ChatGPT is classified as search engine in Europe

What the designation actually triggers

The VLOSE designation brings a distinct set of DSA obligations beyond those that apply to Very Large Online Platforms (VLOPs), including annual systemic risk assessments, independent audits, and structured data-sharing with regulators and qualified researchers. VLOPs, by contrast, face obligations focused more heavily on content moderation and user reporting. OpenAI now has four months, until the end of November 2026, to have those systems working. Non-compliance can trigger fines of up to 6 percent of global annual turnover, and total DSA enforcement fines have already reached approximately €870 million since the regulation's first enforcement action in December 2025.

The risk assessments are where this becomes real. OpenAI must assess and document how ChatGPT affects the spread of illegal content, the protection of minors, users' mental and physical wellbeing, fundamental rights, electoral processes, and public security. An independent auditor must check the findings, which then go to the Commission and, in part, to qualified researchers. These are not theoretical risks. Research has already identified systemic risks in these areas, including some of the same concerns the DSA was designed to address through its VLOSE and VLOP rules. ChatGPT now has to assess those risks formally, report on them, and have the work independently audited.

Lawyers will dispute whether the classification fits. ChatGPT is not a search engine in the way Google is. A user does not enter a query and receive a ranked list of links. The system synthesises an answer, and the web retrieval that sometimes underlies it is not always visible or even active. OpenAI confirmed that its search function operates as a search service under the DSA, which effectively concedes the point for regulatory purposes. But the underlying tension does not disappear. When a system generates an answer rather than indexing sources, the transparency and audit obligations designed for traditional search architectures require genuine interpretation to apply. Requiring OpenAI to document how it ranks results is sensible for a link-list engine; for a model that produces synthesised prose, the equivalent obligation is harder to define and harder to audit.

The Commission knows this. The designation covers ChatGPT's hybrid service, meaning the version that can retrieve information from the web, rather than the base model operating from training data alone. That boundary will be difficult to police in practice as the two modes increasingly blend within a single product interface.

What comes next for other AI assistants

Notably, in addition to ChatGPT, Reddit and Roblox received VLOP designations on the same date. But unlike for ChatGPT, the designations for Reddit and Roblox follow an established template. And so the Commission's willingness to apply the VLOSE category to a generative AI product sets a precedent that will almost certainly reach Gemini, Perplexity, Claude, and any other AI assistant that crosses the 45 million threshold. Which is not so difficult to do at the current rate of growth across these services.

For companies building on top of ChatGPT, there are now two separate compliance questions. The DSA governs OpenAI's obligations as a designated service, while the AI Act can create separate obligations for companies using ChatGPT in their own high-risk systems. The same model can therefore sit inside two different regulatory frameworks.

OpenAI now has four months to put the systems required by the VLOSE designation in place. The bigger question is how those rules will work in practice for an AI system that does far more than traditional search. The regulation was written for search engines. ChatGPT is something broader, and the rules will now have to be applied to it.