In recent months, AI platforms have flirted with allowing advertisements to appear in AI-generated responses to user queries within LLM-powered chatbots.
OpenAI launched a test phase for ads in ChatGPT last month, with Criteo signing on as its first technology partner. While the partnership lends credibility, details remain unclear. Google is also inching toward implementation, building up capabilities for its AI Max for Search tool. Meanwhile, Perplexity has backed away and Anthropic is spurning the idea entirely.
Marketers might accept that this is the direction the industry is heading, but there’s still concern over what performance these platforms can report on their own. Leaders in ad tech and media are thinking through the risks, especially from brand safety and measurement standpoints.
‘Trust but verify’
Nada Bradbury, CEO of AD-ID, believes AI search ads are going to get pushed out before they’re fully realized in an effort to beat the competition. Despite concern from clients, the lack of clarity or regulation won’t slow adoption in this early phase, though it will be expected to evolve its audience measurement and verification capabilities to retrofit industry standards.
“People haven't waited to see an actual purpose or use before getting behind [AI],” Bradbury told Campaign. “People are going to trust that it's going to give them a new avenue to secure more sales and get more exposure.”
She predicts initial ROI measurements will be “sporadic and minimal,” “tightly controlled” by the platform and they’ll keep it that way until marketers start demanding proof that real consumers were reached and driven to purchase, noting that third-party integration isn’t as much a priority as “getting this product out the door.”
“Eventually, we'll get to a place where it’s measured by a third party across the board. But I don't think it'll happen very quickly,” said Bradbury; they're going to “try to protect as much as possible. … It's what we see with a lot of walled gardens today. They're just not willing to let that data escape.”
Richard Raddon, cofounder and co-CEO of Zefr, agrees that marketers will take the “trust but verify” approach to assess the addressability of the AI search ad environment. He doesn’t believe the platforms need to offer much capability at this stage; marketers will give them time to figure it out. But if they want to scale their ad revenue, they’ll need to provide “all the bells and whistles that every marketer is accustomed to getting in every other environment.”
“Test budgets only get you so far,” Raddon told Campaign. “You really hit the big time when you get into CPG, people that are far more ROI-driven and way more disciplined … that's when you really know you've got a scalable ad business.”
Keith Turco, CEO of Madison Logic, predicts that AI technology will help develop new metrics and collection methods entirely.
“There's really no branding element to the way LLMs deliver results,” Turco told Campaign. “The new way [to measure] will have to be performance-based — how they enhance a user's ability to connect with a brand.”
Intimate settings
Raddon noted that the transactional nature of SEO — “I buy keywords, then I show up” — provided a “veil of privacy” that could be lost in the AEO environment. Users are already developing “way more intimate” relationships with LLMs due to the personal nature of the conversations, making safety and suitability much more important.
“The big concern for the platforms is: How do we ensure that users feel like their conversations with the chatbot are private in nature, but also super relevant for the advertiser?” said Raddon. “That ultimately is what Anthropic is poking fun at. They’re saying … ‘because it's a different relationship, we don't think you should ever introduce advertising, because it starts to come off as insincere.’”
Anudit Vikram, chief product officer at Channel Factory, noted the brand safety issues arising within chats that stray into the realm of medical, legal or financial advice, plus the risk of appearing beside hallucinations and misinformation.
“Advertisers and agencies need explainability,” Vikram told Campaign. “In an environment where monetiSation operates inside the same interface that interprets and ranks intent, visibility into decision logic is not optional. Suitability must operate in real time.”
Vikram advocated for clear guidelines that prioritise trust with both consumers and advertisers, including labelling of sponsored content, disclosure to advertisers of paid and organic brand mentions and, crucially, suppressing monetisation in sensitive categories and low-confidence responses.
“The real technical work is not the protocol layer. It is defining a policy and decisioning layer that determines what the agent is allowed to do .… Governance cannot be layered on after the fact,” said Vikram.
AI platforms will likely offer content filters that will suffice for most marketers, following the social media playbook, though Raddon suggested third parties would need to pick up the slack in this realm as well; competition with China will disincentivize industry- and government-level regulation.
“I don't think … there's going to be any appetite to regulate these companies at all,” Raddon told Campaign. “Responsible AI has to be a marketer's concern. … The social media companies have trained [us] to think, ‘When my brand's out in the public square, how do I want to interact?’”
“Marketers are desperate to harness people's attention in these environments,” he said, adding, “If they don’t do it smartly, I think there’s going to be carnage.”
Click-through or clickless?
Experts disagreed on whether AI search ads would continue to erode click-based traffic, as many marketers had seen happen with Google’s AI Overviews.
According to Vikram, LLMs will likely remain a clickless environment, in which selection for inclusion in responses becomes the marketing signal.
“The monetisable unit shifts from pages and impressions to intents, tasks and decision moments inside conversations,” said Vikram. “The primary performance event is not click-through rate. It becomes ‘recommended,’ ‘shortlisted,’ ‘chosen’ or ‘purchased.’”
However, to Turco, a call to action is still necessary, and clicks remain the clearest example.
“The only way to monetise is measurability. The only way to measure is based on action,” he said. “An easy way to start would be the traditional sense of allowing clickability … so you can track and trace that the LLM drove traffic.”
Still, click-through rates in ChatGPT’s test phase were found to trail Google’s significantly.
“I'm curious to see where else they evolve to prove ROI in a way that helps them monetise their platforms,” said Turco, predicting they’ll go “straight to the way we currently measure.”
“Here we are … still buying and measuring on CPC and CTA,” he added. “We haven’t really evolved as much as one would have thought, technologically, over the past 15-20 years. I'm hoping it allows us to evolve in new ways of thinking and measuring.”
Source: Campaign US
