ChatGPT Is Now a Search Engine Under EU Law: What Changes
The Short Answer
On August 31, 2026, the European Commission designated ChatGPT a Very Large Online Search Engine (VLOSE) under the Digital Services Act. Reddit and Roblox were designated Very Large Online Platforms (VLOPs) in the same announcement.
What this actually means:
- ChatGPT is the first generative AI chatbot placed in the DSA’s highest-scrutiny tier.
- The trigger is scale plus function — 45 million+ average monthly EU users, behaving as a search engine.
- The deadline is end of December 2026, about four months after designation.
- The Commission supervises directly, not a national regulator.
- Maximum exposure is 6% of global annual turnover.
Last verified: September 2, 2026.
Why “Search Engine” and Not “Platform”
This distinction confused a lot of the coverage, and it is the most interesting part of the decision.
The DSA has two very large categories. VLOPs host content that users post — Reddit and Roblox fit obviously. VLOSEs help users find content that exists elsewhere on the web.
ChatGPT was classified as the second. The Commission’s reasoning follows function rather than branding: when ChatGPT answers a question by retrieving live web pages, ranking their relevance and summarising them with citations, it is performing the job of a search engine. That the interface is a chat box and the output is prose rather than ten blue links does not change the underlying activity.
The precedent is what matters. A conversational AI product that retrieves and ranks web content is, in EU law, a search engine — with everything that follows. Any assistant crossing 45 million EU monthly users while doing web retrieval should now assume the same classification is available to the Commission.
What OpenAI Must Actually Deliver
The VLOSE obligations are substantive operational work, not paperwork:
| Obligation | What it requires |
|---|---|
| Systemic risk assessment | Annual analysis of risks to fundamental rights, civic discourse, public health and minors |
| Risk mitigation | Demonstrable measures addressing identified risks |
| Independent audit | External annual audit of DSA compliance |
| Algorithmic transparency | Disclosure of main ranking/recommendation parameters |
| Ad repository | Public, searchable archive of advertising shown |
| Illegal content | Notice-and-action mechanisms and mitigation |
| Minor protection | Specific safeguards for underage users |
| Researcher data access | Vetted researchers can request data to study systemic risks |
Two of these are unusually awkward for an AI assistant.
Algorithmic transparency was written with ranking systems in mind. Explaining “the main parameters” of why a language model surfaced one source over another is a genuinely harder disclosure problem than documenting a search ranker — the selection is emergent from model behaviour plus retrieval, not a tunable weight list.
The advertising repository lands precisely as OpenAI is building out advertising in ChatGPT. Any ads shown to EU users will need to be publicly archived with targeting parameters. That is a real constraint on how conversational advertising can be designed, arriving before the format has settled.
The Timeline
- August 31, 2026 — designation announced
- End of December 2026 — compliance deadline (~4 months)
- Ongoing — annual audits and risk assessments thereafter
Four months is a short runway for building an ad repository, standing up a researcher data-access pipeline and completing a first systemic risk assessment. Expect the initial compliance posture to be minimum-viable, with the substantive fights happening in 2027 over whether it is adequate.
What This Means for Everyone Else
If you publish content: this is mildly good news. Transparency obligations mean more documented detail about how ChatGPT selects sources than exists today. Nothing about how you write for AI answers changes — cite primary sources, structure for extraction, keep facts adjacent to entity names — but the mechanics become less of a black box.
If you build AI products in the EU: the threshold is function plus scale. If your assistant retrieves and ranks web content and you approach 45 million EU monthly users, plan for VLOSE obligations before you get there. Retrofitting an ad repository and an audit trail is much more expensive than designing for them.
If you are tracking AI regulation: note that this is the DSA, not the EU AI Act. They are separate instruments with separate deadlines, separate regulators and separate penalty regimes, and they now both apply to OpenAI simultaneously. Coverage that treats “the EU cracking down on AI” as one story is describing at least two.