Ads Made by AI: Do You Have to Tell Consumers? IAB Just Drew a Line

Yesterday, IAB (Interactive Advertising Bureau, the digital advertising industry's trade group) released something: the AI Transparency and Disclosure Framework, version 2.

The name is a mouthful. Let me translate it for you: going forward, when you make an ad, it spells out when you must tell consumers "this was made by AI" — and when you don't have to.

That's it. That's the whole thing.

But don't underestimate it. Marketing teams all over the world are wrestling with exactly this question right now.

First, the Conclusion: Where the Line Sits

What counts as "requires disclosure"?

Most people's first instinct: if you used AI, you label it. Copy written by AI? Label it. Images retouched by AI? Label it. Background music picked by AI? Label that too.

Wrong.

IAB's thinking is nothing like that. The question it asks isn't "did you use AI" — it's "when a consumer sees this ad, will they mistake the fake for the real."

Stop and think about how different those two questions are.

One checks the tools. The other checks the outcome.

Draft a piece of product copy with AI, and the consumer who reads it doesn't come away thinking "a real person was just talking to me." So it doesn't affect their judgment of what's real. No label needed.

But suppose you use AI to generate a video in which someone who looks like a real person does something they never actually did. The consumer watches it and believes it happened.

That one, you must label.

The key was never whether AI touched the asset. It's whether AI changed what "real" means in the viewer's eyes.

That's the line.

This Side of the Line, and That Side

Let's get specific.

On the far side of the line, disclosure required: images and video generated from prompts, certain synthetic voices and digital humans, digital replicas of deceased people, and placing a living person "into" a scene they never lived through — note that ordinary brand endorsements are excluded. Hire a spokesperson to shoot a commercial, and that's an authorized, standard practice. It doesn't count.

Chatbots too. If a consumer reasonably mistakes one for a human customer-service agent, it has to identify itself.

On the near side, no automatic disclosure: routine post-production, internal workflows, copywriting, standard audio cleanup, background music, generic synthetic voices, and avatars that are obviously cartoonish or stylized.

Put plainly: the more real it looks, the more you label. If it's obviously fake, no one's asking you to.

Is that reasonable? I think it is.

Because what regulators care about and what consumers care about turn out to be the same thing: don't get fooled. As for whether you use AI to speed up your backend, nobody cares that much.

More Than Half of People Want That Label

IAB's approach wasn't dreamed up on a whim.

Before the first version of the framework came out, they ran consumer research with Sonata Insights. The results were interesting: attitudes toward AI in advertising are split. Some think it's great for creative work; some think it feels fake.

But when asked about ads "generated entirely by AI, or containing AI-generated images and video," more than half said: the brand should tell me.

See, consumers are precise in what they ask for. They're not shouting "put a label on all AI." They only care about the stuff that might pull one over on them.

The people who want disclosure and the people who want bans have never been the same people. This research proved it once again.

The Real Headache: Every Region Has Different Rules

If it were just about drawing one line, this wouldn't be complicated.

What's complicated is that every jurisdiction is drawing its own line right now, and the lines don't line up.

Count them with me: New York State's law on synthetic performers took effect in June. California's SB 942 and Article 50 of the EU AI Act took effect August 2. South Korea introduced AI labeling requirements earlier this year.

The same creative asset, run in New York, run in California, run in Europe — faces a different set of hurdles in each market.

In the US, IAB offers two options: a standardized "flash" disclosure icon, or a clear piece of text. Note that this is industry guidance; where the law has hard requirements, the law wins.

The EU? Article 50 requires disclosure of covered AI-generated content and deepfakes, but doesn't specify which icon. A voluntary code of practice covers the design details, and a unified European icon is still something they're arguing over.

So the real value of IAB's framework is giving the marketing industry a cross-market common approach: how to draw the line, how to make the call, so every region isn't left figuring it out alone.

For Marketers, This Is Now a Classification Problem

At this point you might be thinking: isn't this the compliance department's job?

Not quite.

It used to be enough to know whether an asset "used AI." Now you have to answer a chain of questions: What did AI do to it? Did it change the authenticity or identity presented in the content? Which market will this asset appear in? Which disclosure requirement applies in that market?

"Used AI" is one-dimensional. "What AI did, where it runs, which rule applies" is four-dimensional.

Knowing that AI touched an asset isn't enough anymore. You need to know what it touched, what it changed, and where the asset ends up.

This means marketing operations just gained another category of asset attribute to manage. Like privacy compliance before it, it won't go away on its own — it will only get more granular.

Yesterday's framework is another step in that direction.

As for when we'll get one universally accepted icon? I don't know. But drawing the line clearly up front beats improvising every single time.

Here's hoping every file in your team's asset library can answer those four questions.