AI visibility audit: what can you conclude from a low score?

An AI visibility report can show that competitors appear for questions where your brand does not. The obvious next question is what to fix, but the report has not necessarily told you what caused the gap.

For the audit, I used a fixed set of eight commercially relevant questions across three AI visibility tools. None of the 104 sampled answers mentioned Steve New. That was enough to establish a weak baseline for this sample, but it said nothing yet about why the absence existed.

A weak result could be compatible with several different explanations: a technical problem, an entity problem, weak or missing content, poor category association, an unsuitable prompt set, limited third-party evidence, or something specific to how a particular AI system retrieves and constructs answers.

Those explanations imply different actions. A useful audit therefore needs to do more than find problems that sound plausible. It needs evidence that makes some explanations less plausible than others.

What changed as I added more evidence

The absence was only the starting point. It did not tell me whether authority, content quality, access or something else was responsible.

The first useful narrowing came from conventional technical checks. The site had scored 91/100 in a conventional technical audit and was indexed, which makes a broad technical or indexation failure a less convincing explanation for the absence. It does not prove that every AI crawler, search layer or retrieval system can access and use the site in the same way, but it narrows the problem.

I also looked at pages appearing as sources around the questions I was testing. Many had estimated conventional page traffic of zero or one.

That weakens another easy explanation: that pages need substantial conventional search traffic before AI systems will use them as sources. High estimated page traffic was not a necessary condition in this sample.

It does not follow that links, site authority, traffic or brand prominence are irrelevant. The evidence only makes one strong version of the argument harder to defend.

A different problem appeared in the measurement setup itself. One platform initially resolved “Steve New” into a footwear context and produced a competitor set that included Steve Madden.

Because the error sat in the measurement setup rather than the site, entity resolution itself became something I had to verify before treating the competitor output as evidence.

I still could not name a single proven cause, but several explanations no longer deserved equal weight.

A diagnosis can be useful without proving one root cause

Suppose six explanations could reasonably account for a weak visibility result. If the available evidence makes four of them substantially less likely, the audit has made progress even if the remaining two cannot yet be separated with confidence.

The next question becomes more specific: what is the cheapest useful observation that would distinguish between the explanations that are still live?

That is a better basis for action than choosing the first plausible cause and building recommendations around it.

Consider two businesses that are both absent for commercially relevant buyer questions.

In the first, the relevant pages barely address those questions while competitors have clear material answering them. Improving or creating the page may be an obvious experiment.

In the second, assume the site already has strong material but the company has very little association with the category outside its own domain. Publishing another similar article may add more of what is already failing to change the result.

Both businesses can produce the same visibility symptom, yet the useful next action depends on which explanation is still credible.

This is why I am less interested in a complete list of everything that can cause weak AI visibility than in finding evidence that separates explanations which would lead me to do different things.

The diagnosis does not always need to be difficult. If the site blocks a relevant crawler, the audit tracks the wrong entity, the tested questions have little to do with the business or there is simply no material addressing the subject, the obvious explanation may be good enough. There is no prize for making the diagnosis more complicated than the evidence requires.

The deeper work matters when several materially different causes remain compatible with the same result.

The intervention should not be bigger than the diagnosis

The amount of uncertainty I am willing to accept depends partly on what somebody wants to change.

If the evidence suggests that one important page does a poor job of answering a relevant question, improving it may be cheap and reversible. I do not need a courtroom standard of proof before testing the change.

The standard should be higher if the recommendation is to publish fifty pages, rebuild the information architecture, change the positioning, launch a substantial digital PR programme or commit an ongoing budget to AI visibility. Before making that kind of intervention, I want considerably more confidence that the diagnosed constraint is real.

That still leaves several legitimate outcomes for an audit. The evidence may support fixing something directly, running a small test or measuring something else before changing the site. The gap may also be real without being important enough to work on yet.

Even a correctly diagnosed visibility problem is not automatically a priority. If the gap concerns questions with little relationship to a meaningful customer decision, fixing it may not justify the time or opportunity cost. Whether AI visibility itself is commercially valuable is a separate question and needs different evidence.

What I want from an AI visibility audit

I do not judge an audit by the number of recommendations it produces.

I want to know what we actually observed, which explanations the evidence has weakened, which important alternatives remain and what evidence would separate them. If the remaining uncertainty would change what we do next, it still matters.

A low visibility score can be a perfectly good reason to investigate. It is not, by itself, a diagnosis.

If several live explanations would lead to materially different fixes, the audit is not ready to prescribe one.