Should you tell clients you use AI? A 2026 disclosure guide

Yes, tell your clients you use AI, and put it in the contract instead of confessing mid-project. Disclosure costs a little trust now. Silence costs more once a client policy, a platform rule or a regulation asks the question for you.

Does disclosing AI lower trust?

Yes, a little, and pretending otherwise would be dishonest. A 2025 paper in the journal Organizational Behavior and Human Decision Processes, by Oliver Schilke and Martin Reimann, found that people who admit using AI are trusted less than people who say nothing, even when the work is good. The authors call it a transparency dilemma.

That sounds like an argument for keeping quiet. It is not, because the penalty only matters while disclosure is optional. Once a client policy or a rule requires it, you are not choosing between disclosing and staying silent. You are choosing between disclosing and getting caught.

Getting caught is the expensive outcome. A client who learns later that you ran their confidential brief through a public chatbot, or delivered "hand-crafted" copy that was mostly generated, does not file it under minor. They file it under "lied to me". A small trust dip you control is cheaper than a large one you do not.

There is also a direction of travel. In an expert panel run by MIT Sloan Management Review, 84% agreed that companies should be required to disclose AI use to customers, according to the published results. Norms are moving toward disclosure as the default.

When disclosure is required, not just polite

Disclosure becomes an obligation in three common situations: a regulation covers your output, the client has an AI policy, or your confidentiality terms limit how you use the tool. Plenty of service providers hit at least one without noticing.

The regulatory case is the EU AI Act. Its transparency rules in Article 50 apply from 2 August 2026, according to the European Commission. Providers of generative AI systems must mark outputs in a machine-readable format. People who deploy those systems must label deepfakes and must disclose AI-generated text that is published to inform the public on matters of public interest, unless a person reviewed it and carries editorial responsibility. The Commission notes that its code of practice is voluntary, but the Article 50 requirements are legal obligations. The full text is on the AI Act Article 50 page.

Most client work will not fall under those rules. But if you produce images, video or published copy for EU audiences, put the date in your calendar and ask the client how they plan to label things.

The second case is the client's own policy. Many companies restrict generative tools internally. If a client bans them for staff, assume they expect the same from the contractor who now holds their data. Ask early, in writing.

The third is the platform you sell on. Marketplaces publish their own rules. Fiverr, for example, keeps a community standard on AI-generated content, so read the current version for any platform where you take orders, and do not market work as hand-made if it was not.

What about the NDA you already signed?

The quiet risk is confidentiality. Most NDAs forbid sharing a client's confidential information with third parties, and a public AI tool is a third party. Paste a brief, source code or a customer list into a consumer chatbot and you may have handed protected material to a service that can retain it.

So there are two questions, and people merge them. One is "did I use AI?", which is about credit and process. The other is "did I put the client's protected data into an outside system?", which is about a possible breach of contract. You can answer the first honestly and still be in trouble over the second.

Read the confidentiality clause in each agreement before you use a tool on that client's material. If it is silent, ask. A one line email that says "I plan to use an AI assistant for first drafts, and I will not enter your confidential data into public tools. Is that fine?" settles it for both of you. Our guide to anti-AI contract clauses covers what to do when the client forbids AI outright.

How to write a short AI-use clause

A good clause says what you use AI for, what you do not do, who is responsible for the result, and how client data is protected. Keep it boring. A clause that reads like a lawyer trying to scare someone makes clients nervous, and one that reads like a normal operating note relaxes them.

Here is a plain version you can adapt:

The service provider may use AI-assisted tools for tasks such as research, drafting and editing. A person reviews and is responsible for all final work. The service provider will not enter the client's confidential information into any public or consumer AI service without written approval. If the client requires that no generative AI be used on any part of this project, the client will say so in writing before work begins, and the service provider will comply.

Four things make it work. It limits AI to assisting tasks, so you are not claiming a machine wrote the thing. It names a person as responsible, which is the ambiguity clients fear. It carves out confidential data, which protects your NDA. And it gives the client a clean way to opt out.

If your deliverable falls under the EU rules above, add a line saying that AI-generated output will be marked as the law requires. This is a template, not legal advice, so check it with your lawyer before you rely on it.

How to bring it up without scaring the client

Say it as a statement about quality and accountability, not an apology. The trust penalty comes from ambiguity, the sense that nobody is clearly in charge of the output. Remove the ambiguity and you remove most of the penalty.

A sentence that works: "I use AI tools to move faster on research and first drafts, and I review and stand behind everything I deliver. Your confidential material never goes into a public tool." It names the benefit to the client, which is speed. It names the person who is accountable, which is you. And it answers the confidentiality worry before they raise it.

Do not list every model you touch, and do not promise the work is free of machine help if that is not true. Overclaiming is its own trap. The MIT Sloan piece makes a related point about customers in general: disclosures should be clear and easy to read, not buried. Say it once, in plain words, in the place the client will actually look.

Timing matters too. Raise it in the proposal or at kickoff, before any work exists. A client who hears it on day one treats it as your method. A client who hears it on day thirty treats it as a confession.

Where the clause should live

In the contract, not in a chat message. A term that lives in a messaging thread is a term you will struggle to point to when a dispute starts. Put it where both sides signed, and keep a copy you can produce.

Many service providers slip here. The clause is fine, the client agreed, and six months later the signed copy is an email attachment, the delivered files are in a share you no longer control, and the brief that shows what data you received is gone. If a question comes up, you want the agreement and the delivery in one place with timestamps.

How Delivvo keeps the agreement and the delivery together

Delivvo puts your AI-use terms inside the contract you send, then keeps the signed copy and the delivered files in the same client portal.

The flow is short:

1. Write your contract with the AI-use clause in it. Every clause is editable, and Business plans and above can upload their own contract PDF and place signature boxes on it.
2. Send it from the client record. The client signs by typing or drawing their signature.
3. Delivvo stores a signing record with a SHA-256 hash, the IP address, the device and an archived PDF, so you can show what was agreed and when.
4. You deliver through the client portal, which opens with a one-time code. Files, versions and approvals sit next to the signed contract.
5. The client approves the work inside the portal, so the record shows who accepted which version.

Say a client asks six months later what you agreed about AI. You open one record and show the clause, the signature, the date and the files delivered against it, without hunting through inboxes. That also helps if you work with a small team, because everyone sees the same terms. The detail a generic e-signature tool does not offer is that the contract, the files and the approvals share one client record, and every document carries a verification code that anyone can check. Clients pay you directly through your own gateway, and Delivvo takes 0%. Read more about contracts and e-signatures and the client portal. Delivvo provides templates and records, not legal advice, so check your wording with a lawyer.

Read more about Delivvo for teams, or start with the CRM and client portal.

FAQ