Paid placement does not move the index

A reasonable question arrived repeatedly after ChatGPT advertising opened to European advertisers on 24 August 2026: does an index of AI citations still mean anything once brands can buy their way into the interface? The short answer is that it means more, not less. The longer answer is worth setting out, because it clarifies what this index actually counts.

Two surfaces, one screen

An assistant's screen now carries two different things. There is the answer, generated from retrieval and training, which names brands, orders them, and often reduces a category to two or three options. And there is a sponsored slot, labelled as such and visually separated from the answer. OpenAI's documentation is direct about the relationship between them: ads do not influence the answers ChatGPT gives, and advertisers receive aggregate information only, with no access to conversations.

The index counts the first surface. A brand earns presence by being named in the answer; it earns rank by where in that answer it lands; it earns coverage by being named across engines rather than one. None of those things is purchasable through an ad platform, because the ad platform does not write the answer.

A benchmark is only useful if the thing it counts cannot be bought. That has become a feature.

What we observed at the launch

We compared cited-brand composition on our buyer-intent prompt set before and after the France launch, holding prompts and run counts constant. The composition did not move. Positions that shifted sat inside the run-to-run variance we already track for those prompts. This matches the stated product design, and it is the result a well-specified index should produce.

We report it anyway, because "the documentation says so" and "we checked" are not the same claim, and only the second one belongs in a benchmark.

Where advertising genuinely touches visibility

There is an indirect path, and it deserves acknowledgement rather than dismissal. Advertising generates traffic, conversation and sometimes coverage. Coverage in independent sources is one of the inputs that models lean on, and source authority is one of the five dimensions this index scores. So a large, sustained campaign can, over time, thicken the source base under a brand and lift citation as a second-order effect.

That path is real but weak per euro spent. It runs through third parties, arrives late, and cannot be steered toward the specific prompts that matter to a business. Working directly on the source base — independent coverage, consistent entity description, documentation the models can parse — reaches the same dimension without the intermediate step.

The plan boundary, which is the underrated detail

Advertising is served to Free and ChatGPT Go accounts. Plus, Pro, Business, Enterprise and Education display none. For consumer categories that boundary is unremarkable. For business categories it is decisive: in organisations that have deployed a business or enterprise plan, the people making vendor decisions never see an ad. They see the answer. That is the surface this index has been measuring since it started, and it is where the shortlist is actually formed.

What would change our position

Two developments would force a revision, and we would rather name them in advance than be seen to move quietly later.

The first is a format change: if sponsored content stopped being visually separated and began appearing inside the answer body, the boundary this index relies on would blur. We would then label sponsored observations distinctly rather than merge them into presence and rank.

The second is an evidence change: if measurement showed that advertising spend in a category systematically preceded citation gains in that category beyond what source-base growth explains, the causal story above would need revising. We are watching for it. We have not seen it.

The practical read

For anyone allocating budget between paid placement and citation work, the distinction reduces to something simple. Paid placement is rented, engine-specific, and restricted to two plan tiers; it works while the budget runs. Citation is earned, spans engines, and persists when spend stops. On short cycles and low tickets, rent. On long cycles where an assistant is compressing a market into a shortlist of two or three names, own the position — and measure it, because it is the only one of the two you cannot verify with an invoice.