The EU just classed ChatGPT as a search engine. What that means for the answers you get
The EU designated ChatGPT a Very Large Online Search Engine on 31 August 2026. Every duty that follows, and what each changes for you.
On 31 August 2026 the European Commission designated ChatGPT a Very Large Online Search Engine under the Digital Services Act. It is the first AI chatbot anywhere in the EU to be put in that category. Reddit and Roblox were designated Very Large Online Platforms on the same day.
The trigger is a number. The Commission's public list of designated services records 159.1 million average monthly active users in the EU for ChatGPT, a figure the page notes is as reported by the provider, against a threshold the Commission words as "at least 45 million average monthly users in the EU". Search Engine Journal reports that the figure covers the six months to 31 March 2026. The same record names OpenAI Ireland Limited as the main establishment in the EU and Ireland as the Digital Services Coordinator, the national regulator already sitting at the centre of most European questions about the company.
The clock is now running. In the Commission's own wording, the three services have "four months, i.e. by January 2027, to comply with the additional DSA obligations".
Most coverage stops at that summary. What follows is each duty on its own, with the same three questions asked of every one: what the rule says, what it changes for a person typing a question into ChatGPT, and whether it is worth anything to you.
1. A public advertising repository
The rule. Designated services must, in the Commission's phrasing, "have a publicly available repository of advertisements". Under Article 39 of the DSA that means the ad itself, who paid for it, the period it ran, whether it was aimed at a particular group and on what parameters, and how many people it reached. The same article bars the repository from carrying any personal data about the people who saw the ad.
What changes when you ask a question. ChatGPT began carrying sponsored cards across 31 European markets on 24 August 2026, a week before the designation, as reported at the time by Digiday, FashionNetwork and EU Perspectives, and we went through that rollout and its tier boundaries in a separate piece on whether ChatGPT Go is worth paying for. Right now the only account of what those cards do, and of who is buying them, is OpenAI's. Once a repository exists, you can look up which advertisers were bidding against which kinds of question, in which countries, for how long. Ask an assistant which running shoe to buy, see a sponsored card, and there will be a public record you can go and check. It will not tell you why you personally saw that particular ad.
Verdict: the one duty on this list a normal person can use directly, and the reason this designation matters more than the usual compliance news.
2. Recommendations that do not run on your profile
The rule. Designated services must "provide an option in their recommender systems that is not based on user profiling".
What changes when you ask a question. In a social feed this is a familiar thing, a chronological toggle instead of an algorithm. Inside a chatbot the shape is genuinely unsettled, because the surfaces that could count as recommender systems are scattered: the stored memories that colour an answer, the suggested follow-ups, the prompt starters, and whatever ad ranking exists behind the sponsored card. The Commission applies this duty where recommender systems are in play, so how far it reaches into a chat interface will get argued over long before anyone announces it. The safest expectation is a settings toggle arriving in January.
Verdict: the most interesting duty here and the least predictable, so judge it on what actually ships rather than on the wording.
3. Someone has to write down what could go wrong
The rule. Designated services must identify, analyse and assess the systemic risks their service creates, then mitigate them. The Commission names the categories: illegal content, negative effects on minors, users' physical and mental well-being, fundamental rights, electoral processes and public security.
What changes when you ask a question. This is the duty sitting underneath every argument about teenagers and chatbots. OpenAI launched ChatGPT for Teens on 18 August 2026, with self-harm and explicit content limits, break nudges and quiet hours, in a year that had already brought family lawsuits and a lawsuit from the state of Florida. All of that arrived as company policy, which a company can revise on a Tuesday. From January it also has to exist as an assessed, documented, mitigated risk that a regulator can hold up against what the product actually does.
Verdict: slow, invisible from the outside, and probably the duty with the largest long-term effect on what the model will and will not say to you.
4. The annual audit
The rule. Designated services must "be audited by an independent auditor at least once a year and adopt measures that respond to the auditor's recommendations".
What changes when you ask a question. Nothing you will see in the interface. What it does is put an outside party in front of claims that have so far been self-certified. OpenAI's stated position is that ads do not influence the answer above them and that advertisers receive no conversation data. Both are plausible and neither has ever been checkable from outside. Whether an auditor can meaningfully test that first claim against a generative system, where the ranking is a model rather than a rule sheet, is an open question nobody has answered yet.
Verdict: worth something only if auditors get at the answer ranking itself, so hold judgement until the first report.
5. Outsiders get to look at the data
The rule. Designated services must "allow vetted researchers to access platform data when the research contributes to the detection, identification and understanding of systemic risks in the EU".
What changes when you ask a question. Indirectly quite a lot, on a long timer. The question everyone actually wants answered is whether the presence of a paying advertiser shifts what an assistant recommends, and by how much. That study cannot be run today by anyone outside OpenAI. Under this duty, university researchers can apply for the data to run it. Expect the first credible findings well after the deadline, and expect them to be argued about.
Verdict: the duty most likely to eventually produce the headline that changes how you use these tools, roughly two years from now.
6. Six percent of global turnover
The rule. Designated services must "share their data with the Commission and national authorities so that they can monitor and assess compliance with the DSA", and "establish an internal compliance function that ensures that the risks identified are mitigated". Behind both sits a ceiling written into the law itself. Article 74(1) of the DSA lets the Commission impose fines "not exceeding 6 % of its total worldwide annual turnover in the preceding financial year".
What changes when you ask a question. Nothing directly, and everything about whether the five duties above are real. The Commission has already shown it will use the instrument, fining AliExpress 550 million euro for DSA breaches, and on 24 July 2026 it issued preliminary findings that TikTok accounts belonging to minors do not meet the safety standards the DSA requires. A 6 percent ceiling applied to a company of OpenAI's size is not a rounding error.
Verdict: the number that turns the rest of this list from a wish into a schedule.
Keep this well away from the AI Act
These two laws get blended together in almost every article about European AI rules, so it is worth being blunt.
The EU AI Act is a different instrument on a different timetable. Its transparency rules started applying on 2 August 2026 and cover three consumer-facing things: an interactive AI system has to disclose that it is AI, deepfakes have to be labelled, and AI-generated or altered content has to carry machine-readable marks so it can be detected. Its high-risk deadlines moved later still when the AI Omnibus entered into force on 27 July 2026, pushing Annex III systems out to 2 December 2027 and high-risk AI embedded in physical products to 2 August 2028.
None of that says a word about advertising. The AI Act places no obligation whatsoever on the sponsored card sitting under a ChatGPT answer. Every duty in this article, the ad repository included, comes from the Digital Services Act. If you read a piece that credits the AI Act with forcing OpenAI to publish its ads, the piece is wrong.
What the designation does not do
Four limits, because the gap between a designation and a change you can see is wide.
It does not ban advertising in ChatGPT. Nothing in the DSA stops a designated service selling ads. The requirement is that it document them.
It does not make answers more accurate. Systemic risk is a legal category covering illegal content, minors, fundamental rights and elections. An assistant confidently inventing a restaurant opening time is not what any of this is aimed at.
It applies to ChatGPT as a search engine, and only as a search engine. The Commission designated it a VLOSE, while Reddit and Roblox got the platform designation, and that distinction shapes which articles of the DSA bite.
It sets a start date for compliance, and only that. January 2027 is when the obligations begin to apply. Nothing in the designation guarantees when a usable, searchable ad repository actually appears in front of the public, or how good it will be when it does.
The verdict
Regulation usually reaches consumers as a cookie banner. This one arrives with an actual artefact attached. From January there should be a public, searchable record of who paid to appear alongside the answers, on a service the Commission's own record puts at 159.1 million average monthly active users in the EU alone, whose advertising business passed a billion dollars of annualised revenue according to an OpenAI statement reported on 31 August, and which has never been obliged to show anyone outside how that works.
That record will be dry. Nobody will read it for pleasure. But it is the first time an AI assistant's commercial layer will be inspectable by someone who does not work there, and the first credible outside answer to the question of whether sponsorship moves the recommendation.
Full disclosure on our side: MultiChats carries no advertising on any plan, the free one included, so a repository would have nothing of ours to hold. You can read what the plans do cover instead. January is the date to put in the calendar, whichever assistant you use.