The EU has put ChatGPT in its strictest tier, and it did so by calling it a search engine

The numbers, and where each one comes from

• 159.1 million average monthly active users in the EU for ChatGPT, the figure recorded against OpenAI Ireland Limited in the Commission’s supervision table, read on 6 September 2026.
• 45 million average monthly users in the EU is the threshold a service has to declare to be designated at all, from the Commission’s press release of 31 August 2026.
• Four months, dated by the Commission as January 2027, is how long a newly designated service has to comply, from the same release.
• 364 million for Google Search and 119 million for Bing, the two search engines already in the tier, from the same supervision table. ChatGPT enters it in second place.
• $0 to $200 a month across the individual rungs of our ChatGPT plans table, checked against learn.chatgpt.com/docs/pricing on 5 September 2026. The designation attaches to none of them in particular.

Why the category matters more than the label

A platform and a search engine carry different duties under this law, and the Commission chose the search engine one. Its reasoning is that a service which answers a prompt by going and looking at the open web is doing what a search engine does, whatever the box looks like. That reading is what makes the designation portable: any assistant that searches the web and clears the user threshold is now describable in the same terms, and the Commission has said in its own words that it will keep watching for services that meet it.

The narrower point is that the figure which triggered this covers the search function. The chat that does not search is a different thing, and it has not been designated in its own right. Anyone reading 159.1 million as a count of European ChatGPT users is reading it as more than it is.

What actually changes, and when

Not the product, and not in September. The obligations start in January 2027, and the visible ones from a reader’s side are the annual risk assessment, the independent audit and the researcher data access. The first two produce documents. The third produces something this site has never been able to get: a route by which somebody who does not work for OpenAI can examine how the service behaves at scale and publish what they find.

Our own ledger will not move because of this. It records prices and allowances, and a designation changes neither. What we will record is the date the first assessment and the first audit appear, because those are the moments this stops being a legal category and starts being a document somebody can read.