seen the likely follow-ups. Article three answers the one on data, article six the one on review, and your tier list answers scope by name.
4 · When disclosure stops being optional
Around the prepared answer, map the places where disclosure is required rather than merely wise.
Contracts first. Some client agreements and RFPs now require proactive notice of generative AI use, and the contract register from article four should flag every one of them, because on those accounts a question that feels voluntary is a compliance event.
Sector rules next. Publishers face submission and licensing contexts with explicit AI disclosure expectations, and regulated professions are watching their own bodies put oversight duties into writing, as the American Bar Association did in July 2024.
Public statements last, because they are the disclosure surface firms forget they own: the website paragraph about innovation, the partner’s conference quote, the proposal boilerplate that nobody has reread since 2024. Our checklist compresses this into a single line, which is that your public statements about AI should match what you actually do. The gap between the two is discoverable, and somebody is now paid to discover it.
☐ We have ready language for when a client asks whether and how we use AI.
☐ Our disclosure position is consistent across the organization, not improvised per person.
☐ We know which clients or contracts require us to disclose proactively.
☐ Our public statements about AI match what we actually do.
Where this closes the loop
This is the last article in the governance series, and the last of the checklist walk, deliberately, because disclosure is where the quiet internal work becomes visible and starts paying for itself. An inventory, a tiered policy, clean data terms, checked contracts, real review gates: a client sees none of those. What a client sees is a confident, consistent, two-paragraph answer to a hard question. That is the moment governance converts from cost into reputation.
Our stake in this is simply stated. Client-facing transparency language is a named deliverable of our AI Transformation practice, built on the assessment that makes the words true. If your firm already has the words and the system beneath them, you are the firm this series was cheering for, and we would be glad to have argued you into nothing.