Guide

What an AI visibility score means — and how to tell whether yours is real

An AI visibility score is the share of AI answers, to a set of buyer questions, that cite or name your business. Whether it means anything depends on three things the score itself does not show: which questions were asked, how many answers it is built from, and how many times each question was asked. A score without those three is a number, not a measurement.

Cited, named, or ranked?

Cited — your site is one of the sources the assistant links to. This is what sends visits.
Named — your business is mentioned in the answer text, linked or not. This is what shapes a shortlist.
Share of voice — your citations as a share of all citations. It moves when competitors move, even if nothing about you changed.

They are not interchangeable: in our study of 8,441 answers, a brand was cited in 1.4% of answers but named without a link in 5.2% (who AI assistants cite). A tool that reports only one of them is telling half the story.

How much a score can move on its own

Suppose your score is 2%. This is the range that score is honestly compatible with, and the smallest rise the next measurement could confirm, at different numbers of answers:

Answers behind the score	Cited in	95% interval	Smallest rise a test can confirm
50	1 of 50 (2%)	0.1% – 10.6%	+14.0 points
108	2 of 108 (1.9%)	0.2% – 6.5%	+7.4 points
500	10 of 500 (2%)	1% – 3.6%	+2.4 points
2,000	40 of 2,000 (2%)	1.4% – 2.7%	+1.0 points

Computed on this page: exact (Clopper–Pearson) interval; smallest rise = the first later count a two-sided Fisher exact test at 5% calls different, with the same number of answers. Our own reports use 108 answers per market — about 12 questions × 5 assistants × 2 asks (1 on ChatGPT).

Two things follow. A score built from a few dozen answers cannot tell 0% from 10%. And a day-over-day change of a point or two is almost never a change: it is the assistant re-drawing its sources. Ask for the interval, and for the test behind any “up” or “down”.

Which questions — the part most scores hide

The same business can score 0% or 30% depending on the question list. A generic category list (“best running shoes”) is answered from editorial sites and big marketplaces, so most brands score zero on it — 63% of brands in our study did. Questions written from what you actually sell, or taken from your own Search Console, are the ones that matter for your revenue. A good tool shows you the exact questions before it measures and lets you change them.

A checklist for any AI visibility checker

Can you see the exact questions, and replace them with your own?
How many answers is the score built from, and is each question asked more than once?
Is there an interval on the score, or just a number?
When it says you went up or down, is that change tested?
Does it separate cited from named, and show which pages were cited instead of you?
Can you download every answer behind the number?

AnswerRange was built to pass that list, and the method is published with its data (methodology, papers). Measured through the providers' APIs, which are not identical to the consumer apps.

Questions

What is a good AI visibility score?

There is no universal benchmark, because the score depends on the question list. Compare yourself with the competitors measured on the same questions, and with your own previous runs — and only where the intervals do not overlap or a test says the change is real.

Is a free AI visibility check accurate?

It can be, if it shows its questions, asks each one more than once and reports an interval. A free check from a single answer per question is a snapshot of one draw.

Why is my AI visibility score zero?

Usually because the assistants answer your buyers' questions from other sites: reviews, publishers, forums and marketplaces. Check that the questions match what you sell, then look at which pages are cited instead — that list is your to-do list.

How often should I measure AI visibility?

Every two to four weeks for most businesses. Measuring daily mostly records the assistants re-drawing their sources; measuring too rarely means you cannot tell which change worked.

See where you stand. The first measurement is free, needs no card, and shows you every question before anything runs. Measure my domain — free

Other guides: Getting recommended by ChatGPT, from measured data · how we measure · who AI assistants cite