Compliance for AI-generated marketing
Your AI wrote the ad.
The rules did not change.
Fair Housing does not care that a model wrote the listing. Reg Z does not care that the email came out of a prompt. The obligations are exactly what they were, the liability is still yours, and the only thing that actually changed is how much copy you now publish and how fast.
The gap
The tools that write it have nothing that checks it.
Every content tool in a regulated industry now generates copy. Almost none of them screen it, because screening needs a rulebook per industry and the generator has no idea which one you are in. So the draft arrives fluent, confident and occasionally unpublishable, and the only thing between it and a complaint is whether somebody happened to read it closely.
What a model will happily write
# a real prompt, a real answer
"Charming home in a safe,
family-friendly neighborhood
near great schools."Fluent, warm, and three Fair Housing problems: safe neighborhood, family-friendly and great schools all read as steering on protected characteristics. Nothing in the sentence looks like a violation to the person who asked for it.
What the gate says
{
"blocking": true,
"label": "Not publishable as written",
"findings": [
"great schools",
"family-friendly",
"safe neighborhood"
]
}Plus the identifiers the piece is missing, which is the half a keyword filter never looks at: the agent's licence number and Equal Housing Opportunity.
Run that exact line through the engine → It answers in about a second, with no sign-in.
What actually changed in law
Less than the headlines suggest, and more than nothing.
Most of what applies to AI-generated marketing is the law that was already there. The Fair Housing Act's advertising provision, Regulation Z's triggering terms, the FTC's rules on substantiation and endorsements, TCPA consent — none of them were written with generative AI in mind and none of them needed to be. They regulate what you publish, not what wrote it.
Synthetic performers have to be disclosed
New York amended General Business Law § 396-b so that a business creating a commercial advertisement must, in the statute's own words, “conspicuously disclose in such advertisement that a synthetic performer is in such advertisement”. It applies where the business has actual knowledge that a synthetic performer appears.
Civil penalties are $1,000 for a first violation and $5,000 for each one after. Signed 11 December 2025 as Chapter 617; effective 180 days later.
Read it yourself rather than taking our word for it: S.8420-A on the New York Senate site. If you use an AI-generated spokesperson, actor or voice in a video ad that runs to a New York audience, this is the one genuinely new obligation to know about.
What we are deliberately not telling you: that every state now makes you label AI copy. California's AI Transparency Act gets quoted that way and it does not say it — it obligates the companies that build large generative systems to embed provenance data and publish a detection tool. It is not a duty on the business writing an advert with one. The bill that would have imposed broad provenance duties, AB 3211, died in committee in 2024 and was never enacted.
This is a screening tool sold on accuracy. Overstating the rules to sell it would be the same failure it exists to catch.
What to do about it
Put a gate between generation and publication.
Not a review meeting. A step: the copy is written, the engine reads it against your industry's rulebook, and what comes out either publishes or comes back with the phrase, the reason and a rewrite. Nine rulebooks, 505 refused phrases, 29 negation words so “we can’t guarantee a closing date” is cleared rather than blocked.
However you write
In the browser
The Chrome extension checks copy where you already write it, including by right-click. Paste an AI draft, see what a reviewer would flag.
Free25 checks a day, no account
Add it to Chrome → Or use the web checker →If you build the tool The one nobody else sells
In your product
One POST between your model and your publish button. No SDK, no implementation project, and the free tier is self-serve — a key comes back in the response.
Freethen $299/mo for 10,000 checks
The Compliance API →If you are liable for a team
Across a brokerage
Every agent's copy screened however they wrote it, with a dated record of what was checked and when.
$299a month, up to 25 seats
Founding Broker Program →Before you rely on it
What it does not do.
It is not legal advice, and it does not make anything compliant
It tells you what a compliance reviewer would flag so a person can decide. The liability for what you publish stays with you — that is true of every tool in this category, including the ones that do not say so.
It reads text. It does not watch what you already posted
Monitoring published social accounts and archiving them for an examiner are different products, and some of them are good. This is the gate before publication, which is the half nobody sells as an API.
It will not catch everything, and no screening tool does
A rulebook is a list of known problems. Novel phrasing that means the same thing can pass, which is why it is a gate rather than a guarantee, and why the record it keeps says what was checked rather than that the piece was safe.
It does not detect whether AI wrote something
That is a different question and a largely unsolved one. This reads the copy and applies the rules, which is the same job whether a person or a model produced it — and the reason the answer does not depend on getting detection right.
The simplest version of all of this: your obligations did not change, your volume did. A gate is how you keep the second from breaking the first.