One number, three completely different fates
Here's the trap. Your AEO dashboard shows visibility climbing, the chart is green, somebody puts it in the board deck. Then sales asks why nothing downstream changed, and nobody has a good answer.
The reason is that most AEO reporting folds three distinct outcomes into one blended score. There's being mentioned, which can mean an AI assistant named you once as a caveat, as in 'some teams use Brand X, but it's pricier.' There's being cited, where the model pulls a fact or a stat from your page as supporting evidence but still recommends somebody else. And there's being recommended, where the answer actually tells the user to go with you.
Those are not three sizes of the same thing. They're three different things. A caveat mention and an outright recommendation sit at opposite ends of the buying funnel, and a single average smooths right over the difference. When the blended score goes up, you genuinely cannot tell whether you won more deals or just collected more footnotes.
What each outcome is actually worth
| Comparison category | Mentioned | Cited | Recommended |
|---|---|---|---|
| What the AI is doing | Names you in passing, often with a qualifier | Pulls a fact or figure from you as evidence | Tells the user to choose you |
| Typical phrasing | "Options include Brand X..." | "According to Brand X, the average is..." | "For your use case, go with Brand X." |
| Buyer intent signal | Low — awareness at best | Medium — authority, not endorsement | High — purchase-adjacent |
| Moves a deal? | Rarely on its own | Sometimes, builds trust | Yes, this is the one that converts |
| Easy to inflate? | Very — caveats are cheap | Somewhat — depends on your data | Hard — has to be earned |
Why the blend flatters the wrong win
Caveat mentions are the cheapest outcome to accumulate and the easiest to mistake for progress. Models hedge constantly. They list alternatives, they add 'some people prefer,' they hedge their hedges. If you only count whether your name appeared, you'll rack up mentions fast and feel great about it.
But a mention that exists only to warn the reader off is worth close to nothing, and in a blended score it looks identical to a recommendation. That's how a team ends up celebrating an 18% lift that's made almost entirely of soft references, while the recommendation rate that actually drives pipeline sits flat at 11%.
Crescive refuses to blend them. It tracks mentioned, cited, and recommended as three separate metrics, scored per answer and per model, so you can see which one moved and which one didn't. When recommendations are flat and mentions are up, you know the real work is converting those hollow references into endorsements, not patting yourself on the back for the aggregate. And because Crescive drafts the fixes behind a human approval gate, you can target the specific gap, like the comparison page that keeps getting you cited but never recommended.
The gap hiding inside a healthy-looking number
62% mentioned, 11% recommended
In this illustrative view of your dashboard, the brand shows up in nearly two-thirds of answers but is actually recommended in roughly one in nine. A blended score reports both as the same kind of 'visibility,' which is exactly why the 51-point gap goes unnoticed until sales starts asking questions.
What to do with three metrics instead of one
Start by separating them in your own reporting, even before you change tools. For every tracked query, log three booleans: were you named, were you cited, were you recommended. The spread between those three columns tells you where you're stuck.
High mentions and low recommendations means you're on the radar but losing the head-to-head. That's usually a positioning and proof problem, not an awareness problem. High citations and low recommendations means models trust your data but hand the deal to a competitor anyway, often because your use-case fit isn't legible to the model. Low across all three means you're not in the consideration set at all, which is a different fight entirely.
Each pattern needs a different fix, and you can only see the pattern if you stop averaging the three together. The single-number dashboard was never going to tell you which lever to pull.
Key takeaways
- Mentioned, cited, and recommended are three different outcomes with three different values; a blended visibility score hides which one actually moved.
- Caveat mentions are cheap to accumulate and easy to mistake for progress, so a rising aggregate can be mostly hollow.
- Track the three as separate metrics per query and per model — the gap between them tells you whether your problem is awareness, proof, or positioning.
FAQ
What's the difference between being mentioned, cited, and recommended in an AI answer?
Being mentioned means an AI assistant names your brand, often in passing or as a caveat like 'some teams use Brand X, but.' Being cited means the model pulls a fact or figure from your content as supporting evidence, usually while recommending a different option. Being recommended means the answer explicitly tells the user to choose you. They sit at different points in the funnel and carry very different value, with recommendation being the outcome most likely to move a deal.
Why is a single blended AEO visibility score misleading?
A blended score averages mentions, citations, and recommendations into one number, so it can rise even when the growth is entirely low-value caveat mentions and your recommendation rate stays flat. Because a passing reference and an outright endorsement count the same in the aggregate, the score can look healthy while nothing downstream changes. Tracking the three outcomes separately shows which one actually moved and tells you whether to fix awareness, proof, or positioning.