Vector Index AI Search Intelligence

Four results, not one

How to get recommended by AI

A business can be the source an answer is built from and still not be the business the answer puts forward. Those are different outcomes with different causes, and a report that blends them cannot tell you which one you have.

The four states

Every answer places a business in one of four positions. We record them separately because the work that fixes one does not fix another.

Absent

The business does not appear. Nothing about it reached the answer.

Mentioned

The name appears in the text. This is the result most often sold as visibility, and on its own it is the weakest of the three positive states.

Cited

A page from the business is listed as a source the answer drew on. The business supplied the material. Being cited does not mean being put forward.

Recommended

The answer advises the reader toward the business. This is the outcome a buyer actually wants, and it is the only one of the four that is a judgement rather than an observation.

Why the last one is handled differently

Absent, mentioned and cited can be read directly out of a stored answer. Recommended cannot. Deciding whether an answer recommended a business is an interpretation of the wording, so we do not write it into the record of what was observed.

Instead the rule for classifying it is written down and frozen before the answers are read, and the verdict is stored separately from the observation, labelled as inference and tied to the run it came from. If the rule were written after reading the answers, we would be choosing the rule that produced the result we wanted.

This is the sort of thing that sounds like process until you see what it prevents. A provider who classifies recommendations after seeing the data can make any campaign look successful.

What this changes about the work

If a business is absent, the question is whether anything about it is reachable and specific enough to use. If it is mentioned but never cited, the material being used comes from somewhere else, and the work is about what those other sources say. If it is cited but not recommended, the business is supplying the evidence for an answer that then points elsewhere, which is a positioning problem rather than a technical one.

Three different diagnoses, three different jobs. This is why a single score is not useful, and why we do not publish one.

What we will not claim

We cannot promise a recommendation, and we would not trust anyone who did. Answers vary between runs, between engines and between model versions. Sampling records what answers contained under recorded conditions. It does not establish why a business was included or left out, and a change followed by a different result is not proof that the change caused it.

What measurement does give you is a defensible statement of where you stand today, which of the four states you are in, and which of them a piece of work is aimed at.

How we measure · Ranking in ChatGPT · What an audit should contain · Our scope and prices

Questions we get asked about this.

What is the difference between being mentioned and being recommended by AI?

Mentioned means the business name appears in the answer. Recommended means the answer advises the reader toward that business. A business can be named among several options without being the one put forward, so the two are recorded separately.

What does it mean to be cited by an AI answer?

It means a page from the business was listed as a source the answer drew on. The business supplied material the answer used. That is a different result from the answer recommending the business, and one can happen without the other.

Can a business be cited but not recommended?

Yes, and it is a common and useful finding. It means the business is supplying the evidence for an answer that then points a buyer somewhere else. The fix for that is usually about positioning rather than about the website.

Can anyone guarantee an AI recommendation?

No. Answers vary between runs, engines and model versions, and no provider controls a model's output. We do not guarantee citations, rankings, traffic or revenue, and a guarantee from anyone else is a reason to ask what evidence supports it.

Why not just report one visibility score?

Because the four states have different causes and different fixes. A single number blends them and hides which problem a business actually has, which makes it useless for deciding what to do next.

We publish what we find.

We're measuring AI answers in Cincinnati and will publish what we find, including our own results, including when they're unflattering. No pitch attached.

One email a month at most, and you can leave with one click. We do not sell your address and we will not add you to anything else.