# European Commission Designates ChatGPT a Very Large Online Search Engine

OpenAI reported 159.1 million average monthly users of ChatGPT search in the European Union, more than three times the threshold that triggers the Digital Services Act's strictest tier.

- Published: 2026-09-01T06:29:06.347Z
- Canonical: https://polylog.news/ai/2026-09-01/european-commission-designates-chatgpt-a-very-large-online-s
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news)

The European Commission has designated ChatGPT as a very large online search engine under the Digital Services Act (DSA), the European Union (EU) law that imposes extra obligations on the largest platforms. It is the [first AI chatbot to receive that designation](https://www.techtimes.com/articles/326086/20260831/chatgpt-becomes-first-ai-chatbot-designated-search-engine-under-eu-law.htm). Reddit and Roblox were designated on the same day under the parallel rules for very large online platforms.

The trigger was OpenAI's own reporting. The company disclosed that ChatGPT search averaged [159.1 million monthly active recipients in the EU](https://www.searchenginejournal.com/chatgpt-is-now-a-very-large-online-search-engine-in-the-eu/587801/) over the six months ending 31 March 2026, well above the 45 million threshold in the law. The Commission characterized ChatGPT as a hybrid service that qualifies as a search engine because it answers user queries partly by searching the web.

The obligations are concrete. Designated services must run annual systemic risk assessments, submit to independent audits, give researchers data access, offer a recommender option not based on profiling, and publish transparency reports. The companies have four months from notification to comply, a deadline the Commission places in January 2027, and penalties reach 6 percent of global annual turnover.

There is an unresolved technical question underneath the legal one. Systemic risk assessment and auditing methods were written for ranked link lists and feeds, not for generated answers where the output is produced fresh for each query. Auditing a model's answer distribution is a different measurement problem than auditing a ranking, and neither the Commission nor the auditing firms have published a method for it.

## What this means

Regulators are now treating the assistant interface, not the model weights, as the regulated object. That shifts compliance cost onto whoever owns the consumer surface, which favors labs with legal and audit staff already in Brussels and raises the entry cost for smaller assistant products crossing the user threshold. It also creates a durable compliance asymmetry between an assistant that searches the web and one that answers only from what is stored in its own model weights (parametric memory), because the first is a search engine in law and the second, for now, is not.

## What to watch

- How the Commission and auditors define a systemic risk assessment for generated answers, since the method chosen becomes the template every assistant vendor must build against.
- Whether other assistants with large European user bases publish user counts or avoid doing so, because the disclosure itself is what triggers designation.
- Whether OpenAI ships a European recommender or answer mode that does not rely on user profiling, which would show how far product design bends to the deadline in January 2027.