I’m not a major LLM user, in general, though I often put some generic shopping prompts through the major systems (ChatGPT, Gemini and Claude, namely) to see what comes out the other side. Mostly it makes me curious about how a big brand marketer might see these theoretical (and real) customer journeys play out in real time – or see, for instance, a competitor getting top placements for in-category searches – and say to themselves, “What the heck can I do to get in the mix here?”
If a brand wants to improve its search visibility, there are all sorts of well-understood mechanisms – paid, organic and earned – to do so. Want better social metrics? Or placement on app store charts? Or in Amazon results?
Those platforms aren’t free, but they have well-marked tollbooths, and every marketer understands that there is a playbook to produce the right content and pay the right sum to appear in their marketplaces.
But for a brand marketer whose CEO says, “We need to appear in these customer journey recommendation engines, and we’re willing to pay for it,” what can they do?
To get an answer, I spoke with a couple of advertisers and two different agency buyers who are participating in the ChatGPT Ads pilot, as well as an ad agency holdco exec who is not spending on ads with ChatGPT but saw the criteria ChatGPT is using to determine its first ad clients.
The easy ones
In general, marketers should sheepishly acknowledge to their CEOs that there’s very little they can do to affect LLM results, even with an open wallet. There is really no paid mechanism to show up in AI responses. The brand can consent to AI bot crawlers and to make its metadata and API feeds as ingestible to LLMs as possible. But that’s mostly the extent of their powers.
The exception is Google’s AI Overviews (AIOs), the AI-generated responses that sometimes appear atop general Google searches.
Google is the only platform that’s truly flipped the switch with advertising. ChatGPT is testing narrow ad use cases with specific brands and retail partners; Google AIO ads are open to all advertisers and, in fact, cannot even be turned off or excluded by advertisers that use Google Search, Google Shopping or Performance Max.
ChatGPT is still looking at drops in a petri dish, while Google has turned on a fire hose.
Ironically, according to one marketer who is an alpha brand partner in the ChatGPT advertising test, the Google AIO ads are going full steam but cannot be scrutinized whatsoever. A PMax or Google Search advertiser doesn’t see in its campaign reports if, when or how often its ads were served to AIOs. Nor is there any way to tie those impressions to conversions (other than trusting Google’s self-attributed ROAS).
AIOs are probably serving tens or hundreds of thousands of impressions that the company can’t analyze at all, he jested, while ChatGPT serves a few thousand ads that are intensely studied. What they have in common, he added, “is that they probably cost a couple hundred thousand dollars each.”
And while ChatGPT Ads has the spotlight right now, Amazon’s Rufus chatbot is the next likeliest to uncork ad revenue in a major way. Adweek reported last month that Rufus has a pitch deck to sell ad placements, which currently feature products only organically. And as with Google, Amazon will quickly open the shopper agent ads to its entire advertiser catalogue.
There is “zero implementation effort,” according to the Rufus ad sales pitch deck obtained by Adweek. “All existing and new Sponsored Products and Sponsored Brands campaigns in the US are automatically eligible” for the new agentic ads – which take the form of sponsored prompts within the chatbot or on product detail cards.
Which means Amazon isn’t going to spend a year or more nibbling around the margin of AI chatbot ads. It’s going full fire hose, just like Google.
The hard case
When marketers talk about getting in front of agentic shoppers and AI search engines, they’re generally talking about ChatGPT.
But ChatGPT is mostly not interested.
ChatGPT is judiciously building up demand in specific verticals, if not among specific brands.
For example, when Target announced its selection as an initial ChatGPT Ads partner, it boasted that “Roundel is unlocking the opportunity for brand partners to show up in relevant moments beyond traditional media placements.”
Many brands carried by Target likely cheered the news.
However, their excitement was premature. Roundel’s portfolio of advertisers are not all serving ads within ChatGPT. There are only a “handful” of vendors from Roundel in the ChatGPT pilot, a Target spokesperson told AdExchanger.
Rather, when a ChatGPT user has a shopping query about a specific item, and if Roundel has a suitable advertiser that is also a ChatGPT pilot partner, then that ad might be served.
This explains why ChatGPT Ads has largely failed to deliver on the overall traffic quantity that advertisers expected.
Both agency advertisers I spoke to said it was frustrating to have such large commitments from clients that are going unspent. Clients put up $200,000, and those advertisers would be happy to see 2,000-3,000 total ads serve. So to be tracking well behind even that modest benchmark “is like sitting on a goldmine and all you have is a beach toy shovel.”
The ChatGPT Ads numbers “have been a trickle,” said one alpha brand advertiser client. Although both he and another direct ChatGPT brand advertiser I spoke with concurred that ChatGPT traffic has been unusually lucrative.
ChatGPT ads have “so far generated markedly larger average cart sizes,” said the same marketer. In other words, he said, shoppers coming via ChatGPT ads are more likely to convert in that session and add more items to their cart than even an organic direct site visitor.
Wow!
Although the limited scope of the ChatGPT Ads beta is directly related to its strong results. ChatGPT knows it has X amount of shopping intent for, say, gaming laptops or air fryers. With one or two pilot partners, it can generate fantastic results. But if the program were to open to all laptop, gaming hardware and kitchen appliance brands, then nobody would excel.
One advertiser compared it to AppLovin’s ecommerce beta program, which has been in a closed, application-only beta since 2024. Meta’s Advantage+ Shopping Campaigns likewise slow-rolled an applicant-only beta program for a year and a half, onboarding new advertisers in seemingly random tranches so its AI platform performance could support the demand.
ChatGPT is also eschewing some of the lowest-hanging fruit, in terms of juicing the short-term value from ad revenue.
For instance, in December 2025, Criteo hosted an investor and press event where it debuted the work from a partnership with a yet-to-be-named LLM, which was using Criteo’s product recommendation tech and catalogue feed to help inform a nascent ad business. How might an LLM gauge the value of a product’s relevancy to the user’s search or in-session series of prompts, the current trendiness for online sales of a product and how much more that brand is potentially willing to pay for the placement? And how might a retailer working with ChatGPT make those decisions when placing ads on behalf of its brands?
Those are the types of questions Criteo engineers are working on, CEO Michael Komasinski said at the time.
But what that LLM was not doing was retargeting based on purchase history or web-browsing history elsewhere.
Criteo refused to name the LLM in question that it was working closely with behind the scenes, although in its presentation the “theoretical” LLM used to demonstrate how the product worked was ChatGPT, and last month Criteo was named as ChatGPT’s first third-party ad tech vendor partner.
Criteo declined to comment additionally on its ChatGPT pilot.
But, long story short, don’t expect ChatGPT to truly open its platform to advertisers, likely even within the next year or more.
As one agency holdco exec who’s participating in the ChatGPT Ads beta told me, if your CEO wants to figure out how to get in front of ChatGPT shoppers, a marketer’s best option is “to place ads somewhere else asking people to go to ChatGPT and prompt for their brand.”
