Your citation counts are up, brand mentions are climbing, and the AI referral traffic dashboard looks better than last quarter. So why does none of it feel like it’s actually moving pipeline?
Because the problem is that AI visibility and AI influence are not the same thing, and most of the AEO market is measuring the wrong one.
When we ran our multi-vertical Dark AI research, we expected brands with higher visibility in LLM responses to also have stronger influence over buyer decision-making. What we actually found was more uncomfortable.
Getting cited or mentioned by an LLM means the influence already happened. It doesn’t tell you whether you’re currently building it. The brands celebrating their citation counts are looking backward. Those numbers tell you where the model ended up, not whether your content had anything to do with it.
Why BOFU Metrics Flatter You
To understand why high AI search visibility can be meaningless, you need to understand something called convergence. This is the process by which LLMs narrow down which brands they recommend as a buyer moves from exploring a topic to actually making a decision.
We measured this using what we call canon concentration scores, a 0 to 1 scale that tracks how often the model names the same brands when you run the same type of prompt multiple times at each stage of the funnel.
- At the awareness stage, the score sits at just 0.37. The model is in exploration mode. It mentions brands loosely, names different players across different runs, and talks in broad category terms. Only about 22% of responses name the same brand consistently.
- By the consideration stage, the score rises to 0.44. The model starts comparing options and evaluating trade-offs, and its language gets more specific. But there’s still enough variability that the shortlist isn’t locked in yet.
- At conversion, the score jumps to 0.82. The model has made up its mind. It names roughly the same brands every time, uses language like “best for,” “recommended,” and “top choice,” and almost never deviates. Variability is essentially gone.
This is also exactly where citations and retrieval finally show up. At the conversion stage, our data shows 94% brand mention rates, 48% retrieval, and 48% citation rates. At the awareness stage, mentions hit 60% but retrieval and citations are at literally zero.
So when you track citations as your primary AEO KPI, you’re measuring something that only shows up after the model has already picked its preferred brands. At that point, the decision is already made. You’re not learning whether your content shaped anything. You’re just learning whether you made the cut.
What Each Visibility Signal Actually Means (and Doesn’t)
The problem gets worse when you actually look at what each visibility type tells you.
Our research identified three separate signals that most people in the AEO market treat as the same thing.
- Brand mentions just mean the model knows your brand exists in a category. That’s it. A mention doesn’t signal authority, endorsement, or accuracy, and it rarely leads to actual selection. Think of it like being listed in a directory. You’re in there, but nobody is choosing you over anyone else because of it.
- Retrieval is when the model actively pulls your content while generating a response. That’s a stronger signal because it means the model finds your content credible enough to reference. But it still doesn’t mean you’ll be cited or recommended. The model is checking whether it can trust you, not committing to you.
- Citations and links are the rarest signal and show up almost exclusively at the bottom of the funnel. Even then, a citation is just the model showing its work. It already decided which brand to recommend; the citation is how it backs that up. Citations don’t signal preference, ranking, or likelihood of selection. They just mean the model needed a source it could point to.
Most of the AEO market treats these three signals like a ladder: mentions are good, retrieval is better, citations are best. Our research says that’s the wrong frame entirely. They’re three different behaviours with three different causes, and none of them tell you whether your content actually shaped how the model thinks about the problem or what criteria the buyer uses to evaluate options.
Where AI Influence Actually Forms
Influence lives in what we call Dark AI. This is the earlier part of the buyer journey where people use LLMs to explore a problem, understand a category, and figure out what they should even be looking for. And it all happens with no clicks, no referrals, and no consistent brand mentions.
No traffic doesn’t mean no impact. Say a buyer asks an LLM “what should I look for in a content intelligence platform?” and the model’s answer highlights criteria that happen to match your product’s strengths. The buyer now has a mental model that naturally points toward you by the time they’re ready to decide.
This is also why LLMs rarely introduce new brands at the bottom of the funnel. They’re drawing from what they already know. If your content wasn’t part of shaping how the model thinks about the problem during awareness and consideration, you won’t be on the shortlist at conversion. It doesn’t matter how many FAQ schemas you’ve deployed.
Which is exactly what makes citation-chasing tactics so counterproductive. Schema markup, snippet-ready paragraphs, listicle formats optimised for citation capture, they all target the stage where the model is most locked in and least open to being influenced.
What to Measure Instead
Our research points to three metrics that measure influence while it’s forming.
- Inclusion Before Convergence: How often does your brand show up in awareness and consideration prompts, before the model has narrowed things down? This tells you whether you’re in the conversation while options are still open.
- Criteria Alignment: When your brand gets mentioned, is it being linked to the things your buyers actually care about? This tells you whether the model connects you to the right buying criteria, not just whether it knows you exist.
- Information Gain: Is your content actually teaching the model something new, or is it repeating what the model already knows? We score this on a 0 to 3 scale. Level 0 means your content is paraphrasing ideas that already exist. Level 3 means it’s introducing a genuinely new way of thinking about the problem. Content sitting at Level 0 won’t change how the model frames anything. It just gets absorbed into the background.
Stop writing content aimed at winning citations at the bottom of the funnel. Instead, write content that shapes how the model compares options at the consideration stage. Give it balanced, specific comparisons that make your strengths the natural baseline for evaluating the category. Stick to specific use cases and constraints rather than broad “best overall” claims, which the model tends to ignore anyway.
And stop chasing specific prompts like they’re search keywords. LLM prompts are long and come in infinite variations. Trying to optimise your content against individual prompts one by one creates a bloated content library and often introduces contradictions that make your overall positioning less clear, not more.
BOFU is not where brands win. The framing that determines which brands make the final cut happens much earlier, in the Dark AI layer where traditional metrics pick up nothing. Your goal should be to shape how the problem is framed in a way that naturally plays to your strengths, downplays your weaknesses, and makes visibility the byproduct.
If you want the full data, the convergence mechanics, the concentration scores, and the complete measurement framework, the Dark AI report is where to go.
The brands that win in AEO won’t be the ones with the highest citation counts. They’ll be the ones whose content shaped how the model thinks, long before it ever cited anyone.
