Should I Tell Clients I Use AI? A Clear Answer

Should you tell clients you use AI? In three situations you have no choice, and outside those three it is a positioning decision rather than an ethical one. Most of the anxiety around this question comes from treating it as a single moral judgement when it is really two separate questions: what are you contractually obliged to disclose, and what do you gain or lose by volunteering more than that.

Take them in order, because the first one is settled by documents you can go and read today.

The three cases where disclosure is not optional

1. Your contract has a confidentiality or subprocessor clause

This is the one that catches freelancers and small agencies most often, and it is not about AI specifically. Standard client agreements contain language restricting who else may see the client's material. Pasting a client's unreleased financials, customer list or source code into a third-party service is disclosure to a third party, whether or not that service is an AI tool.

The clauses to look for, by name:

• Confidentiality: often phrased as "shall not disclose to any third party", sometimes with a carve-out for "employees and contractors with a need to know". A model provider is neither.

• Subprocessors or subcontractors: usually in the data-processing addendum. Many require written approval, or at least notification, before you add one.

• Data location or residency: if the contract specifies where data may be processed, a provider running in another region breaks it independently of anything AI-related.

If any of those apply, using the tool without approval is a breach, and the disclosure conversation is not optional. It also has a clean solution: ask for the tool to be added to the approved list, in writing, once. Then it is covered for every future project with that client.

2. The client is in a regulated sector

Healthcare, legal, financial services, and anyone handling children's data operate under rules that reach through to their vendors. Their compliance team will have a position on AI tooling, and it may be more permissive than you expect or considerably less. Either way, finding out after delivery is the bad version.

For European clients there is an additional layer if what you are building makes consequential decisions about people. That is a use-case classification question rather than a disclosure one, and we cover it in the EU AI Act's transparency rules.

3. You are selling human expertise as the product

If the engagement is explicitly "a senior engineer will review your architecture" or "a copywriter will write your brand voice", and the deliverable is substantially model-generated, the mismatch is a misrepresentation. Not because using the tool is wrong, but because the client bought a specific thing.

The line is what the client believes they are paying for. A client paying for an outcome has no reasonable expectation about your keyboard. A client paying for a named person's judgement does.

Outside those three, it is a positioning decision

Here is where most of the hand-wringing happens, and where the honest answer is: it depends what business you want.

Clients do not hire you for your tools. Nobody asks a designer which version of Illustrator they run, or a developer whether they use autocomplete. Tools are how the work gets done, and the deliverable is what was bought. By that logic, unprompted disclosure is unnecessary.

But that logic has an edge, and the edge is judgement. What clients are actually buying from you is that someone competent decided the output was right. If you are reviewing, correcting and standing behind everything you deliver, you are doing the job regardless of what drafted the first version. If you are forwarding output you have not evaluated, you are not doing the job, and no disclosure policy fixes that.

So the useful reframing: you owe the client accountability, not a tooling inventory.

The case for saying nothing

Some clients have an unexamined negative reaction to "AI" as a word, formed from bad experiences with obviously generated content. Volunteering it invites a conversation about method when you would rather have one about results. If your work is good and you stand behind it, staying quiet is defensible.

The case for being open about it

Three real advantages, in rough order of how much they matter:

• It ends the pricing argument early. Clients who suspect AI involvement and were not told tend to raise it at invoice time, framed as "why did this cost that much". Clients who knew from the start engaged on that basis.

• It is a differentiator right now. "I use these tools, here is how I check the output, here is what I will not use them for" is a more sophisticated pitch than either silence or enthusiasm. It signals that you have thought about it.

• It cannot be used against you later. Disclosure is only awkward before it happens. Discovery is worse, and the damage is disproportionate to the underlying facts.

For freelancers pricing this work, how much to charge for an AI automation project covers the related question of whether faster delivery should mean lower invoices. It generally should not, for reasons worth understanding before this conversation comes up.

What to actually say

If you decide to disclose, do it once, in your proposal, in about three sentences. A policy stated up front is professional. The same information volunteered defensively after a question is not.

Something on this pattern:

> I use AI-assisted tooling in parts of my process, primarily for drafting and research. Everything I deliver is reviewed and edited by me, and I am accountable for it. I do not put client-confidential material into third-party tools without your written approval.

That covers the three things a reasonable client wants to know: that you use it, that you check the output, and that their data is not going somewhere they did not agree to. It takes a paragraph and closes the topic.

Two things to avoid. Do not overclaim expertise you do not have because the tools made a deliverable look sophisticated, and do not attach the words to a specific product roster you will change in two months.

What clients are actually worried about

Worth naming, because addressing the real concern is more effective than addressing the stated one. When a client asks "did you use AI for this", they are rarely making a philosophical objection. They are asking one of three things:

1. Am I paying senior rates for something automated? Answer with what you did: the review, the corrections, the decisions. That is the work.

2. Is my confidential information now in a training set? Answer with your actual practice on data, which is why having one matters. Is it safe to give AI access to my data covers the mechanics.

3. Will this be generic? Answer by pointing at the specifics in the deliverable that only come from understanding their situation.

Each of those has a concrete answer. None of them is resolved by a disclosure policy alone.

FAQ

Do I legally have to tell clients I used AI?

In most jurisdictions there is no general law requiring it for freelance or agency work. Your obligation usually comes from your contract, particularly confidentiality and subprocessor clauses, and from sector-specific regulation if the client is in one.

What if a client asks directly?

Answer honestly and immediately. A straight answer about your process and your review standard costs you very little. Evasion, if it later becomes apparent, costs the relationship.

Should I charge less because AI made the work faster?

Not automatically. You are being paid for the outcome and the judgement, not for hours. If you have committed to hourly billing, bill the hours you worked and let the efficiency show up as capacity for more clients.

Can I put a client's confidential data into an AI tool?

Not without checking the contract, and usually not without written approval where a subprocessor clause exists. Getting a tool added to an approved list once is far easier than justifying it afterwards.

Does disclosing make me look less skilled?

The opposite, when it is paired with a clear account of how you check the output. Discomfort with the topic reads as inexperience. A settled position on it reads as professionalism.

Related: proving AI-generated work is correct to a client