// HOW-TO · COMPLIANCE

How to comply with state AI-ad disclosure laws (2026)

Comply with state AI-ad disclosure laws like Pennsylvania's SB 806: decide if disclosure is required, write a label in the same medium, and ship it intact.

Last verified · 2026-10-24 · by Moe Ameen

State legislatures, not just platforms, now regulate AI in advertising. Pennsylvania's Senate Bill 806 — one of the first state AI-ad disclosure laws — was signed into law on October 22, 2026 (Act 99 of 2026) and takes effect December 21, 2026. It requires a clear, conspicuous disclosure when an ad's AI-generated or AI-modified content creates a false impression about the nature, origin, or authenticity of what's being sold; text-only ads are exempt. California's SB 1050 and moves in New York point the same direction: a patchwork of state statutes stacked on top of the platform labels and the FTC rules you already clear.

This is the operational task of satisfying that layer, step by step: how to decide whether a given creative is in scope, how to write a disclosure that meets the statute's same-medium and legibility rules, how to make it 'permanent or extraordinarily difficult to remove' the way the law asks, and how to run the whole thing across a batch of variants without shipping an unlabeled one. For the policy landscape and why the trigger is deception rather than AI use, read the [state AI advertising disclosure laws](/guides/state-ai-advertising-disclosure-laws) guide; for the separate Meta-platform labeling task, see [how to disclose AI-generated ads on Meta](/how-to/disclose-ai-generated-ads).

This is a practitioner checklist, not legal advice. The core terms in these laws are subjective and the map is changing fast — for regulated categories, large spends, or anything where the AI materially changes how the product reads, confirm the current statute and, if warranted, check with counsel.

The steps

Decide whether the creative is actually in scope. Start with the trigger, because it is narrower than 'I used AI.' These laws attach when AI-generated or AI-modified content creates a false impression about the nature, origin, or authenticity of the goods or services. An honest AI-generated product shot usually is not the target; an AI-altered clip that makes the product look like something it is not is. Write down, per creative, what the AI actually did and whether it changes how a viewer would read the product. That one judgment drives every step after this — and when the terms are fuzzy, treat a clear label as cheaper than the argument.
Map where the ad will run, not just where you made it. Pennsylvania's statute doesn't spell out its own territorial reach, but it hands enforcement to the attorney general and county district attorneys — officials whose job is deceptive conduct affecting the state's consumers, not conduct that merely originated there. The safer working assumption is that production geography doesn't get you out of it: list the states your buy can serve, and treat the creative as governed by the strictest rule among them. For a national paid-social campaign that can reach Pennsylvania, the practical answer is to handle disclosure at the creative level rather than trying to maintain a labeled cut for one state and an unlabeled cut for the rest — and to confirm with counsel if a campaign genuinely never reaches the state.
Check the exemptions before you decide you are exempt. Pennsylvania's law carves out content that is evidently fictional, parody, satire, commentary, or criticism, and content whose AI-touched material is limited to text. Political campaign advertising isn't folded into SB 806 either — it's a consumer-sales statute, and Pennsylvania has separately worked on AI-deepfake rules specific to campaign ads. These are narrower than they look. 'Evidently fictional' does not cover a realistic synthetic testimonial that a viewer could take as a real customer, and because the core trigger is subjective, a loosely-claimed exemption is a risk rather than a safe harbor. If you are leaning on a carve-out for a realistic commercial ad, that is the moment to get a second opinion.
Write a disclosure in the same medium as the ad. The statute specifies form, not just presence. The disclosure must be in the same medium as the ad: on-screen text for video, a visible line for an image or text ad, and a spoken line for audio. Keep the wording plain and unambiguous — that the content was generated or modified using artificial intelligence — rather than a vague hedge. On video it has to stay on screen long enough to be read; in audio it has to be at a volume and cadence a listener can actually catch. A one-frame flash or a mumbled aside does not meet the bar.
Make the label hard to remove — build it into the asset. Pennsylvania's law asks that, where feasible, the disclosure be 'permanent or extraordinarily difficult to remove from the synthetic content.' That points away from a detachable caption, a description-field note, or an ad-platform 'About this ad' panel, and toward baking the label into the asset itself: burned into the video frame, composited into the image, part of the rendered creative. Build it in at production so it travels with the file to every surface it is distributed on, rather than depending on one platform to carry it.
Clear the platform label and the FTC honesty line too. State compliance is one of three layers, and they stack. Still complete the platform step — Meta's AI self-certification where its social or political rules apply, Google's advertiser declaration — and keep any automatic 'AI info' label; those are covered in [how to disclose AI-generated ads on Meta](/how-to/disclose-ai-generated-ads). Separately, the FTC bars fake and AI-generated testimonials and requires substantiation: a state-compliant label does not make a fabricated endorsement legal. A creative passes only when the state disclosure, the platform label, and the honesty of the claim are all satisfied.
Document the call and systematize it across the variant set. At volume you cannot make these judgments once and assume they carried. Record, per creative, the scope decision, the disclosure wording, and that the label is embedded — a short log that answers 'is this in scope and is it labeled' without reconstructing it later. Then enforce it across the whole batch: a fifty-variant test is fifty compliance decisions, and the common failure is labeling variant one and letting forty-nine clones ship clean. Bake the check into your creative-approval gate so it runs before anything is published.

Common gotchas

Treating 'I used AI' as the trigger. The law keys on a false impression about the product, not on AI use itself — honest AI creative may not need the label, deceptive AI creative does.
Assuming you are exempt because you are not based in Pennsylvania. Enforcement keys on where the ad misleads consumers, not where it was produced; a national buy that can reach the state is the safer assumption to treat as in scope.
Putting the disclosure only in a platform metadata panel or description field. The statute wants it in the same medium as the ad, legible or audible inside the creative itself.
A one-frame on-screen label or a mumbled audio aside. It has to stay up long enough to read, or be at a volume and cadence a listener can catch.
Leaning on the parody/satire/'evidently fictional' carve-out for a realistic commercial ad. The exemptions are narrower than they sound and the core trigger is subjective.
Thinking a state label replaces Meta's self-certification or cures an FTC testimonial problem. The three layers stack; clearing one does not clear the others.
Labeling the hero cut and letting the rest of a variant test ship unlabeled. Per-violation penalties make each unlabeled clone its own exposure.

Legal note

This is a practitioner checklist, not legal advice, and these laws are new and subjective. Pennsylvania's Senate Bill 806 (House 163-40 on Oct. 7, 2026; Senate 48-2 on June 10, 2026) was signed into law on October 22, 2026 as Act 99 of 2026 and takes effect December 21, 2026; it requires a clear, conspicuous disclosure when AI-generated or AI-modified synthetic content (image, video, or audio — text-only ads are exempt) creates a false impression about the nature, origin, or authenticity of advertised goods or services, and is enforced by the attorney general and county district attorneys with civil penalties up to $3,000 per violation. Its text does not spell out an explicit rule for out-of-state advertisers, though enforcement is generally understood to reach ads that mislead Pennsylvania consumers regardless of where they were made. California's SB 1050 and other state measures impose different triggers and scopes, and the FTC's rules on fake and AI-generated testimonials apply federally on top of all of them. Confirm the current statute text (the Pennsylvania General Assembly's SB 806 page) and, for high-stakes creative, consult counsel.

Where Kompozy fits

The hardest line in Pennsylvania's law to satisfy by hand is step five — a disclosure that is 'permanent or extraordinarily difficult to remove,' in the same medium, legible on every single variant. A detachable caption or a platform panel does not meet it, and adding an on-screen or spoken label to fifty cuts one at a time is exactly the manual tax that makes a big test non-compliant in practice. [Kompozy](/) closes that gap because the label is applied where the creative is rendered, not bolted on afterward. It is an AI content generation and multi-platform publishing engine, so the disclosure is composited into the asset itself.

Concretely: a [Persona Shorts](/glossary/persona-shorts) or [Marketing Shorts](/glossary/marketing-shorts) ad renders with an on-screen caption layer and, for template-driven formats, a brand-exact [HyperFrames](/glossary/hyperframes) overlay — both are burned into the video frame, so a disclosure line placed there ships inside the file to every surface instead of living in a description field a platform might strip. Author the disclosure wording once for the concept and every variant in the batch carries the same embedded, legible label; a Persona Photo, Quote Graphic, or Carousel can carry the same baked-in line. That is the technical form of 'extraordinarily difficult to remove' the statute is reaching for — it travels with the asset rather than depending on one distribution surface to display it.

The per-post review gate is where steps one, three, six, and the FTC check from step six all land once. Kompozy publishes the same labeled asset across eight social platforms plus blog and email on [Autopilot](/glossary/autopilot) behind that gate, so you confirm scope, disclosure, and claim honesty a single time per creative rather than re-deciding per platform — and the same embedded label is what ships everywhere, which is what matters if a given campaign reaches a state with a disclosure law regardless of where it was produced. Be clear on the boundary: Kompozy renders the disclosure into the creative and distributes it, but it does not make the legal call on whether a given ad creates a false impression, and it does not file Meta's or Google's platform self-certification for you — those judgments and that toggle stay with you. Starter ($199/mo for 5,500 credits) fits a solo advertiser running a few labeled persona creatives; Pro ($499/mo for 18,000 credits) suits an agency embedding disclosures across high-volume AI ad batches for many brands; Enterprise is custom.

Frequently asked questions

Does Pennsylvania's SB 806 require a label on every AI-made ad?

No. The disclosure duty attaches when AI-generated or AI-modified content creates a false impression about the nature, origin, or authenticity of the goods or services — not to every use of AI. An honest AI-generated ad may not need the label; an AI-altered one that misrepresents the product does. Because the terms are subjective, many advertisers label anyway as the lower-risk path.

I'm not in Pennsylvania. Does the law still apply to me?

The statute doesn't contain an explicit out-of-state clause, but enforcement sits with the Pennsylvania attorney general and county district attorneys, whose role is deceptive advertising that reaches the state's consumers — so the safer assumption is that it applies if your ad is expected to run in Pennsylvania, regardless of where it was produced. For a national campaign that can serve a Pennsylvania audience, the simplest compliant approach is to handle disclosure at the creative level rather than geo-fencing a labeled and an unlabeled version; confirm with counsel if your campaign genuinely never reaches the state.

Where does the disclosure have to go?

In the same medium as the ad, inside the creative. For video that means on-screen text that stays up long enough to read; for audio, a spoken line at an audible volume and cadence; for an image or text ad, a visible line. Where feasible the law wants it permanent or extraordinarily difficult to remove, which in practice means building it into the asset rather than a detachable caption or a platform panel.

Does completing Meta's AI self-certification satisfy the state law?

No. Meta's label and self-certification are platform policy inside one ad system; a state statute like SB 806 is a separate legal requirement enforced by the attorney general. They stack. You complete the platform step and add a state-compliant disclosure in the creative, and both sit on top of the FTC's honesty rules, which a label never cures.

What are the penalties for getting it wrong?

Under SB 806 a court can impose a civil penalty of up to $3,000 per violation, plus attorney fees and court costs, enforced by the Pennsylvania attorney general and county district attorneys. The per-violation structure is why volume matters: an untested batch of non-compliant variants is many exposures, not one. Other states set their own penalty ranges.

How do I keep a big variant test compliant without checking each one by hand?

Make the decision once at production and carry it through the whole batch. Decide scope and disclosure wording for the concept, embed the same label into every rendered variant so it travels with the file, and run a single creative-approval gate that confirms scope, label, and claim honesty before anything publishes. The failure mode to design out is labeling the hero cut and letting its clones ship clean.