Major brands are racing to optimize their presence in large language models (LLMs), but they're confronting an uncomfortable reality: Traditional measurement frameworks don't work, and executives demanding ROI metrics may be asking the wrong questions.

"These are the conversations I have to have: They're like, 'Well, where's the money, where's the numbers,' and I'm like, 'They don't exist right now,'" said Jennifer Mennes, vice president and global head of digital marketing and strategy at Mondelez International, during EMARKETER's Ad Buyer Strategies Summit. "For someone who's been in media for almost 30 years to say 'I can't tell you what the ROI is,' it's a very difficult conversation."

Marketing leaders from Mondelez, Bayer, and Seer Interactive discussed how generative engine optimization (GEO) is forcing brands to rethink everything from content strategy to measurement frameworks.

Incumbent brands aren't winning by default

Even the world's most established brands are discovering that decades of marketing investment doesn't guarantee visibility in LLM responses.

Oreo, the number one cookie globally by sales and market share, doesn't rank first when consumers ask LLMs for cookie recommendations, Mennes said. According to her, the issue stems from how brands have prioritized content.

"The LLMs are definitely looking for more organic, the influencer, the recipe content, and the website content," she said. "We've de-prioritized that for more brand advertising and commerce advertising."

The problem extends beyond individual brands. Wil Reynolds, founder and CEO of Seer Interactive, noted that LLMs show a 26x bias toward US results in his firm's research.

"If the training models are learning about your brand on US data, US dosage levels, what ingredients you use from country to country, then your people in those countries could get wrong answers," he said.

Brands are losing control over their narratives

Unlike traditional search where paid advertising could compensate for weak organic rankings, LLMs present fewer control mechanisms.

"We get to control media buying. We get to control how our brand shows up in various environments, to some degree," said Peter Sloterdyk, CMO of Gist. "We're used to being in control and having access to levers to pull up and down, or to invest here, to pull back there, and the LLMs aren't at our beck and call."

The variation in responses compounds the challenge. A single question like "What kind of face wash should I use?" can generate 35 different answer variations across five major LLMs, Sloterdyk noted.

Reynolds highlighted another control issue: timing. "If the training data cut off was January of 2025, and we are in May of 2026, you cannot even judge the work that we tried to do to optimize you, because it's been 16 months of you waiting," he said.

Brands need to audit what LLMs are saying about them before investing in advertising within these platforms. In one case Reynolds cited, a client's ads claimed "we're the best at this" while the LLM response above stated the brand wasn't even a consideration. "That's actually probably not the thing you want to be advertising," he said.

Retail media becomes a priority

Facing measurement challenges in open LLMs, brands are focusing resources on retail media networks where they can track performance and influence purchase decisions.

"Twenty percent of our shoppers in the future are going to be agents. They're not going to be humans," Mennes said. Mondelez is prioritizing optimization within Amazon, Walmart, and other retail platforms where search happens closer to purchase.

The strategy reflects practical realities about consumer behavior.

"If you're really familiar with allergy products, you're going straight to Amazon or Walmart," according to Maria Givens, vice president and head of media and digital platforms at Bayer. But discovery-oriented queries about new health concerns still happen in open LLMs, requiring brands to maintain presence in both environments.

Marketing leaders are pushing back against demands for traditional ROI metrics, arguing that speed and quality matter more than phantom measurements in GEO's early stages.

"I'm kind of rethinking how I might want to galvanize my team," Givens said. "Instead of focusing on some phantom metric that we don't actually know delivers business outcomes, I actually really just want them to focus on pumping out quality content to influence LLMs from many different angles."

Brands need to secure executive agreement on directional progress rather than precise ROI.

"Can we just all agree this is where the North Star is, and let's figure it out along the way," Mennes said. The alternative is paralysis while competitors move forward.

Hidden GEO value

Traditional brand-building activities are showing unexpected benefits in LLM visibility that aren't captured in current ROI calculations.

Research by Seer Interactive found approximately a 3x increase in LLM visibility for banks in cities where they sponsored stadiums. "If you're sponsoring a stadium, are you including in your ROI equation the bump that you're going to get in LLMs in that region?" Reynolds asked. "Most people aren't yet."

Brands should expand ROI frameworks for sponsorships, partnerships, and other brand-building activities to account for their influence on LLM training data and responses. As Reynolds noted, "We can't track much of it. So the stuff you can track, you might want to."

Watch the session.

This was originally featured in the Retail Daily newsletter. For more retail insights, statistics, and trends, subscribe here.