ArticleIssue #191 | Your 2027 AI Ad Plan Needs a Map
Last week, I had a long meeting with a Founder/CEO of a mid-sized regional services business. He’s incredibly well-known in his specific niche (read: he’s a mini-celebrity among those who share his speciality). His business is substantial and growing, and like any good CEO, he’s always looking for opportunities to improve.
That, of course, led us to AI. But not automations or apps or vibe-coded tools to support his business. He was fundamentally bored by those applications. His thing was AI creative. Even more specifically: he wanted to know how AI could be used to generate TV-spot-quality spots of him, tailored for specific audience segments.
Honestly, that was an application I hadn’t really thought of. Obviously we’ve all had (at least) preliminary conversations about how to deploy AI creative on Meta, YouTube & TikTok – but linear TV? CTV? Radio?
His rationale (once he articulated it) made a good bit of sense: he spends $100k-$250k on ad production every time he wants a new set of spots (2-4x per year). That’s $200k-$1M just to make the creatives required for his substantial TV buys. If AI could cut that by 50%, he saves $100k-$500k, without incurring the risks inherent in migrating back-office processes to AI or the legal liability that might arise from ingesting customer data into AI systems. Financially, it only gets better for him after that – the TV spots are then cut down for use on Meta & YouTube, so the same process would allow him to generate more Meta/TikTok creative, too.
It’s a hell of an idea.
Hence, today’s newsletter. Yes, most of the examples will center on Meta, but the fundamental approach extends to Google, TikTok, CTV/OTT/TV & any other platform offering to generate your creative for you. For each platform, the details (which buttons to push, which changes their system makes, how labels are applied) change, but the underlying principles remain the same. So, with that disclaimer, let’s get to it.
Where Is AI Creative Today?
For much of the last 2 years, the AI creative conversation has centered on permission: can we use an AI spokesperson? Will Meta allow it? Will our AI-enhanced ads be labeled as such? Do they need to be? What counts as “AI-generated” vs. “AI-enhanced” vs. “Just an ad?”
Candidly, that conversation is over. It isn’t resolved, but it’s nevertheless over. In 2027, the question isn’t whether you can; it’s where.
Back in Issue #161 (The AI Regulation Wars Begin), we broke down New York’s synthetic performer law, months before it took effect. In #176, we asked whether you should be running AI-generated ads at all. Since New York’s law went live (June 9), California signed its own version (effective January 1, 2027) – and even added something New York’s doesn’t have: a path for private plaintiffs (aside: I’m still working out exactly how those cases are going to be brought, but I can guarantee there will be some incredibly well-funded litigation).
In the midst of all this, a group of states has taken a fundamentally different approach to regulating AI creative: unlike NY + CA (which focus on disclosure), these laws focus on consent. Since that term is less charged (at least in this context) at the moment, those laws have flown under the radar (which, I’m guessing, was the intent).
All of this leads to a progressively more expensive patchwork for brands. The cold economic reality is that this environment disproportionately punishes the 2 groups we should be least eager to harm: consumers + innovators. Larger brands have compliance teams, expansive insurance policies and the resources to defend themselves against litigation. Small/upstart brands? Not so much. When regulation stifles innovation, total value to the consumer declines while costs rise. It’s simple economics.
I say this to share why I’ve spent so much time reading and thinking about these issues. The honest answer is twofold: (1) virtually no one is talking about them and (2) many of our clients fall into the same category as the example I opened this issue with: a regional, privately-owned business trying to use technology to remain competitive. I believe brands like those provide massive benefits to the communities where they operate, and I desperately want to see them continue to succeed. These laws have the potential to make that more difficult for those kinds of companies.
And….What’s Going On?
I’m fortunate to have access to hundreds of ad accounts, spanning companies of wildly different sizes across dozens of industries. The most common setup on all brands with a multi-state footprint is a single standard: there’s a national campaign (or a set of regional campaigns) with a single creative standard. Under the current regime that either means: (a) over-compliance in ~48 states or (b) under-compliance in ~2 states. Neither is a great place to be when the first demand letter shows up.
This week, I want to fix that.
(Quick caveat: I run an advertising agency. While I was accepted to law school, I didn’t go. That means I’m not your lawyer. Treat this as a framework to bring to your lawyer, not a substitute for one.)
2 Kinds of Laws; 2 VERY Different Problems
Before we get to structure, let’s talk about the basics. In the US today, there is NO national standard for AI creative. It does not exist. All AI creative legislation operates at the state level, and at that level, there are two very different kinds of state laws in play. Mixing them up can be extraordinarily problematic.
The first are disclosure laws: these cover AI people who aren’t anyone – a synthetic spokesperson, an AI “customer,” an AI-generated face in a lifestyle shot.
The second are consent laws: these cover AI versions of people who are someone: a cloned voice, a digital twin of your founder, a deepfake of a celebrity.
The first thing to know: a disclosure that a given ad unit or social post or TV spot is “AI-generated” only addresses the first set of laws; it does absolutely nothing for the second.
Here’s the map as it stands today:
As you can see, this is quite the patchwork. No one is quite sure what’s going to hold up, what’s going to be viable and what’s going to be challenged, so everyone is playing the age-old “throw legal spaghetti at the wall and see what sticks” game.
But, of those laws, 2 deserve extra attention:
- New York has been live since June, and there’s no required wording; audio-only ads, translation dubbing and ads for movies or games are exempt.
- California’s SB 1050 is the one that should keep you up at night. It covers synthetic performers used “prominently” (demoing the product, voicing the narration or reacting to the message), it explicitly includes AI voices, and the disclosure has to be substantially similar to “this performance features a synthetic performer” or “no human performer is depicted.” Violations count as false advertising under California law, which opens the door to private lawsuits.
At the federal level, the NO FAKES Act cleared the Senate Judiciary Committee unanimously in June, but it has not been brought before either Chamber (House or Senate) for a vote, and there’s no indication whether or not it would be signed into law by the Administration.
The quick summary: you can split disclosure by state – the AI-generated ads you run in NY + CA must carry a disclosure (California’s wording works for both). Everywhere else? So far, nothing required. BUT: you can’t split consent. Ensure you have consent in writing, every time, everywhere.
Also: the list above is far from fixed. New Jersey has an AI image disclosure bill in committee. Massachusetts had AI disclosure bills this session; none passed, and they’ll need to be refiled in January. Illinois, Texas and Washington have all reportedly discussed synthetic performer bills of their own. The list will only grow, especially if states find a way to monetize it (i.e. a $10 per AI creative “registration fee” or something – it doesn’t exist yet, but give it time).
I Wouldn’t Wait for DC (or CA) To Save You
3 things sit on top of the state laws, but none are going to get you off the hook:
- Thing #1: federal preemption is possible, but it’ll be a 12-round fight. The December 2025 executive order set up a DOJ task force to challenge state AI laws. Plenty of people are betting it wins. Honestly, it might. But an executive order doesn’t invalidate state law on its own. That means until either (a) Congress gets its act together and passes a national law or (b) a court strikes a state law down, whatever laws are on the books in a particular state are enforceable.
- Thing #2: the FTC is still the arbiter of deception/deceptive advertising. Its rule on reviews and testimonials bans testimonials from people who don’t exist. AI fakes are firmly covered by that definition. An AI “customer” describing results they never had is a problem in all 50 states, whether or not it comes with a label. This matters a great deal for any structure you want to implement going forward.
- Thing #3: Meta’s “AI info” label is not your disclosure. Meta auto-labels ads made with its own generative tools, and since June, it’s been detecting third-party AI too. But for most ads, that label lives behind a tap in the “About this ad” menu. I’m not a lawyer, but my professional opinion is that it is exceedingly difficult to call that “conspicuous” with a straight face. The safer route is to treat that as a backstop. The disclosure that protects you is the one on the creative.
I realize this all sounds gloomy-and-doomy, but it’s the reality we live in today. Progressively more states will pass laws regulating AI, as politicians become more and more desperate to appear “on it” about all things AI. How many of them stick is anyone’s guess.
2 Tiers, 1 Floor
With that context in mind, here’s how we’re thinking about the current environment, as it is today. Remember, this CAN change as states change laws, pass new ones and/or the feds do something.
Split your national campaigns into a “strict” tier for NY + CA, and a “standard” tier for everywhere else. The strict tier ONLY uses real people (or consented replicas of your execs/principals/founder/owners). The standard tier allows AI generated people in labeled still images. Underneath both is an immovable floor (based primarily on Point #2 above – the FTC exists).
Before we get into the tiers, let’s clarify what geo-targeting can (and can’t) do: Meta only lets you target people ‘living in or recently in’ a location; Google’s tightest setting is people ‘in or regularly in’. There’s no residence-only option, and both platforms say accuracy isn’t guaranteed. So, a Californian on a work trip to Phoenix can see your standard-tier ad. A Texan in Manhattan for the weekend can see your strict one.
That’s why the standard tier can’t be the “anything goes” tier. The geo split manages risk; it is NOT a firewall.
The Floor: No AI Video
Both tiers share one hard rule: no AI-generated video, motion or assets a platform creates on your behalf.
Why so strict?
Because AI video is where synthetic performers sneak in. The definitions in these laws are broad enough that a platform-generated background extra could arguably qualify as a synthetic performer. And ‘we didn’t know the platform did that’ is not a defense I’d want to test in California.
The catch is that every major platform now has some variant of these features on by default (or one click away). Meta has recently auto-enabled some AI-enhancement features across many accounts, without asking (and yes, per their Terms of Service, they can do that). Effectively, this is an opt-out exercise in every single account.
On Meta, that means turning off Advantage+ creative video generation (image-to-video), image animation and video concepts from Ads Creative Studio, then checking again at every launch, because defaults have a way of coming back.
On Google, it means turning off Performance Max video enhancements, which now include generative aspect-ratio extension and auto AI voice-over (Campaign settings > Asset optimization > Video), and doing the same in Demand Gen at the ad level.
On TikTok, it means keeping Symphony auto-generation and Symphony avatars out of your campaigns, and turning off the Smart+ automation that rotates Symphony assets in by default.
One more potential mine (in the DMZ-style minefield we’re all trying to navigate): if a platform auto-applies an AI asset or an “AI-generated” label you didn’t plan for, pull the ad and find out what slipped through. TikTok labels every Symphony Creative Studio video as AI-generated, so a label on a TikTok ad is a fairly reliable checkpoint.
The One Exception: Your Own People AND Built By Your Team
You might read the Floor and think, “So the CEO from the top of this issue is out of luck.”
Good news: he isn’t.
His use case is arguably the cleanest application of AI video in advertising today.
Both disclosure laws cover the same thing: an AI human who isn’t anyone. New York and California both define a synthetic performer as a realistic AI representation of a person who is not recognizable as any identifiable natural person. A digital replica of your CEO/founder is, by definition, recognizable as an identifiable natural person. Therefore, it sits outside both disclosure statutes AND squarely inside the consent laws. Consent laws exist to protect the person being replicated. When that person is the one commissioning the replica, the hard part of the problem disappears.
The reasoning behind the Floor doesn’t apply here either. The Floor exists because platform-generated video is where synthetic strangers slip into your ads without anyone choosing them. A replica your team builds from your founder’s own footage, voiced with your founder’s own voice, saying words your founder approved, has no unknown people in it. Every element is accounted for.
That’s why the exception is narrow, with 4 conditions that are all mandatory:
- Written consent, scoped to advertising. Not a handshake, not a verbal approval in a meeting, not an email that says “go for it.” You need a signed document that names the replica, the permitted uses (channels, markets, duration) and how consent can be revoked. If the agreement runs through a production vendor that builds or licenses the replica, California’s digital-replica contract rules (AB 2602) may require a reasonably specific description of intended uses and that the person was represented by counsel. The easiest solution is to have a lawyer draft it once and reuse it.
- Approval of every word. The replica says only scripts the principal has approved, in writing, before production. FTC endorsement rules require endorsements to reflect the endorser’s honest views. A digital twin making claims the real person never signed off on is a deception problem everywhere in the US, no matter what your consent document says.
- Built by your team, not the platform. The replica is produced in your production pipeline and uploaded as a finished asset. No Advantage+ animation, no Symphony avatars, no auto-generated cutdowns. If Meta “enhances” the replica, you’re back under the Floor.
- Principals only. Founders, owners & executives with real authority over the business. Employees are a different case: the power imbalance makes “consent” much harder to defend, and that’s exactly the situation the digital-replica contract laws were written for. If you want to replicate a frontline employee, that’s a separate conversation with your lawyer, not something this exception covers.
Meet all 4 and the replica can run in both tiers, including New York and California, with no synthetic-performer disclosure required. You may still want a voluntary disclosure, though, if the spot implies something the replica can’t truthfully claim. An ad that starts with “I was at our Cherry Hill location this morning” is a factual representation, and a replica can’t make it truthfully (unless, of course, the person in the video was there).
For the CEO at the top of this issue, the math holds. He records a single long session per year: every angle, every lighting setup, hours of voice. The production team builds the replica once, and every new spot after that is a script approval, not a $100k-$250k production. The savings he was hoping for are real. They just have to run through a consent and a script-approval process instead of a platform’s generate button.
What About TV?
Everything above assumes you can geo-split, and on Meta, Google and most CTV platforms you can do so (albeit imperfectly). Linear TV and radio are a different animal. Broadcast signals follow DMAs, not state lines. The New York DMA covers northern New Jersey and parts of Connecticut and Pennsylvania – and national cable reaches every state. So for linear, there is no Tier #2: anything running on a national or NY/CA-adjacent buy must conform to Tier #1 standards. Radio adds one more wrinkle: New York exempts audio-only ads, California doesn’t. An AI voice on a California radio spot needs an audible disclosure.
The good news: the principal exception works on TV exactly like it works everywhere else. A consented, script-approved replica of your founder clears both states. And expect your stations to start asking – New York’s law protects the media outlet only until it gets written notice, so AI representations in insertion orders are coming. The brands with a provenance file ready will clear faster than the ones scrambling to build one.
Tier #1: Real People, The Old-Fashioned Way
Based on the laws on the books today, NY & CA must be treated with far more caution, because that’s where the laws have real teeth + where CA’s private right of action can get expensive VERY quickly.
As depressing as this may be for the AI-maxis among us, our rule in these 2 states is simple: no synthetic performers. Neither state actually bans them – both technically allow synthetic performers with a disclosure. But given CA’s private right of action and no settled line between “AI-generated” and “AI-assisted,” the litigation exposure very likely is not worth it. No AI people in any format – no AI video, no AI stills, no AI avatars & no AI voiceovers (remember: CA counts a synthetic voice as a performer).
Every person on screen or on mic is a real human, or a consented replica of one of your principals (see the exception above). That means doing it the old-fashioned way: cast real talent, real customers + real employees. Any time you include a person in a video, get a signed release. Verify every testimonial. Budget for production for everyone who isn’t a principal.
For now, AI can still help behind the camera (scripts, briefs, retouching, scenes with no people in them), but – by the letter of the law – show up as a person. Note: there are plenty of laws under consideration (but NOT in effect) that expand covered definitions from “AI performers” to “AI enhanced” or “AI assisted” ads – and I have absolutely no idea where to draw the line there. Fortunately, that’s a future Sam problem. For now, I’m focused on what is currently on the books.
For ads in Tier 1, the optimal setup is to run these as separate campaigns, not separate ad sets. That prevents budget optimization from reallocating spend across tiers. Actively resist any attempts by Meta to import “high-performing” creative from other campaigns into this one, as doing so could result in either (a) Meta / TikTok / Google applying their AI “recommendations” to those creatives or (b) an AI creative accidentally running in CA or NY.
Tier #2: Still Honest, Just Faster
The other ~48 states get more creative freedom, but not a free pass. The floor still applies: no AI video (outside the principal exception), no platform-generated assets. Self-made AI stills (backgrounds, product scenes, composites your team built and checked) are fine, and they can be reviewed in batches to speed up time-to-market.
At this time, AI people can appear in still images only, provided they’re always labeled on the creative itself. And the same hard lines on consent and fake testimonials apply, because those are enforced by the FTC (which does have a national purview) and they have published a binding rule on the topic (16 CFR Part 465).
I’d highly recommend naming your campaigns so the tier is obvious in reporting, something like “2027 | Prospecting | T1 NY-CA” and “2027 | Prospecting | T2 US-ex NY-CA”
Then use the split to answer the question most brands are guessing at: what does going strict actually cost? Run labeled AI-person stills against real-person stills in a few matched Tier #2 states for 4-6 weeks. That isolates the cost of the “strict” tier on CTR and CPA, and it’ll tell you whether “strict everywhere” is cheaper than you think. Before you launch, make sure each cell will clear enough conversions to resolve a CPA difference in that window (the math is in The Tampering Tax) – most regional tests won’t, so extend the window or widen the states rather than calling it early.
If AI people are a small share of your creative, you could skip the split entirely. Run Tier #1 rules nationally and keep all your learning in one place. If they’re a large share, there’s a 3rd option: a single national standard that allows AI people in stills, labeled everywhere with California’s wording (“this performance features a synthetic performer”). That language satisfies New York too, so you get a single creative standard with no geo split – in exchange for taking on the California disclosure risk Tier #1 is built to avoid.
You Don’t Need A Law Degree To Get This Right
Before your next creative review, run every AI-touched asset through 5 questions. None of them require new software. They just require asking before you press go:
- Is there an AI-generated person in it who doesn’t exist? If yes, it can’t run in New York or California. Everywhere else, it can appear in still images only, with a label on the creative itself (not the caption, not “About this ad”). Pretty simple.
- Is there a person in it who does exist? If it’s an AI version of a real face or voice, you need written consent for advertising use. If it is a real person saying real words, you need a signed release attesting to that and clearing it for use (if it’s one of your own principals, built in-house, see the exception – it can run everywhere). Above all, remember: NY’s law doesn’t technically put the burden on you – the state has to prove the person in your ad is synthetic. But think about how that plays out: if the AG’s office runs your spokesperson through a detection tool and it comes back “AI,” the fastest (and maybe only) way to end the case is to prove he/she is real. And you’re the only party holding that proof. A signed release + the original files turns a fight between experts into a 5-minute resolution – which is MUCH cheaper than a court case.
- Did a platform make any part of it? If Meta, Google or TikTok generated, animated, extended or voiced it, it doesn’t run – that’s the Floor. Things like ‘enhancing’ an image create exposure under the FTC Act (Section 5) and to competitor suits under the Lanham Act when they materially misrepresent the product/service.
- How obvious is the disclosure on a phone? Is it high contrast, inside the safe zones, on every frame/card where the AI person appears (or is audible if the voice is synthetic). From a lawyer’s perspective, “This performance features a synthetic performer” beats “#AI” or a tiny sparkle icon every time.
- Can you prove how it was made? The tool, which elements are synthetic, what disclosure was applied and who approved it. A spreadsheet is fine. Links to emails are fine. Screenshots of Slack Messages are probably fine. Relying on Slack alone is probably not (since messages can be edited).
Bring those 5 answers into every creative review and the entire conversation changes. You’re no longer debating whether an ad “feels” too AI; you’re showing exactly where it can run, and what it would cost if it ran somewhere it shouldn’t.
And, to be honest, the two-tier structure is a bridge, not a destination.
Within a couple of years, either DC wins the preemption fight and we get one national standard (yes, please) or more states copy NY and CA until the “standard” tier isn’t worth running. Either way, the brands in the best spot will be the ones that treated this as a creative brief, not a legal patch.
Marketers have spent the last ~2 years asking how much AI can make for them. The better questions to ask as we look forward to 2027 are: what do your customers need to know about how your content was made + what does AI do to their trust?
