Observation · August 2026

Sponsored placement and cited answers: what changed when ChatGPT ads reached Europe

Advertising inside ChatGPT opened to French advertisers on 24 August 2026, part of a rollout now covering 31 countries. We looked for movement in the brands our panel cites. We found none — and the product documentation explains why.

What we checked

Our France B2B panel is scored on buyer-intent prompts across four engines. For this observation we compared the set of brands named in ChatGPT answers in the days before and after the France advertising launch, holding the prompt set and the number of runs per prompt constant. The question was narrow and answerable: did the composition of cited brands move?

What we found

It did not. The brands named on our cybersecurity and regulated-finance prompts after the launch are the same brands, in broadly the same order, as before it. Where positions shifted, the shifts sit inside the run-to-run variance we already measure for these prompts and are not attributable to the advertising change.

Why this is the expected result

OpenAI's documentation is explicit on the point: advertisements are clearly labelled as sponsored, visually separated from the organic answer, and "do not influence the answers ChatGPT gives you". Advertisers receive aggregate information only and have no access to conversations. A finding of movement would have contradicted the stated product design; a finding of no movement confirms it.

This is worth stating plainly because the market reaction to the launch has often assumed the opposite — that buying inventory inside an assistant would buy a place in what the assistant recommends. On current product behaviour, it does not.

The plan restriction matters more than the launch

The detail with the largest practical consequence is not the launch date but who is inside the inventory at all. Advertising is served to Free and ChatGPT Go accounts. Paid professional and organisational plans display none.

For a consumer brand, the free tier is a large and legitimate audience. For a B2B vendor selling into organisations that have deployed ChatGPT Business or Enterprise — common in cybersecurity, regulated finance and industry — the buyer is not expensive to reach through advertising. The buyer is outside the inventory.

Two economies, not one

The clean way to hold both facts together is to stop treating advertising and citation as competing tactics on the same surface. They are different assets.

• Rented exposure. An ad slot produces impressions while a budget runs, on one engine, on two plan tiers, in one country's inventory. Its value does not accumulate; it resets when spend stops.
• Owned position. A citation sits inside the answer itself, is produced across engines, rests on sources the models consult, and survives a budget pause. Its value compounds as the source base thickens.

Neither is superior in the abstract. Which one matters depends on cycle length and deal size. On a short-cycle, low-ticket purchase, rented exposure converts. On an eighteen-month enterprise cycle where the assistant produces a shortlist of two or three vendors, the position inside the answer is the asset.

What we are watching next

• Whether ad formats stay visually separated from the answer, or move toward placements integrated into the response body
• Whether the announced self-service platform changes who can buy, and therefore which categories fill the inventory
• Whether European targeting moves beyond contextual signals — currently the question asked, location and device type — toward the personalisation available in the US
• Whether any measurable divergence appears between cited brands and advertising brands within the same category

Our scoring method and panel construction are documented on the methodology page; the underlying figures are on the data page.