ChatGPT Ads in Italy and Europe: what can an initial pilot prove?

OpenAI expanded Ads Manager Beta to Italy and thirty other European markets on 31 August 2026. Eligible advertisers can create self-service campaigns with CPM, CPC, and oCPC objectives. For euro accounts, the stated minimum daily budget is EUR 15, but this is a technical threshold. It does not prove that the amount can produce a useful sample.

The word beta matters. Availability, inventory, formats, targeting, and eligible categories can change. A sensible plan does not build annual forecasts on conditions that are still moving. It defines a bounded question, an observable period, and a rule for deciding what happens next.

In search advertising, the starting point is often a short query organised into keywords, match types, and groups. In a conversation, the need develops across several turns: objective, constraints, comparison, doubt, and possible decision. This does not make the channel automatically better. It changes the design unit from the searched term to the context in which a message may be relevant.

Ads Manager supports context hints: topics, words, or examples of relevant conversations. OpenAI states that they are not exact-match keywords and do not guarantee delivery in a particular conversation. Useful work therefore means formulating context hypotheses, not renaming old keyword lists.

That is not a responsible promise. OpenAI states that ads are separate from answers and do not influence what ChatGPT says. Products uploaded through an advertising feed can become eligible for ads, but they do not automatically enter organic conversations.

Paid visibility and organic AI visibility can be observed together, but they need separate measurement. A campaign answers questions about impressions, clicks, cost, and attributed conversions. Organic analysis answers questions about presence, citations, sources, and competitors. Combining the two makes it impossible to attribute a change to the correct cause.

Ads Manager reports impressions, clicks, spend, CTR, average CPC, average CPM, and conversions. Conversions can be sent through OpenAI Pixel, Conversions API, or both. When the same event is sent through both systems, the same event ID enables deduplication. Consent, privacy notice, CMP, and shared first-party data need review before activation.

These figures do not prove incrementality on their own. OpenAI notes that totals can differ from first-party analytics and may include modelled conversions. A pilot should therefore declare its window, event, attribution model, and comparison. The result is not an isolated ROAS figure, but a reading with limitations clear enough to support a decision.

The advertising account should be created by the client using its legal entity, tax details, and billing information. The client can then invite consultants or collaborators. This keeps access, history, and responsibility inside the company and makes it easier to change suppliers without losing control of the channel.

Not every category is eligible. OpenAI policies include exclusions and restrictions, particularly for sensitive or regulated sectors. Category, market, message, and landing-page review comes before creative production. A pilot should not start until it is clear what can be advertised and under which conditions.

A useful pilot starts with one commercial hypothesis: for which need, in which context, and with which proposition the brand could be relevant. It then connects that hypothesis to a coherent page and an observable event. Adding many audiences, messages, and products immediately may increase volume, but makes it harder to understand what worked.

The final readout should support three legitimate outcomes: expand what shows a consistent signal, correct the message or journey when relevance does not become action, or stop when cost and quality do not justify more budget. In a beta channel, stopping for a documented reason is more valuable than continuing by inertia.