When does an AI chatbot legally become a search engine?

As soon as an AI chatbot searches the web and reaches at least 45 million users per month in the European Union, the European Commission can treat it as a very large online search engine. That is exactly what happened with ChatGPT on August 31, 2026: the Commission designated the service as a very large online search engine under the European Digital Services Act (DSA), along with Reddit and Roblox, are considered very large online platforms. For OpenAI, this means that by January 2027, the most stringent requirements of the DSA will apply, ranging from risk assessments to a public advertising registry.

The Background

On August 31, 2026, the European Commission designated three services under the DSA. ChatGPT was designated as a very large online search engine, or VLOSE in technical jargon. Reddit and Roblox were granted the status of very large online platforms, or VLOPs. The three services had themselves declared that they reach an average of at least 45 million monthly users in the European Union, the threshold set by the DSA for designation.

This brings the total to 28 designated very large platforms and search engines. Oversight of these services falls to the European Commission itself, in collaboration with the national digital services coordinators. For ChatGPT and Reddit, that is the Irish regulator Coimisiún na Meán; for Roblox, it is the Dutch Authority for Consumers and Markets—in each case because the service has its European headquarters there.

The designated services will have four months from the date of notification to comply with the additional requirements. Specifically, ChatGPT, Reddit, and Roblox must be in compliance by January 2027.

The ruling

The Commission provides a strikingly brief rationale for classifying ChatGPT. ChatGPT is an AI system that responds to users’ questions and requests, in part by searching the web. As a result, the Commission considers it a hybrid service that falls under the DSA as an online search engine. Reddit and Roblox, on the other hand, allow users to distribute third-party content to the public and are therefore online platforms.

This designation comes with a heavy set of obligations. From now on, ChatGPT must conduct an annual assessment of the systemic risks posed by its service and its algorithms (Article 34 of the DSA): the dissemination of illegal content; negative effects on minors; on the physical and mental well-being of users; on fundamental rights; on electoral processes; and on public safety. OpenAI must then effectively mitigate these risks (Article 35 of the DSA), and an independent auditor will verify annually that this is being done (Article 37 of the DSA).

The Commission further notes that this designation grants it investigative powers. It can now scrutinize the operations of the service and any connected systems. The general DSA obligations for search engines and platforms were already in effect prior to the designation; what has changed is the additional layer of requirements for the very largest players.

Legal analysis and interpretation

Why the label “search engine” rather than “platform” matters

The Commission had two labels to choose from and selected one. ChatGPT generates responses, organizes third-party information, and allows users to share content; thus, it exhibits characteristics of both a search engine and a platform. Within the DSA itself, this choice makes little difference: the obligations for very large search engines and very large platforms are largely the same.

Outside the DSA, the difference is definitely noticeable. The European Regulation on Media Freedom Article 18 imposes specific safeguards regarding the handling of content from recognized media services, but only on very large online platforms. As a search engine, ChatGPT is exempt from this regulation: a news article that is summarized or omitted in a generated response does not enjoy the procedural protections that the same article would receive on Facebook or TikTok.

The search engine label opens another door at the same time. The Regulation to Promote Fairness and Transparency for Business Users of Online Intermediation Services, better known as the P2B Regulation, requires online search engine providers, under Article 5(2), to explain publicly and in plain language which key parameters determine the ranking of search results. The definition of “online search engine” in that regulation is broadly worded. The DSA guidance is not legally binding with regard to the application of the P2B Regulation, but anyone who qualifies as a search engine under one regulation will find it difficult to argue that they do not qualify as one under the other. For online stores and service providers that are ranked in ChatGPT responses or remain invisible, this is an argument for demanding transparency.

Does the hosting exemption still apply to an AI-generated response?

A second question that further clarifies this issue concerns liability for the responses themselves. Article 6 of the DSA exempts hosting services from liability for user-generated content under certain conditions, provided the service remains neutral. The Court of Justice has interpreted that neutrality strictly this year. In the judgment Coyote System (ECJ, June 16, 2026, C-188/24 and C-190/24), the Grand Chamber ruled that a service that uses an algorithm to determine under what conditions, in what manner, and in what order third-party content is displayed thereby exercises control. Whoever exercises control loses the exemption. We analyzed that judgment earlier in our article on the Hosting Exemption and the Recommendation Algorithm.

Apply that standard to an AI chatbot, and you can guess the outcome. ChatGPT selects sources, weighs them, and synthesizes them into a single answer. That represents a higher level of control than a recommendation algorithm that merely organizes existing posts. Even when the answer properly cites its sources, the synthesis remains the work of the system itself. Following the Coyote System ruling, the likelihood that the Court of Justice will apply the hosting exemption to such generated responses appears particularly slim. That question has not yet arisen, but the first case in which an aggrieved party directly sues OpenAI over a generated response will inevitably raise it.

What the Digital Services Act Adds to the AI Regulation

OpenAI was already under European oversight. The Artificial Intelligence Regulation (The AI Regulation) already imposes risk management and transparency requirements on the provider of the underlying model. The DSA shifts the focus: it is not the model, but the service and its societal impacts that take center stage. This results in obligations that are not found in the AI Regulation.

The clearest example is advertising. Since August 24, 2026, ChatGPT has been displaying advertisements in 31 European countries, for now only available to users of the free plan and the Go subscription. Article 39 of the DSA requires a very large search engine that displays ads to maintain a publicly searchable advertising registry: who advertised, what message was displayed, during what period, and to which target audiences. Any Belgian company that advertises on ChatGPT will thus become publicly visible. In addition, a very large search engine with a recommendation system must offer at least one option that does not rely on profiling and must explain the main parameters of that system (Article 38 in conjunction with Article 27 of the DSA). What this transparency looks like when the recommended result is generated automatically is a question that the Commission and OpenAI will have to answer in the coming months.

Access for researchers is also a new provision. Article 40 of the DSA grants accredited researchers the right to access the agency’s data to study systemic risks—a right not provided for in the AI Regulation. The extent to which this applies to a language model—including training data or model weights—remains to be seen. Meanwhile, no one should underestimate the Commission’s willingness to enforce the rules: at the end of 2025, it imposed a fine of 120 million euros on X, the first severe DSA penalty, and the cap is set at 6 percent of global annual revenue.

Specifically, what does this mean?

For providers of AI services and online platforms. The guidance shows that the Commission has no hesitation in classifying generative AI under the existing DSA categories. Anyone offering an AI service with search or platform functions is already subject to the general DSA obligations and must publish its user numbers every six months. If the service grows toward 45 million European users, a designation becomes likely. We previously discussed how those users are counted—and why the count is higher than many providers expect—in our article on the question when an online store is a very large online platform. You can find an overview of all requirements on our page about the Digital Services Act.

For businesses that advertise on ChatGPT. You can expect your campaigns to appear in a public registry starting in early 2027, along with the ad content, the time period, and the target audience parameters used. Competitors and journalists will have free access to this registry. In addition, make sure your ad content complies with standard rules regarding advertising and unfair trade practices: the registry will make any violations easy to find.

For publishers and media companies. As long as ChatGPT is a search engine and not a platform, the media safeguards set forth in Article 18 of the Media Freedom Regulation do not apply. Anyone who sees their journalistic content appear in generated responses must therefore base their position on other grounds, such as copyright, press publishers’ rights, and the transparency rules of the P2B Regulation. The broader framework for AI systems, including the provider’s transparency obligations, can be found on our page about the AI Act.

Frequently asked questions (FAQ)

What is a very large online search engine under the DSA?
An online search engine that reaches an average of at least 45 million active users per month in the European Union and has been formally designated by the European Commission. Once designated, the strictest DSA obligations apply: annual risk assessments, risk mitigation measures, independent audits, an advertising registry, and access for researchers.

When will ChatGPT be required to comply with the stricter rules?
Four months after the notification of the designation on August 31, 2026—that is, by January 2027. The general DSA obligations for search engines had already been in effect for some time.

Does this guidance change anything for regular ChatGPT users?
Eventually, yes. Users will have greater transparency regarding ads and how the system works, a point of contact for complaints, and—in the case of recommendation systems—at least one option that does not involve profiling. Furthermore, the protection of minors carries significant weight in the risk assessment that OpenAI is required to conduct.

Conclusion

With the designation effective August 31, 2026, the European Union will henceforth treat ChatGPT as a very large online search engine, with all the obligations that entails: risk assessments, audits, an advertising registry, and oversight by the European Commission itself. Furthermore, the decision to classify it as a search engine—rather than a platform—determines which related European regulations apply and which do not. For anyone in Belgium who offers AI services, advertises on such platforms, or has content found there, now is the time to assess their own position under the DSA.