CASTE STRATEGY
Let's talk ↗︎

Blog / Trend

Advertising on ChatGPT in Italy: how ChatGPT ads work and when they're worth it

Since 31 August 2026 a company based in Italy can advertise on ChatGPT on its own. How it works, who sees the ads and when it's worth it, before you spend anything.

· Andrea Castegini · 6 min read

Yes: today a company in Italy can advertise on ChatGPT, and it can do it on its own. Since 31 August 2026, Ads Manager, the tool for creating campaigns, has been open for self-service in Italy too. On 5 October OpenAI announced a new visual format and new measurement tools. But the real question isn't "can we?". It's "is it worth it for us, now?".

The short answer: it's worth it if you sell something people ask ChatGPT about while they're making up their minds, and if you already know how to measure what a customer brings in. For everyone else it's a small test to run, not a budget to move.

What happened, in order

How it works, without the jargon

Ads are shown only to people who use ChatGPT for free or on the Go plan. Plus and Pro subscribers don't see them. Ads are labelled as sponsored and kept separate from the answer.

Three of OpenAI's rules matter if you're investing:

  1. You can't buy the answer. OpenAI states that ads do not influence ChatGPT's answers. If your product isn't in the answer, paying won't put it there: it appears next to it, as an ad.
  2. Advertisers don't see conversations. They only receive aggregated data, such as views and clicks.
  3. Some topics are off limits. No ads next to conversations about personal health, mental health or politics, and no ads for accounts identified as belonging to under-18s. Health advertisers can advertise if they meet strict eligibility criteria.

In Europe there's an important difference from the United States: personalised ads are not switched on in the European Economic Area or Switzerland. The ad is chosen based on the current conversation, the language and the approximate location, not on the user's history. On 2 October 2026 the data protection authorities of Hamburg and Berlin warned about the personalisation OpenAI plans to introduce in Europe: there's no start date yet.

When it makes sense to try it

People ask ChatGPT specific things: which software to choose, how to plan a trip, what to buy for the home, how to solve a problem. If your customer arrives with a question like that, the ad reaches them while they're thinking it through, not while they're scrolling a feed.

It makes sense to try it if:

It makes less sense if you work in health (ads don't appear next to personal health conversations, and health advertisers must meet strict eligibility criteria), if you only sell to people who already know you, or if today you can't say where your customers come from. In that case, the problem to solve comes before the channel.

Advertising and GEO are not the same thing

It's the easiest mistake to make right now. Appearing in ChatGPT's answer and appearing next to it as an ad are two different jobs. The first is built with a clear website, content that answers real questions and a company described consistently everywhere: that's GEO, which we wrote about in this article. The second is bought.

The two help each other: an ad that leads to a confusing page wastes money, and a clear page performs better both in answers and in ads. But paying doesn't replace the first job.

How to run a test that tells you something

  1. Pick one question. Write down the exact sentence one of your customers would type into ChatGPT before buying. Ask it yourself, and see what ChatGPT answers today and who gets mentioned.
  2. Build a page for that question. Not the home page: a page that answers in two lines and asks for one action only.
  3. Decide in advance what you'll measure. Quote requests, sales, bookings. OpenAI has added integrations with a range of measurement and attribution tools: set up tracking before you start, with privacy consent in order.
  4. Set a duration and a threshold. For example, four weeks and a maximum cost per lead above which you stop. Decide it beforehand, not once the campaign is over.
  5. Compare it with what you already do. The number that matters isn't the click. It's the cost of a customer compared with Google and Meta.

The duration and threshold are an example of method, not a benchmark: adapt them to your sector and to what one of your customers is worth.

It's the same principle we apply to every decision: before you spend, test it small. A new channel isn't a good idea because it's new. It becomes one when the numbers from a test say so.

In short

If you want to know whether this channel makes sense for your company and how to set up a first test, let's talk.

Sources

All articles ↗︎