Starting December 21, 2026, an ad shown in Pennsylvania has to say it was made with AI when the AI changed its picture, video or audio in a way that leaves customers with a wrong idea of what they are buying. The effective date is 60 days after the governor signed it. It is easy to read this as a rule that every AI-made ad needs a label. The text says something narrower, and the narrower version is the one that matters for anyone who makes ads.
The test is what the customer believes
The law covers images, video and audio that AI has significantly modified or generated. The notice has to say the ad "contains content that was modified or generated using artificial intelligence", and it has to appear in the same medium as the ad: on screen for a video, spoken for a radio spot. Content that is evidently fictional is exempt, and so are parody, satire and anything that is only text. The bill itself also says the label does not limit any other legal remedy.
Picture three restaurants that sell wings. The first asks an image tool for a photo of its Friday special, and the tool draws a heaped basket that is twice what the kitchen sends out. That is the case the law was written for, so it needs the label. The second uses the same tool to draw a cartoon chicken in a chef's hat, which is evidently fictional, so it needs nothing. The third has a chatbot write the text of an email offer, and since that is only text, there is nothing to label.
The label does not rescue the first restaurant. A photo that shows twice the food was already a problem under older false-advertising rules, and a note saying AI made the photo does not make the basket bigger. A phone picture of the real basket takes a minute. Use AI for the background, the layout and the cartoon, and photograph the product itself.
Why the label will probably end up everywhere
The word "false" in the law has no stated comparison. One senator asked what the ad is supposed to be false compared with, and the language stayed as it was. A company with lawyers handles a rule like that the way California companies handled Proposition 65: when the warning is free and a mistake is expensive, they warn without testing and stop worrying about whether the warning was needed.
Pennsylvania has less pressure behind it, because there are no private lawsuits. But national advertisers have a second reason to label everything. A commercial made for the whole country will also air in Pennsylvania, and New York has required a disclosure on ads with AI-generated people since June 2026, with a different trigger. Sorting each ad against each state's test costs more than adding one line to every ad AI touched. Within a couple of years, most large brands' AI-touched ads will probably carry that line, including ads that never needed it, and shoppers will read it the way they read cookie banners.
What it means for a small business
The law grew out of fake AI ads that scammed older people, so the attorney general is more likely to go after those than after an honest flyer. A shop that uses a real photo and an AI-drawn border will probably never hear from anyone. A shop whose AI picture oversells its food is different, because a customer complaint can reach it, and the label it added would not help.
So the useful habit is a question you ask before you publish: if the customer came in and compared the picture with the product, would they feel misled? If the answer is yes, a label is the wrong fix, and a better picture is the right one.