AI Visibility Is Not One Score: What You Actually Need to Measure

Being found is no longer enough. AI has to understand you. But when businesses try to measure that, most end up with a single number, and that’s usually where the measurement goes wrong.

A business can be recognized by an AI system, cited as a source, and still never get recommended. It can be described accurately in one context and confidently wrong in another. Measurement exists to work out which problem you actually have.

What Measurement Needs to Tell You

• AI visibility isn’t one result. Recognition, understanding, inclusion, citation and recommendation can move differently.
• A score can summarize without diagnosing. Two businesses with the same score can have completely different problems.
• Measurement only matters if it identifies the constraint. Otherwise it tells you something moved, not what to work on.

Why “Are We Visible to AI?” Is the Wrong Question

When a client asks me this, they usually mean: does an AI system know we exist? That’s answerable, and it’s usually the easiest of the five dimensions to satisfy. A business can be correctly named, described and located, and still miss most of the opportunities that run through AI.

The five dimensions influence each other, but they aren’t a funnel. A system can cite a business it barely understands, or recommend one without surfacing any citation. Strength in one is never proof of strength in another.

Collapse them into a single yes-or-no, or one number from one tool, and a business can spend months fixing the wrong thing.

What We Can Actually Observe

Nobody outside these systems can inspect what a model “knows.”

What we can observe is behavior: how consistently a system identifies a business, what it associates with it, whether it includes it in relevant answers, which sources it surfaces, and when it puts the business forward. Repeated over time, those observations give a practical picture of how a business is being represented.

The Five Diagnostic Questions

Each is a separate, observable question. None can be reliably inferred from the others.

Recognition

Does the system identify the correct entity? I usually start here, because poor recognition makes the other dimensions harder to diagnose. Much of the entity problem lives here and in understanding.

Understanding

Understanding is the set of attributes, expertise, geography, services and relationships a system associates with the business. A business can be correctly recognized and still wrongly understood. That can be worse than absence: absence leaves a buyer looking elsewhere, while inaccurate understanding can send them in the wrong direction.

Inclusion

Does the business enter the consideration set when someone asks about the problem it solves, rather than about the business by name? Missing from a branded question usually points to recognition. Missing from a buyer’s real question is a commercial problem.

Citation

Which sources appear when the system answers, and do they support, weaken or contradict what the business says about itself? That’s also what a website’s validation role either produces or fails to.

One page we worked on stayed uncited for weeks despite being live, relevant and well written. We reframed its opening paragraph to answer the target question directly, leaving the subject matter largely unchanged. Within ten weeks it had become the fourth most-cited page on the site. We can’t establish why from outside the systems. The presentation changed, the substance mostly didn’t, and citation behavior changed materially afterward.

Recommendation

Under what circumstances does the system actually put the business forward, rather than just mention it? Recommendation is competitive: the observable question is where the business gets put forward instead of the alternatives the system also recognizes.

Dimension	Diagnostic Question	Observable Signal
Recognition	Does AI identify the entity correctly?	Correct name, role, location, company/entity relationships
Understanding	What attributes does AI associate with the entity?	Expertise, services, markets, geography, credentials
Inclusion	Does the entity enter relevant consideration sets?	Appearance across non-branded buyer and problem questions
Citation	Which sources appear alongside answers?	Sources cited, source concentration, conflicting evidence
Recommendation	When does AI put the entity forward?	Recommendation frequency and context relative to alternatives

Why These Five Don’t Move Together

In one competitive luxury real estate dataset, a highly credentialed professional performed strongly on recognition and understanding while underperforming on citation. Two live biographies also described the same experience differently, a common result when the same fact has to be maintained by hand in more than one place. We can’t establish that the duplicate information caused the citation gap. It was an avoidable weakness in the evidence around an otherwise strong record.

That divergence is the point.

The Same Score Can Hide Completely Different Problems

Picture two businesses a simple tool might score identically:

• Business A: strong recognition, understanding and citation, but weak inclusion and recommendation.
• Business B: weak recognition, inconsistent understanding and scattered citation, but the occasional recommendation.

An aggregate score might put them in the same range. Their problems are completely different. Business A has a competitive problem. Business B has a more fundamental representation problem.

The score may be useful as a summary. It isn’t a diagnosis.

Not Every Business Needs the Same Level of Diagnosis

None of this means every business needs an elaborate measurement program. For a smaller company, a thin-margin business or a straightforward buying journey, a standardized tool may give enough to make sensible decisions: whether the business is appearing more or less often, and roughly how it compares with competitors. The economics have to work.

The problem changes as the business becomes more complex. More services, markets, customer types, high-value opportunities and decision contexts create more ways for the five dimensions to diverge, and more reason to know which one is holding things back, in which context, and whether the movement matters commercially.

The more complex the buying environment, the more valuable diagnosis becomes.

What Measuring This Actually Looks Like

The purpose of measurement isn’t to produce a grade. It’s to identify the constraint.

• Weak recognition: publishing more content may be premature.
• Wrong understanding: more visibility can amplify the wrong interpretation.
• Weak inclusion: branded success tells you very little.
• Weak citation: the evidence around the business needs attention.
• Weak recommendation, everything else strong: the problem is competitive, not identity.

None of these shows up in a single prompt, and none holds still. A citation that appears once can disappear just as easily, which is why I don’t put much stock in a single test.

One engagement tracked a luxury brokerage’s share of AI citations against its two closest competitors weekly for about fourteen weeks, across roughly fifty representative buyer and seller questions. It moved from about 30% of the three-way share to a peak near 40%, then held around 38-39% as overall citation activity cooled with the season. That isn’t a position the brokerage now owns, and a snapshot taken in the wrong week would have told a different story.

The question set matters as much as the tracking. Measurement tends to drift toward what’s easiest to count, and that isn’t always what matters commercially. You can measure the wrong questions extremely precisely. A useful question set reflects real customer problems and real decision contexts, not just the prompts that are convenient to measure. Some questions name the business; most don’t. All of them should matter commercially.

Measurement Is Still Not the Business Outcome

The five dimensions describe how a business is represented inside AI systems. They don’t say what that’s worth.

A rise in citation share may matter enormously in one buying environment and very little in another, and a recommendation isn’t a sale. The commercial question is whether improved visibility is changing consideration, opportunities, pipeline or qualified demand, and that needs its own measurement.

The diagnostic dimensions tell you what’s happening. Business measurement tells you whether it matters. Confuse the two and an AI visibility metric quickly becomes another vanity metric.

Do You Know Which Problem You Have?

A business that can’t say whether its constraint is recognition, understanding, inclusion, citation or recommendation is optimizing blind. Three questions show where you stand:

• Do you know which of those dimensions is currently the constraint, or do you only know whether your overall visibility is up or down?
• Are the questions you’re tracking representative of the decisions real customers bring to AI, or mostly prompts that are convenient to measure?
• Do you have enough history to tell real movement from a good or bad snapshot?

If you can’t answer those, you probably don’t have an AI visibility diagnosis yet. You have a collection of observations.