Update of 13 October: an image format, in the US for now
On 5 October OpenAI announced a new ad format built around images. It is being tested first during image generation in ChatGPT, over the course of October and only in the US, with an initial group of advertisers. According to OpenAI these ads, too, are labelled and kept separate from the generated image, and the rule that ads do not influence ChatGPT's answers applies to them as well.
Alongside it come tools meant to show advertisers what their ads achieve: connections to services for conversion data and attribution, and first attempts to measure the added effect through regional experiments. For a business here, everything described below stays as it is for now.
What is new
OpenAI has been testing ads in ChatGPT since 9 February 2026, starting in the United States. Its own country list now marks Germany, Austria and Switzerland as available: advertisers here sign up in Ads Manager directly, provided the legal entity being billed sits in one of the listed countries. Trade press dates the European rollout to 24 August 2026.
Where the ads sit
OpenAI describes the placement precisely. Ads can appear below the end of a response, they are clearly labelled as sponsored and visually separated from the response. How many appear depends on the conversation; in longer threads the system weighs the overall context.
Who actually sees them
Ads appear for accounts on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu stay ad-free, as do accounts OpenAI identifies as belonging to people under 18, and temporary chats. Logged-out visitors can see ads as well; there, according to OpenAI, only ads suitable for all audiences are shown.
In Germany there is one addition, in that personalised ads are initially excluded across the European Economic Area and Switzerland. Selection here runs on the current conversation plus coarse signals such as language and general location, rather than on past chats and stored memories.
Where the advertiser's influence ends
The influence ends at the ad itself. OpenAI describes advertising and the chat model as separate systems: advertisers shape their ad, and the answer above it is produced independently of it. One sentence from the help page matters more to readers than to advertisers: seeing an ad does not mean OpenAI endorses that advertiser.
For work on visibility that leaves a clean split. The place inside the answer and the place below it are two different things, and only the second one carries a price.
What we measured ourselves
We query ChatGPT for our clients while logged out, so that no account and no stored memory colours the result. Across the 34 runs between 7 and 14 September 2026, not one carried an ad. A fresh probe on 28 September, asking which roofer in Lüneburg does roof renovations, produced the same picture, and so did three runs on 13 October, already on the new GPT-6 model, asking where to buy a bike in Lüneburg.
What did appear is worth naming, because it is easily mistaken for advertising: a map with business cards, star ratings and photos, and below it the answer naming five businesses along with its sources. Those cards are part of the answer itself; the sponsored label belongs elsewhere.
That holds for the questions we ask; it says little about the market as a whole. For a trade in a town this size there are presumably few advertisers bidding on such conversations.
What this means for a local business
Since late August two routes lead into a ChatGPT conversation, and they work differently. Being named still follows from what can be found about a business online, which is exactly the work we do. Being shown is a purchase through Ads Manager, with a budget and a run time.
Anyone weighing ads is better off settling the other question first: does the business turn up in the answers its customers are already asking? That takes ten minutes to check, and it costs nothing.