Should your agency tell clients it used AI?

I write this for the owner or head of a small agency, two to thirty people, whose team now uses AI image, video, voice and writing tools. The question "should we disclose?" sounds simple, but most advice mixes up three things that have different answers. Once you separate them, the policy is easy to write. This is general information, not legal advice. Check local law, your client contract and each platform's current rules.

Three different questions hidden in "should we disclose?"

1. Do we tell the client? This is about trust and the contract between you and them.
2. Do we label the work for its audience? This is decided by the ad platform, and in some places by law.
3. What do we promise in the contract? This covers ownership, uniqueness and who carries risk. See what to put in your contract when you use AI tools.

Pages that blur these give bad advice, such as "label everything" (not needed in many cases) or "no law requires it, so say nothing" (which ignores the client, the contract and the platform). Take them one at a time.

When the law or a platform decides for you

These are the rules I could confirm from official or primary pages. For the rest, I tell you to check the current page, not to rely on a summary.

• YouTube. Its help page requires creators to disclose realistic content that was generated or meaningfully altered with AI, for example a real person appearing to say or do something they did not, or a realistic scene that did not happen. Non-realistic content and minor edits (filters, upscaling, AI help with scripts or captions) are exempt. It says repeated failure to disclose can lead to removal or suspension from the YouTube Partner Program.
• EU AI Act, Article 50. The transparency article says people who use AI to make deepfakes (realistic fake images, audio or video) must disclose that the content was artificially generated or manipulated, clearly and at the latest when people first see it, with a lighter duty for clearly artistic, creative, satirical or fictional work (disclosure in a way that does not spoil the viewing). Providers of AI systems must also mark outputs in a machine-readable way. Article 50 applies from 2 August 2026. The 2026 Digital Omnibus (Regulation (EU) 2026/1744, published 24 July 2026) did not delay the duties for people who deploy AI to make content. It only gives providers of generative systems already on the market until 2 December 2026 for the machine-readable marking duty, which falls on the tool makers, not on you. This is the EU's AI law, and it can reach agencies making ads for EU audiences. This is information, not legal advice; the Official Journal text is the source to check.
• Fake reviews in the US. The Federal Trade Commission's final rule on fake reviews and testimonials covers fake reviews, including AI-generated ones that misrepresent who wrote them, and allows civil penalties against those who knowingly break it. Practical meaning: never script an AI presenter as a real customer. For the full treatment of AI avatars in ads, see AI avatars versus real creators in UGC-style ads.
• Meta and TikTok. Both publish rules on labelling AI-made or AI-altered content and these change often. They are not summarized here because they change often. Read the current policy page of each platform before launch, and see short-form video ads for clients for the production side.
• Copyright. Not a disclosure rule, but related: the US Copyright Office has said prompts alone do not make someone an author. See its AI page for the reports. Tell clients plainly that raw AI output may not be protectable.

Also note that some tools' own terms require disclosure. The Higgsfield terms, for example, ask users to disclose AI generation where law requires and not to represent output as human-made. Read each tool's terms for yours.

When it is good practice: the materiality test

Where no law forces your hand, a simple test works. Writers' groups and agency policies point to the same idea. The UK's Professional Copywriters' Association guidance puts it as: would the client's view of the work's value, or how they use or publish it, change if they knew exactly how AI was used? If yes, or if you are unsure, disclose. The agency Ignite Social Media describes a similar rule in how they built their AI policy: disclose up front when AI materially shaped the work (imagery, data conclusions, concepts), do not routinely disclose built-in features such as grammar checks, bid optimization or resizing, and answer fully whenever asked.

My own addition: the real cost of silence is not the law, it is the day the client finds out. A client who learns from a vendor, a forum thread or an image-check tool that the hero image was AI-made, when you never mentioned it, starts doubting everything else you made. Telling them first costs one sentence.

What to tell the client, by kind of AI use

What you did	Tell the client?	Also check
Spell-check, resizing, background tools built into your editor, transcription	No routine notice. Answer honestly if asked.	Client's contract, if it bans "any AI"
AI used for research, brainstorming, internal drafts that a person rewrote	Mention in your process summary or Tool List	Client confidentiality
AI-generated images, video, music or voice in the final work	Yes, before approval	Ad platform label rules, tool terms
A realistic synthetic person, voice, or a real person's likeness	Yes, and get written consent from the person	Platform label rules, local law, tool terms
Client confidential material going into a tool	Ask permission first	Whether the tool may train on inputs; see AI tools and training on your data
Testimonial or review style content	Never use an AI presenter as a real customer	FTC rule on fake reviews, platform ad policies

This table matches the materiality idea in the sources above, but the lines are a judgment, not a legal test. Move them to suit your clients.

Check the client's contract first

Before you use any tool on a client's job, search their master agreement, brand guidelines and any supplier terms for the words "artificial intelligence", "generative", "machine" and "automated". Digiday has reported that clients are adding no-AI clauses, with the general counsel of one creator marketing company saying requests for them rose some 30 percent or more in new advertiser agreements (a figure from that one company, so treat it as a signal). If you find such a clause:

• Ask in writing what it covers. Does it include grammar tools, resizing, upscaling, automatic captions?
• Record the answer in the project file.
• If the ban is broad and you want to use AI, ask for a written exception for named uses. Do not quietly work around it.

Add an intake question to onboarding: "Do you have any restrictions on how we use AI tools on your work?" It takes ten seconds and prevents a dispute.

A short AI use policy you can copy

Put this in your proposal or master agreement, adjusting the details. Written fresh for this page, not legal text:

How we use AI. We use AI tools for some steps, such as drafting concepts, creating image and video variants, and quality checks. A named person on our team reviews every deliverable before it reaches you. We keep a list of the tools we use on your project and will share it on request.

What we tell you. If AI-generated imagery, video, voice or copy forms a material part of the final work, we will tell you before you approve it. If a platform or law requires the ad to carry an AI label, we will tell you which and why.

Your material. We do not enter your confidential information into AI tools without your agreement. If a tool's terms allow its provider to train on submitted content, we will pick a setting that turns this off, or ask you first.

Your say. You may restrict AI use for particular deliverables by telling us before work starts.

Pair it with a contract clause on ownership and uniqueness, since you should not promise that AI output is exclusive; see the clause wording in the contract guide linked above.

What to say when a client asks

Answer in three parts: what, who checked, what you did to protect them. Scripts you can adapt:

Plain question. "Yes. We generated the background with an AI image tool, and our designer built the layout, fixed the text and checked the logo against your brand file. Your approved version is on file."

Worried client. "Good question. We use AI for first drafts and variants. Everything goes through a named reviewer, we never put your confidential material into a tool without asking, and I can send you our tool list. If you would prefer no AI on this project, tell me and I will scope it without."

Suspicious client. "I can show you the process. Here is the brief, the options we generated, what we rejected and why, and the final edit. I would rather you see it than guess."

Price question. "AI helps us make drafts faster. You are paying for the direction, the checking and the responsibility for what goes live." The pricing side is in how to price AI-made creative.

Never say "no AI was used" unless it is true, and never say the work is 100 percent original. Those are the two sentences that come back to hurt you.

What not to promise

• Uniqueness. Some AI tools' terms say outputs may not be unique across users.
• Copyright. Raw AI output may not be protectable in some countries.
• Zero training. Unless you control the tool settings or have a contract that guarantees it, do not promise that no provider will ever use client content to train models.
• Labelling everything. Promise to follow platform and legal requirements, not a blanket rule.

Keep a simple record

A shared sheet is enough. One row per deliverable: tool, purpose, human edits, reviewer, date, whether the client was told and when. It protects you if a client or platform asks later, and it makes the "what did you use?" conversation a one-minute job. The approvals side lives in a creative approval workflow.

Who can skip a formal policy

If your team only uses AI for internal research and grammar, never puts AI output in a deliverable and has no client with a no-AI clause, a one-line answer ready for when a client asks is enough. Write the full policy once AI output reaches a deliverable.

Who should get a lawyer involved

• Clients in regulated fields: health, finance, legal services, children's products.
• Any work with realistic people, voices or testimonials.
• Campaigns running across several countries, where labelling rules differ.
• Any client that sends its own long AI rider.

How I'd approach this

I would write the policy once, one page, send it to every current client with a friendly note, and add the intake question for new ones. Then I would make a habit of the record sheet. Most disclosure trouble is not about the law. It is about a client being surprised.

If you want the Tool List, approvals and records to sit in one place that your team and clients both use, Blur Studio builds repeatable workflows, client approval portals and shared asset libraries around these tools. Details are on the business page, with no hard sell.