Four Events, One Dashboard Number
A visibility score promises one number you can watch go up. The AEO dashboards selling that score don't agree on what makes a brand visible inside an AI answer, so they fold four different events into the number and ship it as a single trend line.
The four events aren't interchangeable. A mention is your brand's name appearing somewhere in a generated answer, no link, no endorsement, just present in the text because the model has seen you somewhere in its training data or retrieval set. A citation is the model pointing at you as a source, usually with a link, usually because it's naming where a fact came from. A recommendation is the model deciding you're the answer, the thing to buy or use, picked over competitors who were also in the running. A lead is a person acting on that answer, a click that turns into a visit or a signed-up email.
Those four sit on wildly different cost curves. A mention costs you nothing to produce beyond existing in a category the model has indexed; it happens by accident of coverage, dozens or hundreds of times a month for anyone with a real web footprint. A citation sits one step up: the model has to treat you as where a fact came from, which takes being the kind of page it reaches for, though not the winner of anything. A recommendation requires the model to rank you above rivals on a question with real stakes, a much narrower outcome that doesn't happen just because you exist. A lead requires a person to act, which is rarer than either. Roughly: the cheaper the event is to earn, the more often it happens, and value runs in almost the reverse order.
Average those four into a percentage and the math does something predictable. Weight the four by count and the cheapest event owns the score outright. Weight the four rates equally and it still owns the trend, because it's the only term with the volume to move month to month. Either way the score mostly tracks mentions while wearing the costume of a single unified metric. A business staring at "visibility score: 61%, up 9 points this quarter" has no way to tell, from that number alone, whether the gain came from ten points of fresh mentions or two points of new recommendations. The dashboard answers "did something happen more," not "did the thing that pays rent happen more."
Run the hypothetical: a brand gets 400 mentions, 40 citations, 4 recommendations, and 1 lead in a month. Either of those weightings leaves you with a mention score wearing three pieces of decoration. Weight them by what each is actually worth and the same month might barely move the needle, because 4 recommendations and 1 lead are the only numbers that matter and they're small. Two completely different stories, same underlying activity, depending on which weighting the vendor picked when they built the formula, a choice that almost never gets disclosed in the sales deck.
This is the part that should bother anyone buying the dashboard, not the methodology, the incentive. A rising score is more sellable than four flat or declining lines, and averaging four unlike quantities is a reliable way to produce a rising score out of the cheapest one even when nothing that matters has changed. A company can watch its visibility climb for two quarters running on expanding mention volume while recommendations sit flat and leads don't move at all, and the single percentage will never surface that, because it was never built to keep the four apart. You'd have to ask for the breakdown by name, and the breakdown is a smaller, less flattering, less sellable product than the one number everyone actually bought.
What the compression hides is exactly what a buyer needs before paying for the thing: not whether visibility went up, but which of the four events moved. A score that can't answer that isn't measuring visibility. It's measuring whichever of the four was cheapest to generate, and calling it all of them.