On 31 August 2026, the European Commission designated ChatGPT a Very Large Online Search Engine under the Digital Services Act. Not a platform. A search engine, in the same statutory tier as Google Search and Bing, and the first AI chatbot ever placed there.
Every marketing forecast written for 2026 predicted roughly the same four things: AI personalisation, voice search, sustainability, and augmented reality. None of them predicted that the definitional question would be settled by a regulator in Brussels, or that the answer would carry an advertising disclosure requirement.
Key Takeaways
- ChatGPT is now regulated as a search engine in the EU, effective 31 August 2026.
- The designation carries a public ad repository requirement where ads are shown.
- Vetted researchers gain data access, the first regulated route into how an assistant picks sources.
- Sponsored placements inside AI answers are already scaling, on measured data.
- Google and Microsoft both moved advertisers onto automated campaign types in September.
What the Commission Decided
The Commission’s designation covers ChatGPT’s search service, with Reddit and Roblox named Very Large Online Platforms in the same announcement. OpenAI’s own disclosure put ChatGPT search at 159.1 million average monthly active recipients in the EU for the six months ending 31 March 2026, more than three times the 45 million threshold that triggers designation.
The reasoning is narrower than the headlines suggest, and worth understanding precisely. The Commission treated ChatGPT as a search engine because of what it does, responding to queries partly by retrieving from the open web. Not because of the model, and not because of the chat interface. That reasoning is portable to any assistant that retrieves and answers at sufficient scale.
Obligations follow roughly four months from notification, putting the deadline around the end of December 2026: systemic risk assessment, annual independent audit, data sharing with the Commission and vetted researchers, a recommender option not based on profiling, and a public repository of advertisements where ads are shown. Penalties for breach reach 6% of global annual turnover. The designation itself is a classification, not a finding of wrongdoing.
Two Clauses Marketers Should Read Twice
Most coverage focused on content moderation. The two provisions that change working practice are elsewhere.
The ad repository. Designated services must publish a searchable repository of the advertisements they carry. If sponsored placement inside AI answers grows into a real channel, that repository becomes the first public record of who bought what. Competitive intelligence on a channel that currently has none.
Researcher data access. Vetted researchers gain access to platform data. Applied to a system that decides which sources to cite in an answer, that is the first regulated route into how an AI assistant selects what it surfaces. Everything currently sold as generative engine optimisation rests on inference from outputs. Within a year there may be actual evidence, and some of the current advice will not survive contact with it.
Ads Are Already Inside the Answers
The regulatory timing is not coincidental. Measurement firms have started tracking sponsored placement inside chat responses, and the curve is steep.
Comscore, announcing an extension of its AI measurement product on 3 September 2026, disclosed the finding behind it: across hotel-related ChatGPT prompts with identified source links, sponsored-ad presence rose from 6% in March 2026 to 14% in April and 24% in May. One vertical, one assistant, one quarter. Four times the presence in three months.
That is the story the 2026 forecasts missed. They predicted AI would personalise advertising. What actually happened is that AI answers became advertising inventory, and the measurement infrastructure arrived to sell against it.
Meanwhile the Ad Platforms Took the Wheel
September brought a second shift, quieter and more immediately expensive.
Google began automatically migrating legacy Search campaigns to AI Max from 1 September. Microsoft made its equivalent generally available and switched it on by default for new Search campaigns. Neither required advertiser action, which is the point. Defaults decide behaviour for most accounts, and most accounts will not fight them.
The trade is the same in both cases and it is not unreasonable: automated systems beat manual settings on average. The cost is explanatory. When performance moves, you have less ability to say why, and less room to distinguish a campaign that is working from one that is spending efficiently on the wrong people.
What the Forecasts Got Wrong
Worth an honest audit, since the same predictions are being recycled for 2027.
Voice search. The claim that half of all searches would be voice-activated by 2026 traces back to a single 2016 prediction that was never supported and never came true. It has been repeated in trend pieces every year since. Voice is a real interface with real use cases, and it is nowhere near half of anything.
Augmented reality. The oft-quoted figure that AR makes shoppers dramatically more likely to purchase has no traceable methodology behind it. Meanwhile Meta closed its Spark platform in January 2025, removing third-party AR effects from Instagram and Facebook entirely. The forecasts had AR expanding. The largest AR distribution channel in social shut down.
AI personalisation. Directionally right, wrong mechanism. The prediction was better targeting of ads at individuals. What arrived was AI mediating the discovery step itself, so the contested ground is whether your brand appears in the answer at all.
What to Do This Quarter
Four things, none of them speculative.
Check whether your campaigns were migrated to AI Max, because it happened by default and the notification was easy to miss. Establish a baseline for how assistants currently describe your brand and your category, so you can measure drift once researcher data starts producing real findings. Watch for the ad repository going live around the end of December, since it will be the only public record of who is buying placement in AI answers. And treat every generative engine optimisation claim you are sold as provisional, because the evidence base for it is about to change.
Conclusion
The forecasts were not wrong about AI mattering. They were wrong about where it would matter, and they missed that the decisive move would come from a regulator rather than a product launch.
The useful lesson is about the genre itself. Annual trend pieces reward plausible extrapolation, and plausible extrapolation is exactly what misses discontinuities. A designation nobody was modelling has done more to define marketing’s next two years than any of the four trends everyone agreed on.
Read Next
- Personal Branding on Social Media: What Actually Builds Authority
- How to Evaluate an SEO Agency: What to Check Before You Sign
- How the Instagram Algorithm Works in 2026: Signals, Surfaces, and What Actually Changed
