For most of the last decade, an agency's job at the final stage of production was fairly simple: check the copy, check the specs, check the brand guidelines, and ship it. That final checkpoint just got a new line item, and it is not going away. Starting this month, the two largest ad platforms on earth are asking a question no creative brief used to answer: was this made by a machine, and does anyone actually know?
On July 9, Google added a feature called 'How this ad was made' to My Ad Center, the panel any user can open by clicking the three dots on an ad running across Search, YouTube, or Discover. If an advertiser built the ad using Google's own generative tools, the disclosure switches on automatically. If the ad came from anywhere else, Photoshop's generative fill, Midjourney, a freelancer's Sora export, it is on the advertiser to flip a manual control saying so. Google has been explicit that it will not check. There is no verification layer, no algorithmic sniff test. It is an honor system sitting on top of a trillion-dollar ad business.
Meta moved first and moved harder. Its 'About this ad' panel now carries an 'AI info' label for any creative made or substantially edited with generative tools, and unlike Google, Meta is not relying purely on advertiser honesty. The company rolled out detection based on industry-standard provenance signals, the same C2PA-style metadata trails that Adobe, OpenAI, and others have been embedding in AI-generated files. Meta's Advantage+ tools label themselves automatically. Everything else needs a manual checkbox in Ads Manager. And this is where it stops being a UX footnote and starts being a compliance problem: as of this year, undisclosed AI content is the third most common reason ads get rejected on the platform, responsible for 14 percent of all rejections. The penalty structure escalates fast. A first violation is a strike and a pulled ad. A second within 90 days triggers a 24-hour hold on the whole account. A third can get the account suspended.
Layer state law on top of that. New York's synthetic performer disclosure statute took effect June 9, the first of its kind in the country. It requires advertisers to conspicuously disclose when an ad features a 'synthetic performer,' a generated figure meant to read as a real, though not identifiable, human being, provided the advertiser actually knew AI was used. The law does not specify font size or placement, just 'conspicuous,' which the FTC has separately defined through its own disclosure doctrine as prominent, legible, and hard to miss regardless of screen size or scroll speed. Fines start at $1,000 for a first offense and jump to $5,000 after that. Audio-only ads are exempt. Straight translation work is exempt. The broadcasters and platforms carrying a noncompliant ad are exempt too, and the liability sits with whoever made the thing.
And underneath all of that sits the FTC itself, which has spent the last year and a half extending its existing deception and endorsement rules to cover AI-generated content. Guidance issued in late 2024 made clear the agency sees no meaningful difference between a lie told by a person and a lie told by a model: AI involvement in an ad has to be disclosed, claims made by AI-assisted content need the same substantiation a human claim would need, and an AI-generated endorsement or testimonial has to be flagged as fabricated rather than presented as a real customer's words. The agency has shown it will act on this. On December 22, 2025, FTC staff sent warning letters to ten companies over likely violations of the Consumer Review Rule, the rule covering fake reviews, undisclosed incentives for positive posts, and unlabeled insider testimonials, all territory that overlaps directly with AI-generated endorsement content. Penalties under that rule run up to $53,088 per violation, and because each individual post or review can count separately, the exposure scales fast for anyone running a real volume of content.
The picture is more complicated than a simple tightening of enforcement, and that complication is worth understanding because it cuts against the tidy narrative. That same month, the FTC reopened and set aside its own 2024 consent order against Rytr, an AI writing tool the agency had previously accused of helping customers generate fake reviews, on the grounds that the order unduly burdened AI innovation. Read the two decisions together and they complicate the simple story that 'the FTC is cracking down': the agency is actively working out where the line sits between policing deception and being seen as hostile to a technology the current administration has made a point of wanting to move fast. An agency or advertiser planning around a stable, settled set of rules is planning around a target that is still moving in both directions at once.
Four rulebooks, one campaign
Step back and look at what this actually adds up to. An agency running a national campaign today has to satisfy Google's honor-system toggle, Meta's provenance-based detection with escalating account penalties, a New York statute with its own definition of 'synthetic performer' and its own fine schedule, and a federal standard that treats AI disclosure as a live wire under existing deception law. None of these four regimes use the same definitions, thresholds, or enforcement mechanisms. A creative asset can be compliant on one platform and exposed on another, in one state and not in the next.
This is the part worth sitting with: in practice, this plays out as a production problem more than a legal one. Somewhere between the concept meeting and the media buy, someone now has to know, asset by asset, whether AI touched it, how much it touched it, and whether that fact was written down anywhere. Most agencies do not currently have that person, or that system. Creative production has always been collaborative and a little bit improvisational; a junior designer runs a background through a generative fill tool to save two hours, nobody thinks twice, and the file moves on. That casualness is what the new rules are targeting, whether or not that was the intent.
The trust gap underneath the legal one
There is also a trust problem sitting underneath the regulatory one. The IAB's State of Data 2026 report found 57 percent of consumers say they are concerned about encountering fake or misleading ads built with generative AI, a figure corroborated separately by Kantar's Media Reactions research. That is a high enough number to be shaping how people evaluate brands before a platform label ever appears. Set against it is a real governance gap from the same IAB report: AI-related clauses currently appear in only around 40 percent of brand-agency and partner contracts, though the report expects that share to roughly double over the next one to two years as the market catches up. Even at 40 percent, most client-agency agreements still say nothing about who is responsible for disclosing AI use, and that gap, consumers already bracing for AI deception while the paperwork underneath most agency relationships hasn't caught up, is exactly where lawsuits and platform penalties come from. When something goes wrong, the client will ask the agency why nobody flagged it, and the agency will ask the freelancer or the stock-content vendor the same question, and if none of those relationships were ever documented, everyone finds out together, usually after the ad has already run.
Stay Ahead of AI Regulations and Trends
What agencies should actually do
The honest answer is that disclosure needs to stop being a legal afterthought and start being a production step with its own owner. That means keeping a real asset log, not a folder of files but an actual record of which tools touched which piece of creative and at what stage, so that when a client or a platform asks, the answer exists instead of needing to be reconstructed from memory. It means writing AI disclosure obligations directly into scopes of work and freelancer agreements, so a contractor who runs a background through a generative tool is contractually required to say so, rather than agencies discovering it after a platform detects it independently. It means adding a disclosure check to the same pre-launch review where legal already checks trademark and substantiation, because at this point AI provenance belongs in that same category of things that get checked before the button is pressed, not after. And it means having an honest conversation with clients about where the actual risk sits: platform enforcement can pull a campaign mid-flight, a state statute can generate a fine per violation, and the FTC's per-post penalty math means a single unlabeled batch of social creative could cost more than the campaign was worth.
The industry built its entire AI adoption over the last three years around speed and cost, and the platforms and regulators are now building the accountability layer underneath that adoption after the fact. Agencies that treat disclosure as paperwork will keep getting caught by it. Agencies that build it into how work actually gets made, quietly, as part of the process rather than a compliance checkbox bolted onto the end of it, will be the ones still standing when a client asks who signed off on that video and whether anyone can prove what made it.
