LLM visibility is the new version of appearing on the first page of Google Search results. No one wants to be invisible.
So when generative engine optimization (GEO) entered the conversation, marketers instinctively treated it like a new version of search engine optimization (SEO). But GEO doesn’t react to the same technical levers as SEO.
Research shows that LLM visibility can be influenced with consistently good marketing across touchpoints, because AI models absorb information in enough different ways that no single optimization tactic reliably moves the needle.
The problem is, these conclusions don’t give marketers the answers they are looking for.
**The data problem **
The core issue with LLM visibility platforms is that they are not using real data.
These platforms run synthetic queries at scale and organize the outputs into marketing dashboards. The results are limited to dozens or hundreds of permutations—far fewer than a real user set. The platforms then produce “scores” that create an illusion of precision.
In other words, they’re measuring performance against a small synthetic test, not real human queries.
The IAB’s measurement framework, released this month, backs this up: measurement programs running fewer than 50 queries don’t even qualify as directional. Most vendors on the market operate well below this volume, meaning GEO data is minimally sufficient to characterize a category, let alone rigorous enough to act on.
The GEO industry has a strong incentive to ignore this—and the consequences are already showing up in practice.
I have one client who uses four different LLM visibility tools simultaneously. Each one tells her something materially different about where her brand stands. I recently watched two agencies pitch a large enterprise bank on GEO strategy. Each arrived at opposite recommendations, both grounded in data that contradicted the other’s.
This isn’t isolated bad luck; as the IAB notes, two providers measuring the same brand can produce results that diverge enough to be unusable.
When the foundational measurement layer is this unreliable, building a content strategy on top of it is basically just guesswork with a dashboard, not optimization.
The content problem
Marketers have spent the last year asking “How do we show up in AI answers,” but chasing visibility with unreliable data only eats up valuable resources. A better question to ask is, “What happens when we do?”
A visitor referred by an AI platform isn’t a cold click. They’ve already had a conversation, asked follow-up questions, and formed a point of view before they ever land on your homepage. They arrive with context and intent that traditional search traffic never carried. Most websites aren’t built for this visitor.
Racing to improve AI visibility, marketers are making their content more LLM-readable. We’ve seen clients chase GEO scores by updating their websites into more “LLM-friendly” formats. As it turns out, these can actively hurt the on-site experience for humans, reducing engagement and conversions in the process.
The fix isn’t a content strategy, a GEO strategy, or a website strategy. It’s AI-native infrastructure, built for two audiences at once.
On the surface, the website adjusts based on what the visitor is likely looking for, instead of everyone starting from the same generic homepage. But there’s also a hidden layer underneath, written specifically so that AI systems, not humans, can read and understand it.
The marketing world is moving at a crazy pace, and it’s easy to get pulled into keeping up with every new score and tool.
But the brands playing that game are optimizing for a number nobody can agree on.