The short versionOpenAI told advertisers late on Friday, October 16, that ChatGPT ads are getting new visual formats, including more prominent, full-width images on mobile and, soon, a multi-image format. Those tests use existing campaigns and creative, with no action required. Separately, Google is starting to test ads in AI Mode for users who connect their apps through Personal Intelligence. Google says those ads won't be based directly on Gmail or Google Photos content. Both are tests, not full rollouts.
Ads are moving into the answer box. In the same week, OpenAI made ChatGPT ads more visual and Google started testing ads in a more personal version of AI Mode. Both are tests. Here's what's actually happening.
What happened
OpenAI announced a set of ChatGPT Ads updates "late Friday," October 16, in an email to advertisers, Search Engine Roundtable reports. The new visual formats include wider cards, larger images on desktop, "more prominent, full-width images on mobile" and, "soon, a new multi-image visual format." OpenAI said these "tests use existing campaigns and creative, with no action required from advertisers." Coverage of the email says the multi-image format will be tested first during image generation in the US with select advertisers.
On the Google side, Search Engine Land reported on October 16 that Google is starting to test ads in AI Mode for users who connect their apps through Personal Intelligence. Connecting apps is opt-in.
The details
Google's limits. Google says these ads will not be based directly on the contents of a user's Gmail inbox or Google Photos library, and that advertisers can't access users' connected apps. Search Engine Land notes Google hasn't announced new advertiser controls or a wider rollout.
OpenAI's other updates. The same email covered a redesigned conversions page in Ads Manager, weekly performance emails and more measurement partners, including Northbeam, DV Rockerbox and INCRMNTAL, according to Search Engine Roundtable.
The one number. In an October 5 post, OpenAI cited a single partner result: "According to DV Rockerbox, WeightWatchers' attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark." That's one advertiser, not an average, and no spend or dates were published with it.
Why brands should care
People are asking AI tools the same questions they used to type into search: what to buy, where to eat, who to hire. If ads sit inside those answers, that's new inventory, and early inventory is usually less crowded. But the formats, controls and reporting are still changing week to week.
What to do
Our suggestions, not platform guidance:
- If you already run ChatGPT ads, check your images. The new formats use existing creative, so make sure your images look good large and full-width on a phone.
- Test small. Treat this like any new channel: a capped budget, a clear conversion goal and a few weeks of data before you decide.
- Don't plan around 15.3%. It's one brand's result. Measure your own.
- Keep your basics clean. AI answers pull from your website and listings, so make sure prices, hours and services are current.
Sources
- Search Engine Roundtable: ChatGPT Ads new ad formats, updated conversion page, performance emails (Oct 19, 2026)
- Search Engine Land: Google begins testing ads in AI Mode for users with Personal Intelligence enabled (Oct 16, 2026)
- OpenAI: New ChatGPT ads format and measurement (Oct 5, 2026)
FAQ
Do I need to change my ChatGPT ad campaigns for the new formats?
No. OpenAI said the visual format tests use existing campaigns and creative, with no action required from advertisers.
Are Google's AI Mode ads based on my Gmail or Photos?
Google says ads for Personal Intelligence users will not be based directly on the contents of a user's Gmail inbox or Google Photos library, and connecting apps is opt-in.
Are ChatGPT ads cheaper than search ads?
There's no industry figure. OpenAI cited one advertiser: per DV Rockerbox, WeightWatchers' attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark.
