ChatGPT Ads arrives in Switzerland: what to know before you start

You are buying less of a query than a place for a brand at the moment a need is being expressed. That is the real change in this new channel.

ChatGPT Ads is now available directly to European advertisers, Switzerland included. After several months of gradual rollout, OpenAI opens a new playing field for advertisers.

The ads are currently visible to Free and Go users. They appear separately from ChatGPT’s answers and do not influence the answers the model gives. In January we wrote that it was “not yet possible to pay ChatGPT to sponsor content”; since the ads stay separate from the answers, being cited in them remains a matter of Answer Engine Optimisation.

Three campaign objectives

The structure deliberately resembles that of Google Ads, Meta and the other large advertising platforms. A campaign holds ad groups, which themselves hold the ads.

Three objectives are currently available: Reach, to maximise visibility; Clicks, to generate traffic; and Conversions, to steer delivery towards the users most likely to act after their click.

The targeting: the real novelty

This is probably where ChatGPT Ads becomes particularly interesting.

Unlike Google Ads, you do not work only with keywords matching a query. Advertisers provide context hints, which let ChatGPT understand the conversations, subjects, needs or intents in which an offer may be relevant.

For Bright, for instance, we can work around terms such as digital marketing agency Switzerland, media agency or digital marketing Switzerland.

But the interest lies above all in being able to go further into how the need is expressed:

Context hints are not exact-match keywords. They let OpenAI understand the different situations in which an offer can answer the need the user expresses.

It is an interesting change: you are buying less of a query than a place for a brand at the moment a need is being expressed.

Geographic targeting is also available. In Switzerland you can target by canton, town or postcode.

There is, however, no documented language targeting as a campaign setting yet. In a multilingual market such as Switzerland, campaigns, context hints and ads therefore have to be structured accordingly.

What does an ad look like?

The format stays deliberately simple: advertiser name, favicon or logo, a headline, a description, a landing page and an image.

OpenAI already recommends running many ads, with different variations of headline and description, so as to give the system more scope to identify the right message for the context.

The ad and its landing page are analysed together: the advertising promise therefore has to match the destination.

And conversion tracking?

To optimise a campaign for conversion, you obviously have to be able to measure it.

OpenAI offers its Pixel and a Conversions API. As on the other large advertising platforms, actions taken on the site can therefore be passed back: purchases, leads, sign-ups or other conversion events.

That is why tracking is an integral part of the setup we recommend before any campaign launch.

Optimisation still limited

This is probably one of the main limits of ChatGPT Ads at launch.

While the platform lets you follow impressions, clicks, CTR and conversions, it does not yet give access to the equivalent of the Search Terms report in Google Ads.

In other words, we can tell ChatGPT the subjects, needs and intents on which we want to place a brand, but we cannot know precisely which search or which conversation triggered an ad, a click or a conversion.

For now, then, it is impossible to identify a particularly well-performing query and put more behind it, or conversely to exclude precisely the searches that consume budget without converting.

Optimisation happens more at the level of the ad groups and the intents tested, by looking at their CTR, their conversion rate, their CPA and their volume of conversions.

So you can steer by performance, but at a granularity still well below that of Google Ads. An important point to keep in mind for this new channel.

Careful: not every advertiser is eligible yet

This is another important limit of ChatGPT Ads today.

OpenAI is initially focusing the platform on categories such as consumer goods, local services, travel and experiences, digital products and education. Several regulated sectors remain heavily restricted.

Financial services are generally prohibited. That covers banks, investments, credit, mortgages, insurance and financial planning services.

For insurance, the situation is therefore worth looking at case by case. A purely informative approach may have more scope than a campaign directly encouraging someone to take out an insurance product. But eligibility is not guaranteed: the best way to settle it today remains to submit the case and test OpenAI’s validation.

Casinos, lotteries and other gambling activities are prohibited.

Property is allowed, with one important limit: promoting a specific individual home for sale or to let is prohibited. An agency can, however, promote its platform or its services, as long as neither the ad nor its landing page refers to a specific property listing. That opens the door in particular to campaigns for recognition, property valuation, searching for homes in an area, or general promotion of an agency’s services.

Legal services, alcohol, tobacco and nicotine, dating and sexual content, drugs and political content are also among the categories currently prohibited or heavily restricted.

Why test now?

Because we are still at the beginning.

We have today no Swiss benchmark robust enough to announce a CPC or a CPM in advance. And that is precisely why we began by launching our own Bright campaigns.

Our approach will be the same as on the other platforms: test, measure, understand, adjust, and scale when the results justify it.

For advertisers who want to begin, the aim is simple: test without committing to a heavy management setup, measure the first results, and then decide what place ChatGPT Ads can take in their media mix.

ChatGPT Ads is a new channel. The logic of performance does not change.