There are two different things people mean by an ad in an AI answer, and they carry different risks. The first is a bounded unit sitting below the generated text, sold by the assistant, visually separate from it. The second is a brand, product or recommendation that has been paid to appear inside the answer itself, in the model's own voice, with nothing to mark it apart. Every disclosure regime in force treats both as advertising. Only the first has a working label.
What the existing rules already say
The Federal Trade Commission set out its position on generative AI advertising in May 2023, in a business blog post called The Luring Test. It is not a rule, but it is the clearest statement of how the FTC reads its own authority in this context, and it anticipated the question before the products existed. The post says that "it should always be clear that an ad is an ad, and search results or any generative AI output should distinguish clearly between what is organic and what is paid." It goes further than a labelling instruction. It says that "People should know if an AI product's response is steering them to a particular website, service provider, or product because of a commercial relationship."
That second sentence is the one that matters for the second kind of ad. The obligation is not satisfied by marking a unit. It attaches to the steer itself, wherever the steer occurs.
The UK position is the same and is stated more flatly. Section 2 of the CAP Code requires that marketing communications are obviously identifiable as such, and the Committee of Advertising Practice notes that these rules "apply to all marketing communications, regardless of the targeting or medium". On new formats specifically, CAP's guidance is unambiguous about who carries the duty: "It is the responsibility of advertisers and publishers to cross-apply these principles as soon as they start using a new platform or format." The ASA does not wait for a format to mature before regulating it, and it does not accept novelty as a defence.
In the European Union, Article 26 of the Digital Services Act imposes the hardest version of the requirement. Providers of online platforms presenting advertisements must ensure that, for each specific advertisement presented to each individual recipient, the recipient can identify "in a clear, concise and unambiguous manner and in real time" that the information is an advertisement, "including through prominent markings", who the advertisement is presented on behalf of, who paid for it if that is a different party, and meaningful information about the main parameters used to decide who sees it. Article 26 also prohibits presenting advertisements based on profiling using special categories of personal data.
Three jurisdictions, three instruments, one answer. There is no gap in the law. There is a gap in practice.
What the AI-specific law does not cover
It is tempting to assume the EU AI Act settles this, because its transparency chapter came into force on 2 August 2026 and applies to exactly these systems. It does not settle it, and the distinction is worth getting right.
Article 50 requires providers to ensure that people are informed they are interacting with an AI system, unless that is obvious to a reasonably well-informed observer. It requires providers of systems generating synthetic audio, image, video or text to ensure outputs are "marked in a machine-readable format and detectable as artificially generated or manipulated". It requires deployers to disclose deepfakes, and to disclose AI-generated text published to inform the public on matters of public interest unless a person holds editorial responsibility for it. All of that is about the provenance of the content: machine or human, generated or authentic.
None of it is about money. Article 50 does not ask whether a passage was paid for. A sponsored recommendation delivered inside a fully Article 50 compliant assistant, correctly marked as machine-generated, is still undisclosed advertising if nothing tells the reader it was bought. Article 50(6) is explicit that its paragraphs are "without prejudice to other transparency obligations laid down in Union or national law", which is the Act telling you to go and read Article 26 of the DSA and your national consumer law as well.
The AI-specific statute governs whether a machine wrote it. The advertising rules govern whether someone paid for it. Compliance with one says nothing about the other.
How ChatGPT handles it today
OpenAI began testing ads in ChatGPT in the United States on 9 February 2026, and its published approach is worth reading closely, because it is the most consequential design decision anyone in this market has made so far.
OpenAI's model is separation. Its help documentation states that ads "can appear below the end of a response" and that "Ads are clearly labeled as sponsored and visually separated from ChatGPT's response." Its stated principle of answer independence goes further than placement: "Ads do not influence answers. Ads are separate and clearly labeled." Advertisers, it says, have no ability to shape, rank or alter responses, and receive only aggregated reporting such as views and clicks rather than any access to conversations. Ads are limited to the Free and Go tiers, are not shown to accounts identified as under 18, and are not eligible to appear near sensitive or regulated content including personal health, mental health and politics. Political advertising is not currently allowed at all. Personalised ads were not initially available in the European Economic Area or Switzerland.
Whatever one thinks of ads in an assistant, this is a defensible compliance posture, and it maps cleanly onto Article 26 and onto the FTC's organic-versus-paid line. The unit is bounded, marked and separable. A reader can tell where the answer stops and the commerce starts.
It also quietly concedes the harder point. OpenAI has drawn the boundary where it has because the alternative, letting payment reach into the response, is the thing that cannot currently be labelled.
The part nobody has solved
Now take the second kind of ad. A brand appears in a recommendation. It was placed there, or its facts were supplied, as a result of a commercial arrangement. The sentence reads in the assistant's own register, indistinguishable from the surrounding text, and the surrounding text is the reason the reader trusts it.
Every element of the disclosure requirement applies. The FTC's formulation covers it precisely, because the reader is being steered to a product because of a commercial relationship. The CAP Code covers it, because it is a marketing communication and is not obviously identifiable as one. Article 26 covers it, because it is an advertisement presented on an online interface and the recipient cannot identify it as such in real time.
What does not exist is the format. A banner has a frame. A sponsored search result has a fixed label slot. A native article has a standing "advertisement feature" bar. Generated prose has no frame, no slot and no bar, and the disclosure problem is genuinely hard rather than merely neglected. A label placed at the top of a response over-marks an answer that is mostly organic. A label at the bottom arrives after the reader has absorbed the steer. Inline marking interrupts the register that makes the format work. Machine-readable marking in the response payload is invisible to the person the rules exist to protect.
There is a second problem underneath the first, which is that in most of these arrangements the assistant is not the seller. When an assistant sells its own inventory, one party controls the answer, the ad and the label. When a brand pays a third party to influence what assistants say about it, the payment happens several steps away from the surface where the disclosure would have to appear, and no party in the chain both knows about the payment and controls the pixel.
What the industry standard covers, and what it does not
If you go looking for a trade body answer, the nearest thing is the IAB's AI Transparency and Disclosure Framework V2, published on 18 August 2026. It translates a risk-based, materiality-driven approach into practical protocols for when AI involvement should be disclosed, and it covers AI-generated and AI-assisted text, imagery, video, audio, synthetic voices, digital twins and AI-powered consumer interactions. It explicitly tries to balance transparency against what it calls disclosure fatigue, and IAB is running a workshop on it on 28 October 2026.
Read the scope carefully. The framework is about disclosing that AI was used to make the advertising. It is not about disclosing that advertising was inserted into the AI. Those are opposite problems that happen to share a vocabulary, and conflating them is the most common mistake in this discussion. As of this writing there is no published industry standard for marking a paid placement inside a generated answer, from IAB, IAB Tech Lab or anyone else. The regulatory requirement exists without a technical standard to satisfy it, which is an uncomfortable place for any market to operate in and usually a short-lived one.
What publishers and advertisers should do now
Three practical positions follow, and none of them requires waiting for a standard.
Treat the disclosure duty as attaching to whoever inserts the paid element, not to whoever renders it. That is the CAP Code's position on its face, since it names advertisers and publishers rather than platforms, and it is the reading most likely to survive contact with a regulator. If your commercial arrangement puts a brand into an answer, you are the party that has to be able to explain how a reader could have known.
Keep the paid element separable at the point of insertion. A unit that can be bounded, attributed and marked is one you can label later when a standard arrives, and one you can withdraw from a market whose rules tighten. A brand fact dissolved irreversibly into generated prose is neither. This is the practical reason the separation model matters even for parties who are not OpenAI: separability is what makes compliance retrofittable.
Log the commercial relationship on your own infrastructure. Article 26 requires identification of the party on whose behalf an advertisement is presented and, separately, the party who paid for it where those differ. In a chain involving a publisher, an intermediary and an assistant, nobody can answer that question from platform reporting alone. The record has to be yours.
This is the constraint that shapes the products in this category, including ours. blankspace serves a paid unit into the retrieval moment at the CDN edge and records it on the publisher's own logs, which means the impression is bounded, attributable to a named advertiser and auditable by the publisher rather than only by a platform. That is a precondition for disclosure, not a substitute for it. Any mechanism that places commercial content near an AI answer inherits the same obligation, and a vendor that tells you otherwise is describing a risk it intends you to carry.
Frequently asked questions
Is there a law that specifically requires ads in AI answers to be labelled?
Not one written for AI assistants, and one is not needed. The FTC applies its general prohibition on deceptive and unfair practices, and said in May 2023 that generative AI output should distinguish clearly between what is organic and what is paid. The UK CAP Code requires marketing communications to be obviously identifiable regardless of medium. Article 26 of the Digital Services Act requires online platforms to let recipients identify advertising in a clear, concise and unambiguous manner and in real time. All three are media-neutral by design.
Does the EU AI Act require disclosure of paid placements?
No. Article 50, in force since 2 August 2026, requires people to be told they are interacting with an AI system and requires synthetic content to be marked as artificially generated. It addresses provenance, not payment. A sponsored recommendation can be fully Article 50 compliant and still be undisclosed advertising under the Digital Services Act or national consumer law, and Article 50(6) states that its obligations are without prejudice to other transparency rules in Union or national law.
How does ChatGPT label its ads?
By separation. OpenAI's documentation says ads can appear below the end of a response, are clearly labelled as sponsored and are visually separated from the response, and that ads do not influence answers. Advertisers cannot shape or rank responses and receive only aggregated performance data. The test began in the United States on 9 February 2026, is limited to the Free and Go tiers, excludes accounts identified as under 18, and does not place ads near sensitive or regulated content such as personal health or politics.
Who is responsible for the disclosure, the publisher or the platform?
Under the CAP Code the responsibility is placed on advertisers and publishers, who are told to cross-apply the recognition principles as soon as they use a new format. Article 26 of the Digital Services Act places its obligation on providers of online platforms. In practice both are exposed, and in an arrangement where payment happens several steps from the rendering surface, the party that can actually evidence the commercial relationship is usually the publisher or the intermediary rather than the assistant.
Is there an industry standard for marking a paid brand mention inside generated text?
Not yet. The IAB's AI Transparency and Disclosure Framework V2, published on 18 August 2026, is the closest reference point, but its scope is disclosing that AI was used to produce advertising rather than disclosing that advertising was placed inside an AI response. No published standard from IAB, IAB Tech Lab or any other body currently specifies how to mark a paid placement within a generated answer, which leaves a live regulatory requirement with no agreed technical format.