Pennsylvania will require labels on AI-generated ads, with exceptions

Pennsylvania will require labels on AI-generated ads after Governor Josh Shapiro signed legislation on October 22, 2026, but the requirement contains significant exceptions. The Artificial Intelligence Deceptive Advertising Act targets synthetic advertising that creates false consumer perceptions about goods or services. It does not establish an unconditional label requirement for every advertisement involving AI.

That distinction puts advertisers between a transparency standard and an unresolved interpretation problem. Businesses must consider both how they produced an advertisement and what consumers will believe after seeing it. A generated product image, fictional mascot, and text-only promotion do not necessarily receive the same treatment.

New York already requires disclosures for certain synthetic performers, digitally created people appearing in advertisements. Pennsylvania takes a different approach, connecting disclosure to representations about products and services. The immediate pressure falls on advertisers and agencies deciding which creative assets need labels before the law takes effect on December 21.

Pennsylvania will require labels on AI-generated ads starting in December

Shapiro’s office confirmed the signing of Senate Bill 806 in its October 22 legislation signing notice. The act takes effect after 60 days, giving advertisers a transition period rather than imposing immediate disclosure obligations.

The change is a specific disclosure duty tied to deceptive synthetic advertising, with exceptions that materially limit its reach.

The law defines synthetic content as information that AI generated or significantly modified. That definition includes images, video, audio, and text. However, its exceptions separately exclude content limited to text, an important distinction between the definition and the operative requirement.

The AI advertising legislation connects disclosure to false perceptions about the nature, origin, or authenticity of advertised goods or services. Those words matter more than the mere presence of an AI tool somewhere in production.

A consumer seeing a realistic product image might reasonably interpret it as evidence of what the seller actually offers. A clearly imaginary character creates a different impression. The law makes that consumer interpretation central to deciding whether the disclosure requirement applies.

The notice must communicate that the advertisement contains content generated or modified using AI. It must appear when consumers first encounter the synthetic content and use the same medium. Where technically feasible, it must also resist removal.

That is more demanding than adding an explanation to a separate webpage. The disclosure belongs with the material influencing the purchasing decision. Whether consumers can notice and understand it forms part of the standard.

The text addresses readability, placement, contrast, duration, and audibility. An audio notice needs an understandable delivery. A visual notice needs enough visibility and screen time for consumers to comprehend it.

These provisions make the December deadline an implementation issue as well as a legal one. A production team must accommodate disclosures in the finished advertisement, including versions created for different placements.

The important timing distinction is straightforward: Pennsylvania has enacted the law, but its requirements have not yet taken effect as of October 25. Advertisers have a preparation window, while consumers should not interpret every currently unlabeled advertisement as a violation of this new act.

Nor should readers assume the law prohibits AI advertising. Its structure permits synthetic advertising while imposing disclosure obligations under defined circumstances. The dispute concerns where those circumstances begin and how consistently advertisers and enforcement officials will interpret them.

Advertisers face decisions that an AI tool cannot make for them

The Pennsylvania AI advertising law puts the central decision at the advertising stage. An image generator can identify that it produced an asset. It cannot, by that fact alone, settle what the completed advertisement communicates about the seller’s offering.

The pressure falls on the business presenting the claim, because production history and consumer perception require different judgments.

Consider a hypothetical furniture advertisement. A team could generate a room around an accurate product photograph or generate the furniture itself. Both workflows involve AI, but they create different questions about what the customer believes is available.

A generated setting might imply nothing about the product’s construction. A generated chair could depict dimensions, materials, or features the seller cannot supply. The factual implications depend on the finished advertisement, including its accompanying claims.

This example illustrates the decision problem; it does not establish how Pennsylvania officials will resolve a particular case. The act’s wording requires attention to context, and that context extends beyond a software label.

Agencies face an additional handoff problem. A client might receive a finished video without knowing which scenes used generated material. An editor might later replace a background, alter a product, or remove an existing disclosure during adaptation.

The practical implication is that asset provenance, a record of where material originated and how it changed, becomes useful operational evidence. It helps reviewers identify which questions need answers before distribution.

That record does not automatically establish compliance. It does, however, help prevent a review from beginning with uncertainty about whether the product image represents an actual photographed item.

For teams maintaining campaign documentation, a searchable knowledge base can connect approved assets, source files, and review decisions. The useful capability is retrieving the reasoning behind an approval when someone changes the advertisement.

Smaller advertisers face the same interpretive issue without necessarily having dedicated legal or production departments. A local business using a generated promotional image still needs to consider whether that image creates a misleading impression.

The sponsor’s office identifies a concrete concern: synthetic product imagery that consumers mistake for a depiction of an actual product. Its consumer disclosure standard presents transparency as compatible with businesses continuing to use AI.

That position explains the policy’s intended balance. Consumers receive information about synthetic representations, while businesses retain access to the technology. The difficulty lies in translating that balance into repeatable decisions across individual advertisements.

The commercial pressure therefore starts before any enforcement action. Advertisers must decide whether to revise existing creative material, build labeled alternatives, or seek clarification about borderline uses. Waiting for a definitive interpretation leaves less time to implement whatever that interpretation requires.

Transparency depends on more than attaching a label

The law’s central tradeoff concerns transparency without treating all creative uses of AI as equivalent. A broad rule would make identification simpler. A contextual rule can distinguish deceptive product representations from uses that do not create the same consumer concern.

A label identifies something about production; it does not establish that the advertisement’s underlying claim is true.

Imagine a hypothetical advertisement showing a jacket surviving conditions that the real product cannot withstand. Disclosing that AI created the demonstration would reveal its origin. It would not supply evidence that the jacket performs as depicted.

This distinction separates two questions that advertisers sometimes compress. Did software generate the representation? Does the representation accurately communicate what the product delivers? Answering the first does not resolve the second.

The Federal Trade Commission has long maintained that existing protections against deceptive practices apply to AI. Its 2023 AI enforcement statement rejected the idea that the technology operates outside established law.

Pennsylvania’s act likewise preserves other legal recourse. The disclosure requirement adds an obligation under specified circumstances; it does not turn a misleading representation into an acceptable one.

For consumers, that limits what an AI notice can promise. It can warn that an apparent photograph, scene, or voice has a synthetic component. It cannot certify product quality, validate a testimonial, or guarantee that the seller fulfilled other obligations.

For advertisers, the same distinction argues against treating disclosure as the final approval step. The content still needs factual review. A notice cannot substitute for checking whether the advertised feature exists.

There is also a difference between a technically present disclosure and an effective one. A notice that viewers cannot read contributes little to an informed decision. The act’s presentation requirements recognize that visibility and comprehension matter.

The same-medium requirement reflects the moment of persuasion. Someone hearing an audio advertisement should encounter the relevant disclosure through that experience. A person viewing an image should not need to discover a separate explanation elsewhere.

Removal resistance introduces another practical concern. An advertisement can pass through editing, resizing, and distribution before reaching its audience. A disclosure included in an original file might not survive every transformation.

The act qualifies its permanence requirement by technical feasibility. That qualification acknowledges an implementation limit without eliminating the underlying obligation. Advertisers still face the question of how the disclosure accompanies the content people actually receive.

The broader significance is a shift toward examining synthetic material as part of the advertisement’s message. The issue is not simply whether a production team used AI. It is whether the result borrows the apparent credibility of authentic evidence while obscuring how that evidence was constructed.

New York shows why AI disclosure laws are diverging

Pennsylvania is entering a regulatory landscape where disclosure requirements already differ by state. The closest comparison is New York’s synthetic-performer law, which took effect on June 9, 2026.

New York’s synthetic performer disclosures focus on advertisements containing certain digitally created people. The governor’s office presents the measure as serving both consumer transparency and protections for creative workers.

Pennsylvania centers its rule on misleading representations of goods and services, while New York centers its rule on a defined advertising participant.

Those approaches overlap, but they ask different initial questions. A performer-based rule begins with who appears in the advertisement. Pennsylvania’s language begins with synthetic content and the false perception it creates about the offering.

A hypothetical advertisement featuring a generated human spokesperson illustrates the distinction. The person’s synthetic nature is central to a performer-based analysis. Under Pennsylvania’s framework, the advertisement’s implications about the advertised goods or services also require examination.

Now consider a generated product scene without any people. A performer-focused requirement does not begin from the product image itself. Pennsylvania’s wording directly raises questions about synthetic representations of the offering.

Neither comparison supports declaring one state uniformly stricter. Each approach draws its boundary around different conduct. A requirement can be broader in the types of content it addresses while remaining conditional in when it applies.

For businesses advertising across states, the result is an asset-management problem. A campaign approved under one rule does not automatically answer the questions raised by another. A common label might be operationally convenient without resolving every substantive obligation.

That distinction matters when companies discuss a national standard internally. They can choose consistent production practices, but those practices do not erase differences in legal scope. Advertising teams still need to understand what each relevant requirement addresses.

The comparison also clarifies the policy choices available to lawmakers. A performer rule makes synthetic human presence the transparency concern. A product-perception rule focuses on what advertising leads people to believe they are buying.

Those choices reflect related but distinct interests. Consumers might care whether an apparent spokesperson is real. They might separately care whether the advertised product, setting, or demonstration represents something the seller can actually provide.

Pennsylvania’s approach therefore should not be described merely as another version of the same AI label. Its implementation will test whether a deception-linked requirement delivers useful notice while distinguishing synthetic material that falls outside its scope.

That is the relevant industry comparison. Different states are choosing different triggers, forcing advertisers to examine the relationship between creative technique, audience interpretation, and the underlying commercial claim.

Exceptions leave the most consequential questions unresolved

The act excludes evidently fictional content, along with parody, satire, commentary, and criticism. It also exempts content limited to text. These provisions prevent the broad synthetic-content definition from functioning as a universal disclosure mandate.

The hardest question is what creates a false perception, because that threshold determines whether disclosure becomes mandatory.

Libby O’Neill, deputy chair of advertising, marketing, and promotions at Loeb & Loeb, identifies competing readings in her disclosure trigger analysis. One interpretation emphasizes synthetic product depictions; another requires an additional misrepresentation about the product.

O’Neill also notes that existing false-advertising law already addresses misleading product claims. Her critique exposes a drafting question: what additional conduct does the new disclosure duty capture?

That question does not negate the law. It explains why a headline about mandatory labels cannot settle whether a particular advertisement requires one. An operative standard needs an interpretation that businesses and enforcement officials can apply.

The fictional-content exception creates a related boundary. An obviously imaginary scene might surround a realistic depiction of an actual product. Whether the overall context makes the relevant representation evidently fictional remains a fact-sensitive question.

For example, a hypothetical animated creature holding a beverage is different from a realistic generated image suggesting an ingredient the beverage lacks. Calling the advertisement creative does not answer what consumers understand about the product.

Conversely, the existence of a generated element does not establish deception. A synthetic setting might communicate no false product attribute. The statute’s conditional language makes that distinction relevant rather than allowing AI use alone to decide the issue.

The text-only exception is especially important for accurate reporting. Although the definition of synthetic content includes text, the exception means the act does not impose the same disclosure duty on text-only content.

That does not authorize false written claims. It identifies a limit within this particular law. Other applicable protections remain a separate question, just as they do for labeled images or videos.

Enforcement also shapes the law’s practical reach. The act assigns authority to the attorney general and district attorneys, with injunctions and civil proceedings among the available mechanisms. It does not create a general private enforcement route under this act.

The distinction affects how interpretations develop. Public guidance, enforcement priorities, and cases brought by authorized officials will matter when businesses assess uncertain boundaries. The existence of a statutory duty alone does not reveal which borderline examples will receive attention first.

The skeptical reading is therefore specific: disclosure offers useful information, but an unclear trigger can produce inconsistent decisions. A consumer might encounter different labeling practices because advertisers interpret the threshold differently, rather than because the underlying production methods differ.

Watch guidance, campaign changes, and early enforcement

The next phase concerns implementation. Pennsylvania will require labels on AI-generated ads under the act’s conditions, but the December deadline arrives before every interpretive question necessarily receives an answer.

The most useful evidence will show how officials and advertisers apply the threshold, rather than simply whether more labels appear.

The first signal is guidance from Pennsylvania enforcement authorities before the effective date. Examples involving realistic product images, fictional scenes, and mixed-media advertisements would help distinguish the competing readings.

Guidance explaining that realistic synthetic depictions generally require notice would support a broader practical approach. Guidance requiring a distinct false product implication would support a narrower interpretation. Either development would give advertisers a clearer basis for reviewing creative material.

The absence of guidance would leave the statutory language doing more work. Businesses would still need to make decisions, but readers should avoid mistaking those individual choices for an authoritative statewide interpretation.

The second signal is how advertisers change campaigns approaching December 21. The relevant question is whether they adopt consistent labels, revise specific synthetic representations, or maintain different versions for different distribution contexts.

Broad voluntary labeling would suggest that operational simplicity outweighs the burden of making individual threshold judgments. Selective labeling would suggest greater reliance on distinctions between covered and exempt material.

Those outcomes would reveal business responses, not establish the law’s final meaning. A company can disclose more than the law requires. Another company’s decision to omit a notice does not independently prove that its interpretation is correct.

The third signal is early enforcement activity after the effective date. Initial complaints or proceedings could identify the representations officials consider most problematic. Clear product misrepresentation cases would provide different information from disputes involving otherwise accurate synthetic depictions.

A complaint would still contain allegations rather than a final judicial determination. Readers should distinguish an enforcement theory, a negotiated resolution, and a court’s interpretation before describing any example as settling the law.

For AI product users and knowledge workers, the immediate lesson concerns evidence. An advertisement can look photographic without documenting a real object or event. A disclosure can help identify synthetic production without validating what the seller promises.

For enterprise buyers, that distinction applies when evaluating demonstrations and promotional claims. The useful follow-up is whether the advertised result reflects an available capability under stated conditions. A production notice answers a different question.

The Pennsylvania AI advertising law makes that separation more visible, while AI advertising disclosures will test how well labels communicate it. Neither an unlabeled asset nor a labeled one should end the inquiry into the underlying representation.

Before treating a campaign as evidence, compare the depiction with the actual offering and examine what the advertiser substantiates. If your team produces campaigns, preserve the source material and approval reasoning alongside the finished asset. Pennsylvania’s AI-generated ad labels will provide another signal about production, but the purchasing question remains concrete: does the product or service match the impression the advertisement creates? Follow the coming guidance and early cases with that distinction in mind, rather than assuming the presence of a label resolves the truth of the claim.