I keep getting asked some version of the same question: Do we need to disclose that this image was made with AI?
And the slightly annoying answer is: maybe. But the more useful answer is that we probably should have talked about it before we made the image.
AI disclosure sounds like something you deal with when you’re publishing. Add a label, check a box, put something in the caption and you’re done. In practice, a lot of that decision gets made much earlier, when we’re deciding what belongs in the frame, how realistic we want the image to feel, and where the campaign is eventually going to run.
Now that New York’s synthetic performer law and the EU AI Act’s Article 50 requirements are in effect, that’s becoming part of how I think about the shot list itself. Not the funnest line item on a shoot brief, admittedly, but a useful one.
THE QUICK VERSION
If you’re creating AI imagery for advertising, pay attention to three things: people, realism, and placement.
New York has disclosure requirements around AI-generated people in ads. California’s new AI transparency requirements primarily sit with the AI platforms themselves, not brands. And if you’re publishing into the EU, Article 50 can create disclosure obligations for the company actually using the content.
So the question isn’t simply, “Was AI used?” It’s, “What exactly did we make, and where are we putting it?”
The easiest place to start: what’s actually in the image?
Let’s say we’re creating a campaign for a skincare brand. We make one image of the bottle sitting on a bathroom counter and another with a completely AI-generated woman holding that same bottle.
Creatively, those might feel like two images from the same shoot. From a disclosure standpoint, they’re not necessarily the same thing at all.
If there’s an AI-generated person in the image, pay attention to New York.
New York’s synthetic performer law took effect June 9, 2026 and covers advertising that uses a synthetic performer, essentially an AI-generated person who appears to be a human performer but isn’t recognizable as an actual person.
So an invented model holding your product is where the law becomes relevant. A product sitting by itself on seamless, a flat lay, or a beautifully art-directed kitchen counter with your supplement next to a glass of water and suspiciously perfect morning light are different.
That’s why I don’t love treating “AI imagery” as one giant category. What you choose to put in the frame matters.
New York also requires the disclosure to be conspicuous, and ArentFox Schiff’s breakdown of the law notes penalties of $1,000 for a first violation and $5,000 for subsequent violations.
There isn’t detailed state guidance yet telling brands exactly what “conspicuous” needs to look like, so for now we’re working with a reasonable standard: visible, understandable, and not tucked somewhere a normal person would never see it.
California is different, and this one has been getting confused a lot.
I’ve seen California’s AI Transparency Act (as of August 2, 2026) summarized as another law requiring brands to label their AI-generated ads but that’s not actually what it does.
The requirements we’re talking about here are aimed at large generative AI providers, which now have responsibilities around things like provenance data, labeling options, and tools that can help identify AI-generated content.
So California is putting more responsibility on the tools creating the content rather than simply telling every brand using AI imagery to slap an “AI generated” label on its Instagram ad.
The part I find more interesting for creative teams is that the files themselves are increasingly carrying information about where they came from, whether or not that’s something the brand is actively thinking about.
We’ll come back to that.
Then there’s the EU, and brands should pay attention to this one.
Article 50 of the EU AI Act also began applying August 2, 2026, the same day as California’s law, but with a very different implication for brands.
The European Commission distinguishes between the AI provider, which has responsibilities around marking AI-generated content, and the deployer, which is the company actually using the AI system or content. If you’re a brand publishing the work, you can be the deployer. Orrick has a good breakdown of that distinction if you want to get into the weeds.
For the disclosure requirement we’re talking about here, the important questions include whether the image is AI-generated, whether it resembles real people, objects, places, or events, whether someone would reasonably interpret it as authentic, and whether it’s obviously artistic or fictional.
Which brings me to something I now want answered before we start making anything.
Would someone think this was a real photograph?
This is an interesting creative question because, for a lot of commercial AI work, that’s exactly what we’re trying to accomplish.
We want the lighting right, the product sitting naturally in the hand, the skin to look like skin, and the shadows and reflections to behave. We want someone scrolling past the ad to see a beautiful campaign image, not stop because somebody suddenly has seven fingers and the bottle cap appears to have entered another dimension.
Realism is part of the craft, but it can also affect the disclosure conversation.
A surreal image of a 20-foot serum bottle floating over Los Angeles is obviously constructed. A completely synthetic lifestyle image designed to look exactly like something we photographed on location is a different situation.
Neither approach is inherently better. We just need to know which one we’re making.
And separately from disclosure, the ordinary rules about advertising claims haven’t gone anywhere. If a synthetic model’s impossibly even skin is doing the selling in a skincare ad, that’s a claim about your product, the same as it would be if we’d photographed a real model and retouched her into the same place. A disclosure label doesn’t change that part.
We also need to know where it’s actually going to run.
This is the part that makes a universal AI disclosure rule difficult because the exact same image can have different considerations depending on where and how it’s used.
A US-only campaign isn’t necessarily the same as a European campaign. Paid social isn’t necessarily the same as organic, and a one-off social post can very easily become the hero image for an international campaign six months later.
That last one happens more than you’d think.
So I’ve started thinking about disclosure the same way I think about crops, file sizes, or alternate layouts: plan for the places the asset may need to go before you need it there.
This is why I would make two masters.
If we’re creating an AI image that could require disclosure, I want two finished versions: one clean master and one disclosed master.
And I want the disclosed version to actually be designed. Same typography, same spacing, same attention to the composition. It should feel like it belongs there, not like someone remembered compliance 14 minutes before the ad was scheduled and dropped eight-point Arial into the corner.
If we’re already producing the image, creating that second master is easy. Going back through 37 final assets three months later because the campaign expanded into another market is decidedly less fun.
It’s a tiny production decision that can save a completely disproportionate amount of aggravation later, and it’s one of the things a directed shoot handles that a tool subscription doesn’t.
The platforms may have opinions too.
Legal requirements aren’t the only thing we’re planning around. Platforms are also developing their own systems for identifying and labeling AI-generated content, and AdExchanger has a useful overview of how those platform policies and state laws are beginning to overlap.
Some platforms can read provenance information or Content Credentials attached to a file, while others have their own disclosure requirements for certain kinds of content. I would expect those systems to keep changing much faster than legislation does.
That’s another reason I prefer having our own disclosed version ready. If a disclosure is going to appear with the creative, I’d much rather we’ve thought about how it appears.
The provenance technology itself is already very real. OpenAI has been adding Content Credentials to generated media as part of a broader effort to make the origin of AI-generated content more identifiable.
That doesn’t mean provenance is magic, though. Metadata can disappear when files are downloaded, resized, reformatted, screenshotted, or passed through different systems, and different technical approaches survive different amounts of manipulation.
So I wouldn’t build a brand’s compliance strategy around the assumption that a file will always perfectly announce its entire history. We’re not quite there.
What I’m actually checking before an AI campaign goes live
This doesn’t need to become a 42-point compliance process. For most of the work we’re doing, I want six questions answered:
- Is there an AI-generated person anywhere in the campaign?
- Would a normal person assume these images were photographed?
- Where is the campaign actually running?
- What does each relevant market and platform require?
- Do we have a clean master and a disclosed master where we may need one?
- Is somebody actually documenting that decision?
I want those questions asked while we’re planning the shoot, not while someone’s uploading the final assets.
If you want something you can keep next to the brief, we’ve also put together an AI Disclosure Guide that walks through the decision process, what to check, and what to have ready before an AI-generated asset goes live.
One important boundary here
I’m a creative director, not your attorney, and these rules are new enough that a major campaign involving synthetic people, especially one running internationally, is absolutely something I’d run past counsel.
What I can tell you is where this belongs in the creative process, and that’s early.
Whether something legally needs to be disclosed may ultimately be a question for counsel. Making sure the campaign is built so that disclosure is easy is a production question, and that part is very much our job.
It’s also a good question to ask anyone producing AI imagery for your brand.
Not as a gotcha. Just ask them, “How are you handling AI disclosure and disclosed masters?”
You’ll learn a lot from the answer because good AI production isn’t just about making a gorgeous image. It’s knowing how that image is going to survive the rest of its life: different crops, platforms, markets, retailer requests, campaign extensions, compliance questions, and all the other decidedly unsexy things that happen after everyone approves the pretty picture.
It’s the same reason I care so much about who is accountable when a deliverable is wrong when a founder is choosing an agency or creative partner. The final JPEG is only one piece of the job.
And if you’re thinking more broadly about where AI belongs in your brand, I wrote about what I’ve been noticing about brands using AI right now, including which parts of the brand I think are worth protecting and where AI can genuinely make the work better.
One of the biggest changes I’m seeing as brands get more comfortable with AI is that the question starts to change. The first few shoots are about Can we make this? and then, pretty quickly, it becomes How do we use this well?
I think that’s the much more interesting stage because we’re not really talking about AI anymore. We’re talking about building a smarter creative production system.
Sometimes, after looking at the brief, the smartest answer is AI. Sometimes it’s a live shoot. Very often it’s both.
If you’re planning your next campaign and want help figuring out which is which, book a shoot call. We’ll look at what you need to create, where it needs to go, what makes sense to produce with AI, and what deserves a camera, a crew, and the real thing.
No allegiance to the tool, just the best way to make the work.
