Agentic advertising: the legal rules applicable to sponsored AI agents

In brief. No specific legislation currently targets conversational advertising. In France, however, a sponsored AI agent falls under consumer law (deceptive and aggressive commercial practices), the obligation to identify all online advertising, and, as of August 2, 2026, the transparency requirements of Article 50 of the European AI Act. A "Sponsored" label at the time of the click is necessary, but it is not sufficient. The advertiser, who configures the agent, is the primary party at risk.

What is agentic advertising?

Agentic advertising refers to advertising that takes the form of a conversation with an artificial intelligence agent funded by an advertiser, rather than a static message (such as a banner, video, or sponsored link).

On September 16, 2026, OpenAI announced "Sponsored Agents" in ChatGPT. After viewing an ad, the user can start a conversation with an AI agent sponsored by a brand, in a thread separate from the ChatGPT response and clearly identified as such. Advertisers shape and refine their agents themselves using their own sites and services. According to OpenAI, the agent does not allow for purchases to be made. The service is currently being tested in the United States.

At the same time, the European Commission is preparing the Digital Fairness Act, which aims to update European consumer protection law in response to digital practices.

Why advertising labels are not enough

A traditional advertisement is a static message that is reviewed and approved before distribution. A sponsored agent produces new content with every exchange, tailored to what the user says. Therefore, every response from the agent is, in itself, a commercial practice subject to the Consumer Code.

French law requires that all online advertising be clearly identified as such, along with the advertiser on whose behalf it is created (Article 20 of the Law for Trust in the Digital Economy). The Consumer Code goes further: a practice is deceptive when it "does not indicate its true commercial intent, provided that this is not already apparent from the context" (Article L. 121-3). In a long conversation, the commercial context of the initial click may fade. The commercial intent must remain perceptible throughout the entire exchange.

Who is responsible for the AI agent: the advertiser or the AI system provider?

The advertiser is the one who configures the agent: instructions, sources, tone, and products highlighted. Therefore, it is highly likely that their liability will be the first to be pursued by an aggrieved consumer or by the DGCCRF, without excluding the liability of the AI system provider or the platform where the AI agent is available, if that entity is different from the agent's provider.

A salesperson who oversells in a store leaves little evidence. An AI agent, however, produces a written record of every exchange. Proving an unfair practice is therefore much easier to establish.

Practices to avoid for a sponsored AI agent

Letting the commercial intent fade. It must remain clear throughout the entire conversation, not just at the moment of the click. It is also prohibited under all circumstances to give the impression that the professional is not acting for commercial purposes.

Deceiving, including by omission. This includes inaccurate claims regarding features or price, withholding substantial information, or subtly disparaging competing products. When a competitor is identifiable, such comparisons must also comply with the conditions for comparative advertising: they must be objective, relate to comparable products, and target verifiable characteristics.

Crossing the line into aggressive practice. Following up, insisting, or exploiting information that the user has shared during the conversation can constitute "undue influence." When the practice targets a vulnerable audience due to age, infirmity, or credulity, it is assessed based on the discernment capacity of that audience.

Failing to disclose that the interaction is with an AI. As of August 2, 2026, Article 50(1) of the European AI Act requires that individuals be informed they are interacting with an AI system no later than the start of the exchange. According to the Commission's July 2026 guidelines, this obligation complements, rather than replaces, existing consumer protection rules.

What are the penalties for unfair commercial practices?

Misleading commercial practice : two years' imprisonment and a €300,000 fine. Penalties increase to five years and €750,000 when the offense is committed online. The fine may reach up to 10% of the average annual turnover (Article L. 132-2 of the Consumer Code).

Aggressive commercial practice : two years' imprisonment and a €300,000 fine, which may also be increased to 10% of the average annual turnover, with mandatory publication of the decision (Article L. 132-11).

Frequently asked questions

Is a sponsored AI agent considered advertising under French law?

Yes. Any communication from a professional directly related to the promotion of a product or service constitutes a commercial practice. An agent funded by a brand to present its products qualifies as such, even if it does not facilitate direct purchases.

Is the "Sponsored" label sufficient?

No. While mandatory, the commercial intent must remain clear throughout the conversation, and the content of the responses must comply with the prohibition of misleading and aggressive practices.

Who is liable if a sponsored agent misleads a consumer?

The advertiser, who controls the agent's configuration, will most likely be the primary target. However, the platform's liability is not ruled out.

Must I disclose that the user is speaking to an AI?

Yes. Since August 2, 2026, Article 50(1) of the European AI Act requires that individuals be informed they are interacting with an AI system no later than the start of the exchange.

Will the European Digital Fairness Act change these rules?

The proposal is currently being drafted by the European Commission. It is expected to supplement consumer law, but the rules described here are already in effect.