The EU designated ChatGPT under the Digital Services Act. It classified it as a search engine, not a platform.

On 31 August 2026 the European Commission designated ChatGPT under the Digital Services Act. It did not designate it as a platform. It designated it as a Very Large Online Search Engine.

Reddit and Roblox were designated the same day, as Very Large Online Platforms. ChatGPT was not put in that category, and the difference is the part of this story worth understanding.

The Commission's wording is unambiguous: "the Commission has designated ChatGPT as a Very Large Online Search Engine (VLOSE), as well as Reddit and Roblox as Very Large Online Platforms (VLOPs), under the Digital Services Act (DSA)".

The definition it was designated under, clause by clause

What triggered it, and what happens now

The threshold is mechanical and it is the same for both categories. These services "declared that they reach at least 45 million average monthly users in the EU and thus meet the threshold for designation". That is a declaration by the service, not a Commission measurement.

The clock is four months from notification. The Commission puts the date plainly: "these services have four months, i.e. by January 2027, to comply with the additional DSA obligations for VLOPs and VLOSEs".

What those obligations cover is listed in the same sentence, and the list is broader than most summaries suggest. Providers must assess and mitigate "the systemic risks stemming from their service and algorithmic systems related to the dissemination of illegal content, the negative effects on minors, users' physical and mental well-being, fundamental rights, electoral processes and public security".

For a generative assistant, "systemic risks stemming from their algorithmic systems" is a very large sentence indeed.

The enforcement ceiling is in Article 74(1) of the Regulation: the Commission may impose fines "not exceeding 6% of its total worldwide annual turnover in the preceding financial year".

Why the category matters even though the deadline does not change

Read the Commission's sentence carefully and it applies "the additional DSA obligations for VLOPs and VLOSEs" to all three services together. So this is not a case where ChatGPT drew a lighter regime. The headline obligations and the January 2027 date are shared.

The category matters for a different reason: it establishes what kind of thing the Commission thinks a conversational assistant is.

The Digital Services Act defines an online search engine at Article 3(j) as "an intermediary service that allows users to input queries in order to perform searches of, in principle, all websites, or all websites in a particular language, on the basis of a query on any subject in the form of a keyword, voice request, phrase or other input, and returns results in any format in which information related to the requested content can be found".

That definition was drafted with Google and Bing in view. Applied to ChatGPT, most of it fits comfortably. A prompt is a query. "Or other input" is drafted broadly enough to cover conversational text without any strain at all.

Two clauses are doing real work.

The first is "perform searches of, in principle, all websites". A language model does not maintain an index of all websites. It answers from its weights, and searches the web selectively when it decides to. That is the weakest fit in the definition, and it is the one a provider would contest if it wanted to.

The second is the one that makes the designation work: "returns results in any format in which information related to the requested content can be found". A synthesised prose answer with no list of links is, on those words, results in a format. Five words carry the entire question of whether a chatbot is a search engine, and they were written before anybody needed them to.

This is not the AI Act, and the difference is not cosmetic

A good deal of the coverage has folded this into a general story about Europe regulating AI. It is worth separating, because the two instruments have different regulators, different obligations and different dates.

Two instruments that are being conflated

	Digital Services Act	AI Act
What it regulates here	The service as an intermediary: systemic risk, minors, elections, illegal content	The AI system and its provider, by risk tier
What happened on 31 August 2026	ChatGPT designated a Very Large Online Search Engine	Nothing. This designation is not an AI Act action
The deadline that follows	Four months, which the Commission states as January 2027	Its own separate timetable, unaffected by this
Maximum fine	6% of total worldwide annual turnover, Article 74	A different penalty regime under its own articles

If you are briefing a board or a compliance function this week, the correction worth making is small and specific: the obligation that lands in January 2027 is a systemic risk assessment under the DSA, and it is owned by whoever owns online safety and platform compliance, not by whoever owns your AI Act readiness programme. In most organisations those are different people, and in several they do not currently talk.

Who this actually affects

Directly, three companies. Indirectly, everyone building a conversational product with European users, because the designation establishes a reading of Article 3(j) that other services will be measured against.

The question for a product team is no longer whether their assistant is a search engine in some intuitive sense. It is whether it "returns results in any format in which information related to the requested content can be found", which is a far easier test to meet, and whether they are approaching 45 million average monthly EU users, which is a number a growing product can reach without noticing.

There is a second-order point for anyone running retrieval-augmented systems internally. The Commission has treated the synthesis of an answer as the returning of results. That is a reasonable reading, and it is one that does not obviously stop at consumer products.

The position

The interesting thing here is not that Europe regulated an AI product. It is that it did so without needing the AI Act at all.

A definition written in 2022 for services that index the web has been applied to a system that does not index anything, and it fits because of a clause about output formats rather than a clause about search. That is regulation working the way it is supposed to when it is drafted around function rather than technology, and it is a warning to anyone who has been planning their compliance calendar exclusively around AI-specific law.

The rules that catch you first are frequently the ones that were not written about you.