How AI Visibility Is Measured

AI visibility is measured by asking AI platforms the questions customers actually ask, recording who is named and which sources are cited, and repeating the same questions over time. A score summarizes the result. It is not the result.

See What a GEO Audit Measures Read the AI Visibility Index

By JoLyn Laney, FounderLast updated: September 27, 2026

What is AI visibility?

Short answerAI visibility is whether, and how, a business appears in the answers AI platforms give, such as ChatGPT, Gemini, Perplexity, Claude, and Google's AI answers. It covers being named, being recommended, being described accurately, and being cited as a source.

How is AI visibility measured?

Short answerAI visibility is measured by asking AI platforms the questions customers actually ask, recording who is named and which sources are cited, and repeating the same questions over time. Avante records six levels for each question: presence, recommendation, accuracy, citation and source context, competitive share, and commercial outcome.

Avante calls this method the Avante AI Visibility Measurement Framework. It is the structure behind every assessment we deliver and behind our own published research, Avante's AI Visibility Index.

The framework is built on one idea. No AI platform publishes a ranking or a score for a business. What can be observed is the answer itself: who was named, what was said, and which sources were shown. Everything else is a summary of those observations, and a summary is only as good as the record underneath it.

The Avante AI Visibility Measurement Framework

Six levels, measured in order. Each level answers one question, and each depends on the one before it.

1
Presence
Does the business appear at all?
What is recorded: Whether the business name shows up anywhere in the answer, for each question, on each platform, in each run.
Why it matters: Presence is the floor. A business that never appears cannot be recommended, described, or compared.
2
Recommendation
Is the business actually suggested as an option?
What is recorded: Whether the business is offered as a choice, and in what position. Being named in passing, or as an example of what to avoid, is recorded separately from being recommended.
Why it matters: Customers act on the short list. Appearing in an answer and being suggested by it are different results.
3
Accuracy
Are its services, category, location, service area, positioning, and other important facts correct?
What is recorded: Each factual statement the AI makes about the business, checked against a list of facts the business has confirmed. Errors, omissions, and outdated claims are logged individually.
Why it matters: A recommendation built on wrong facts can send the wrong customer, or send the right customer somewhere else.
4
Citation and source context
Which sources are being cited or appear to support the response?
What is recorded: Every source the platform shows with the answer, grouped by type: the business website, competitor websites, directories, review platforms, publishers, community discussions.
Why it matters: Sources show where an answer came from. They are the most direct evidence of what to work on, and of what is outside the business's control.
5
Competitive share
Which competitors are being recommended instead?
What is recorded: Every business named across the question panel, how often, and in which positions, so the business's share of answers can be compared with named competitors.
Why it matters: Visibility is relative. The useful question is rarely whether a business appears. It is who appears in its place.
6
Commercial outcome
Does increased AI visibility contribute to qualified website visits, calls, leads, reservations, bookings, or revenue?
What is recorded: Visits referred by AI platforms in analytics, and the calls, form submissions, and bookings that follow, where tracking is in place to see them.
Why it matters: This is the level the business cares about. It is also the hardest to observe, because many AI answers are read without a click.

What is an AI visibility score, and is it the same as results?

Short answerAn AI visibility score is a single number that summarizes many individual observations. It is not the same as results. It is useful for tracking direction over time, but it is not a measurement any AI platform publishes and it is not the business outcome. The outcome is qualified visits, calls, leads, bookings, and revenue.

Scores are useful. One number is easy to track, easy to report, and easy to compare against a baseline. Avante uses a scorecard in its own audits for that reason.

A score also hides things. Before relying on one, ask three questions:

What was asked? A score from branded questions, where the customer already knows the business name, will look far better than a score from the unbranded questions a new customer asks.
What was counted? A score that counts mentions will be higher than one that counts recommendations.
Can you see the answers behind it? If the questions, dates, platforms, and responses are not available, the score cannot be checked or repeated.

A rising score with no change in qualified visits, calls, or bookings is a reporting result, not a business result. That is why commercial outcome is a level of the framework and the score is not.

What is a good AI visibility score?

Short answerThere is no universal benchmark. Every tool calculates its score differently and no AI platform publishes one, so a number from one tool cannot be compared with a number from another. A score is meaningful against your own baseline and against named competitors measured the same way, on the same questions.

For scale, Avante's own research found that 12.0% of the businesses an AI listed in a market appeared in its first five recommendations. Being absent from most answers is the normal starting point for a local business, not a sign that something is broken.

What is the difference between an AI mention, a recommendation, and a citation?

Short answerA mention is any appearance of a business name in an AI answer. A recommendation is a mention in which the AI offers the business as an option to choose. A citation is a source the AI platform links or lists alongside its answer. A business can be cited without being named, and named without being cited.

Mention: A mention is any appearance of a business name in an AI answer. It says the business was present in the response, and nothing about whether it was endorsed.
Recommendation: A recommendation is a mention in which the AI offers the business as an option to choose. Every recommendation is a mention. Most mentions are not recommendations.
Citation: A citation is a source the AI platform links or lists alongside its answer. A business can be cited without being named, and named without being cited.
Visibility score: A visibility score is a single number that summarizes many individual observations. It is useful for tracking direction over time. It is not a measurement any AI platform publishes, and it is not the business outcome.

These four are often reported as if they were the same thing. A report that says a business “appeared in 40% of answers” could be describing mentions, recommendations, or citations. Ask which.

A sample measurement table

This is the shape of the record behind an assessment: one row per question, per platform, per test date. The firm is Castillo & Whitfield Injury Lawyers, the fictional firm used in Avante's sample report. Every value below is illustrative and competitor names are placeholders.

Illustrative AI visibility measurement table for a fictional law firm
Query tested	Platform	Date tested	Brand named?	Brand recommended?	Competitors named	Sources and citations observed	Accuracy notes	Recommended action	Change from baseline
best personal injury lawyer in Las Vegas	ChatGPT	Sept 8, 2026	Yes, 2 of 3 runs	No	Firm A, Firm B, Firm C	A legal directory, a best-of list, Firm A's website	Practice areas correct	Complete and correct the firm's legal directory profiles	Baseline
who handles truck accident cases near Henderson	Perplexity	Sept 8, 2026	No, 0 of 3 runs	No	Firm B, Firm D	Firm B's truck accident page, a community thread, a legal directory	Not applicable, the firm was not named	Publish a truck accident service page with location and case types	Baseline
Castillo & Whitfield reviews	Gemini	Sept 8, 2026	Yes, 3 of 3 runs	Not a recommendation question	None	The firm's website, a review platform	Office address out of date in 2 of 3 runs	Correct the address on the three listings that still show the old office	Baseline
best personal injury lawyer in Las Vegas	ChatGPT	Dec 8, 2026	Yes, 3 of 3 runs	Yes, 1 of 3 runs	Firm A, Firm B	A legal directory, the firm's website, a best-of list	Practice areas correct	Hold. One run in three is inside normal variation. Re-test next cycle.	Named 2 of 3 to 3 of 3. Recommended 0 of 3 to 1 of 3.

Illustrative example, not client data. The last row repeats the first question three months later, which is how a change from baseline is recorded.

Why does AI visibility need repeated testing?

Short answerAI answers vary from one run to the next, so a single test can show a business as present or absent by chance. Asking the same fixed set of questions several times, and again at a later date, is what separates a real change from normal variation.

A search ranking can be checked today and again tomorrow with about the same result. An AI answer cannot. Ask the same question twice and the platform may name different businesses and show different sources. One answer is a single moment, not a position.

Why the query panel has to stay consistent

A query panel is the fixed list of questions used to test a business. It is written before any testing starts, from the questions real customers ask, and it is weighted toward questions that do not include the business name.

The panel only works as a measuring instrument if it does not change. If a question is added, dropped, or reworded after the first results are in, the next set of results describes a different test, and any difference could come from the questions instead of the business. When a panel does need to change, the change starts a new baseline.

Why each question is asked more than once

Running each question several times turns a yes or no into a rate: named in two runs of three, recommended in none. A rate can be compared with the next measurement. A single yes or no cannot.

How a real change is told apart from noise

Small movements happen with no change to the business at all. Avante reports rates with a range, records the platform, model, and date of every test, and reports movement only when it is larger than the normal variation between runs. A change that does not clear that bar is reported as no change.

How commercial outcome is measured, and where it stops

The first five levels are measured by testing the AI platforms. The sixth is measured in the business's own data: visits referred by AI platforms, and the calls, forms, reservations, and bookings that follow.

This level has real limits. Many people read an AI answer and then search for the business by name, call it, or walk in, and none of that is labeled as coming from AI. Referred visits are a floor, not a full count. Avante reports what can be observed and says so where the data stops.

Outcome measurement also depends on tracking that works. If calls, forms, and bookings are not being recorded reliably, that is the first thing to fix, before any channel is judged. See GEO vs. SEO vs. Google Ads: where to invest first.

How can I check my AI visibility?

Short answerAsk AI platforms the questions a new customer would ask, without your business name, and record who is named and which sources are shown. Ask each question more than once, on more than one platform. Then ask about your business by name and check every fact.

That manual check costs nothing and takes an afternoon. Avante also offers a free AI visibility checkthat runs real buying questions on two platforms and shows who was named and which sources were cited. For the owner's checklist, see the local AI visibility checklist.

How do AI visibility tools work?

Short answerMost AI visibility tools send a set of prompts to one or more AI platforms on a schedule, record which brands and sources appear in the answers, and summarize the result as a score or a share. They differ in which prompts they use, which platforms they cover, whether they test through a developer API or the consumer app, and how many times each prompt is run.

Those four differences explain why two tools can report very different numbers for the same business on the same day. Before trusting a tool, find out what it asked, where, how, and how often. The same questions apply to a person or an agency, and they are listed in how to choose a GEO agency.

What this framework cannot tell you

It cannot show why an AI platform chose one business over another. The platforms do not disclose that, and sources shown with an answer are evidence, not proof of cause.
It cannot cover every question a customer might ask. A panel is a sample of real questions, chosen and documented in advance.
It cannot predict what a platform will say next month. Platforms change their models and their sources, sometimes within days.
It cannot guarantee that any change will produce a recommendation.

Where the framework is used

The same framework sits behind two things Avante publishes. The Avante AI Visibility Index applies the presence, recommendation, and source levels across 60 local markets. A GEO auditapplies all six to one business, on the questions its own customers ask. For the owner's view, see AI visibility for local businesses.

Published September 27, 2026 by Avante Visibility. Written by JoLyn Laney, President and Founder.

See What a GEO Audit Measures

All six levels, measured on the questions your customers ask, across the major AI platforms, with a prioritized plan and a re-measure against your baseline.