On the EU's own register, ChatGPT search reaches more people in Europe than Bing.

RBy Ronnie Huss, founder of SearchScore

The European Commission designated ChatGPT a Very Large Online Search Engine on 31 August 2026. It is the first generative AI product to receive the classification.

The label is being reported everywhere today. The number underneath it is not, and it is the more useful fact.

Three entries on a register

The Commission publishes the list of designated services. Until this morning the search engine half of it had two entries and had done since April 2023.

By the regulator's own register, ChatGPT search reaches more people in Europe than Bing.

One honest caveat, because it matters and almost nobody will attach it. Those figures were declared at different times. Google's and Microsoft's are 2023 declarations; OpenAI's covers the six months to 31 March 2026 and OpenAI states it was calculated solely for DSA compliance. So this compares numbers sitting on one register. It is not an audit of three services measured on the same day.

It still means something. A business deciding whether AI search visibility is a real concern is no longer weighing a vendor's claim. It is looking at the second entry on the European Commission's list of designated search engines.

What the designation actually requires

Four months, to January 2027. The obligations are not cosmetic.

The Commission requires designated services to address "systemic risks stemming from their service and algorithmic systems", naming illegal content, effects on minors, users' physical and mental wellbeing, fundamental rights, electoral processes and public security.

Alongside that: an independent audit at least annually, data sharing with the Commission and national authorities, an option for recommender systems that does not rely on profiling, and a public repository where the service carries advertising. Penalties reach 6% of global annual turnover.

Designation is not a finding of wrongdoing, and enforcement is a separate process. It is a statement about scale and the scrutiny that follows from it.

The clause that matters most, and it is not the label

Buried in the obligations is the one with consequences for anybody trying to measure AI search: designated services must give vetted researchers access to platform data for studying systemic risks.

Consider what measuring AI visibility involves today. Nobody can see how an assistant chooses its sources, so everyone samples from the outside: ask a question many times, record what comes back, infer the behaviour underneath. We do it. Everyone in this category does it. It is the only method available, and it carries the limits of any black-box measurement, which is why we publish variance figures alongside results.

Researcher data access is the first legal route inside. It is narrow, it covers systemic-risk research and excludes commercial analysis, and it will be argued over for years. But the direction is unmistakable: how a designated search engine selects and ranks sources moves from something the industry guesses at to something documented, audited and, for some researchers, inspectable.

The systemic-risk requirement points the same way. A service that must identify and mitigate risks arising from its algorithmic systems has to describe those systems to somebody.

What changes for a business, and what does not

Very little changes this week. ChatGPT does not rank sources differently because of a designation, and no site becomes more visible because a regulator wrote a letter.

What changes is the standing of the question. Three things follow.

The category is settled. Whether AI answers count as search was an argument. In the European Union it is now an administrative fact, with a threshold, a register and a compliance deadline.

The scrutiny is asymmetric. Obligations land on the service. Nothing here requires OpenAI to tell a business why it was or was not cited, and no part of the DSA gives you a right to appeal an absence from an answer.

And the measurement gap stays open for now. Until researcher access produces something public, knowing whether an assistant recommends you still means asking it, repeatedly, and recording what it says. That has not changed today. It has just become harder to argue is unnecessary.

The honest summary

A regulator has looked at a chatbot, counted 159.1 million monthly users in Europe, and filed it next to Google and Bing.

For anyone still describing AI search as an emerging concern, that is the end of the argument. For anyone measuring it, the interesting date is not today. It is whenever the first vetted researcher gets to look inside.

Ronnie has written about the other half of this, [what it means when a credibility layer acquires a regulator](https://ronniehuss.co.uk/age-of-illusion-v-the-layer-gets-a-regulator/), including the parts we got wrong in our own scoring.