The Regulation Protects the Thing That Stopped Mattering
OpenAI launched advertising across 31 European markets with the targeting switched off. The constraint is the story — and it may protect less than it appears to.
OpenAI launched advertising across 31 European markets with the targeting switched off. The constraint is the story — and it may protect less than it appears to. What happened On 24 August 2026 , OpenAI began serving ads in ChatGPT across 31 European markets , six months after the US pilot began. Three things were constrained deliberately: Tier-gated. Ads appear for Free and Go users only. Plus, Pro, Business, Enterprise and Edu remain ad-free. Placement-constrained. Ads sit below ChatGPT's responses, are labelled sponsored , and are visually separated from the answer. OpenAI states they do not influence answers and that advertisers receive no access to user conversations. Targeting-constrained. Personalised ads are unavailable in the EEA and Switzerland at launch , because GDPR requires explicit consent for personalised targeting. Europe launched as a contextual advertising business. A note on precision: this is 31 European markets , not "the EU." Norway is on the list and is not an EU member. Why it matters The tier structure tells you what this is. Ads run on Free and Go — the tiers that cost OpenAI money and produce little or no subscription revenue. This is not a plan to monetise the user base. It is a plan to make the free tier fund itself. Which answers a question we left open four weeks ago. In our 11 August scan we examined OpenAI's unlimited free tier and asked who pays for unlimited free inference at population scale. The answer arrived thirteen days later: advertisers do. Unlimited free chat and an advertising business are one strategy, not two. For our audience the immediate consequence is a new acquisition channel opening at the precise moment a user articulates a need — with no established benchmarks and, in most categories, no crowded auction yet. Under the hood Here is the part worth thinking carefully about. What is a profile actually for? On a social network, the platform observes behaviour and infers a desire. That inference is lossy in both directions — people research mortgages for their parents and browse shoes they will never buy. Years of engineering went into narrowing the gap. The profile is the accumulated residue of that effort, and it exists because the user will not say what they want, so the system has to guess. In a chat interface, the user says it. Someone typing "my startup is running out of runway and I need to cut costs" has disclosed more commercial intent in one sentence than a year of browsing history could reliably infer. There is nothing left to guess. European data-protection law draws its line between contextual matching (permitted) and profiling (requires consent). That line was meaningful when "context" meant the article you happened to be reading, which told an advertiser almost nothing about you personally. It means something rather different when context is a paragraph you wrote about your own circumstances. The legal category did not change. The thing inside the category did. The question nobody h