Telling Clients You Use AI on Their Work

A client is paying for an outcome, and how you reach it has always been partly your business. That is why nobody discloses which template they started from or which colleague they asked. AI changes the calculation because of scale, because of where the data goes, and because the time it saves is time the client may believe they are paying for.

Three separate questions get collapsed into one and then answered badly. Whether to tell them, what happens to their data, and what happens to your rate. They have different answers.

When telling them is required rather than polite

Disclosure stops being a courtesy and becomes an obligation at three points.

When their confidential information leaves your control. Pasting a client's strategy document, customer list or financials into a third-party tool is a disclosure of their information to that vendor, whether or not the vendor does anything with it. Most consulting agreements contain a confidentiality clause that names who you may share information with. Read yours before deciding this is a judgement call, because it may already have been decided.

When your agreement says so. Master services agreements increasingly carry an AI clause. Some require notice, some require written consent, some prohibit it for defined categories of work.

When the finished work would mislead them about where it came from. A research summary presented as your analysis, where the analysis was generated and lightly edited, misrepresents what they bought. The test is not whether a tool was involved. It is whether the client's belief about what they are getting would change if they knew.

Outside those three, it is a judgement call about the relationship rather than a rule.

What to actually say

The failure mode is a paragraph of hedging that makes routine tooling sound like a risk you are confessing to. Short and specific works better, because it sounds like someone who has thought about it.

For a proposal or engagement letter:

I use AI tools in parts of my process, including research synthesis and first drafts. Your confidential information is not entered into any tool that trains on it, and everything I deliver is reviewed and edited by me. If you would prefer I not use these tools on your work, tell me and I will not.

Three things are doing the work there. It names where the tools are used rather than gesturing at "our process". It makes a specific claim about data handling that you have to be able to stand behind. And it hands the decision back, which is what turns disclosure into consent.

The last sentence is the one people want to remove. Leave it. A client who says no has told you something valuable about how to work with them, and a client who says nothing has now agreed rather than just been informed.

A clause is not a consent

Two mechanisms get confused because both live in the same document.

An AI clause in your engagement letter establishes that you may use these tools. It is a term of business.

A recording consent is separate law, not contract, and it applies the moment you record a client call to run a notetaker over it. It turns on where the participants are rather than where you are, and several states require every party to agree rather than just one. A signed agreement permitting AI use does not satisfy it.

The regulated version of this distinction, written for investment advisers but structurally identical for any consultant recording client conversations, is in the AI clause in a client engagement letter.

The rate question

This is the one people actually worry about, and it only bites if you bill hours.

If a piece of work took ten hours and now takes two, hourly billing means a pay cut for getting better at your job. That is not an AI problem. It is a pre-existing problem with hourly billing that AI made impossible to ignore, and it points at the same answer it always did.

Price the outcome. A market analysis is worth what the decision it informs is worth. It was never worth the hours.

Where hourly is contractual, be straight about the reduction. A consultant who bills the old ten without saying so is running a risk that scales with how good the tools get and how obvious that becomes. A consultant who says the work now takes two and prices accordingly has a conversation about scope instead, which is a better conversation.

The compressed time is the asset. Eight hours back is eight hours for the part a client cannot buy elsewhere, which is the part where you tell them the thing they did not want to hear. More of that is worth more, and it is worth more per hour.

What to keep out entirely

Independent of tier or vendor, some categories do not belong in a general-purpose tool on a client engagement:

• Material non-public information, which carries securities exposure separate from any confidentiality term
• Personal data about identifiable individuals, which brings the client's own privacy obligations into your process
• Anything under a protective order, an NDA naming permitted recipients, or a regulator's document hold
• Credentials and access, a security question and the more common breach

The rest is a judgement call informed by what tier you are on and what its terms say.

The check worth doing once

Before the next engagement, three things. Read the confidentiality clause in your standard agreement and see whether it already answers this. Confirm which tier of each tool you are on and what it says about training, because the consumer and business tiers of the same product frequently differ. And write your two sentences of disclosure once, so you are not composing them under pressure in a proposal.

That is an hour, and it converts a recurring anxiety into a settled position you can state clearly when a client asks.