What Is a Good AI Visibility Score? Benchmarks by Rank

Three Numbers Called “the Score”

Before you compare your score with anyone's, check which number it is. Three different measures go by the name, and they sit on very different scales:

This post benchmarks visibility on one engine at a time, the simplest of the three to compute and to compare. Share of voice reads lower because every answer names several brands, and it depends on which brands count as competitors. Ranqo's Share of Voice counts more brands: one named only once still counts if you watch it, if AI gave its website or if Ranqo confirmed its website. More brands in the count can only lower your share, so on the same answers Ranqo's figure reads the same as the figure here or lower. Our guide to measuring AI share of voice covers the choices that change it.

In advertising, share of voice is read against market share: in research published by the IPA, Les Binet and Peter Field found that brands whose share of voice exceeds their share of market have historically tended to grow. You can make the same comparison with AI answers, though we know of no published study that tests whether it predicts growth there.

A composite score is only comparable within one tool, which is one reason AI visibility tools show different numbers for the same brand. A composite that includes sentiment also moves more between runs: in Ranqo's research paper, sentiment flipped 6.7 times as often as whether a brand was named at all, which is why Ranqo reports visibility, position and sentiment as separate numbers.

How We Measured

The benchmarks come from the answers in Ranqo's tracking, read for every brand each answer named, not only the company that tracks the questions:

AI Visibility Benchmarks by Rank

The leader's score is the ceiling your category sets. Across categories and engines, the leader typically appeared in 47.4% of the answers, and the middle half of leaders scored between 30% and 70%. Second place typically scored 40%, third 30%, fifth 20% and tenth 10%.

The field is likely wider than your watchlist. In a typical score set, an answer named 5.1 brands on average, and one company's questions on one engine turned up a median of 45 different brands. So the brand ranked third in your answers may be one you have never thought of as a competitor, and a score read only against the rivals you chose can look better than it is.

The Same Score, a Different Rank

Your own score does not fix your rank. Among the companies asking these questions, a score of 20-30% on one engine ranked anywhere from #1 to #8, with a median of #3, and about half reached the top three. A score of 10-20% ranked from #2 to #13.

Rank depends on how widely the answers spread their mentions. Where they keep naming the same few brands, the leader scores high and a modest score sits mid-table: in the quarter of score sets naming the fewest different brands, the leader typically scored 70%. Where they name many different brands, no one is named often and a modest score can lead: in the quarter naming the most, the leader typically scored 35%.

Higher scores usually put a company in the top three: two in three of those scoring 30-40% got there, and nearly all of those scoring 40-60%. They did not settle first place, though: fewer than half of the companies scoring 30-60% ranked first. At 60% or more, every company ranked first or second. Under 10%, the best rank any company reached was #9. Across every company and engine, the company asking the questions was named at all in 57.9% of score sets, ranked in the top three in 39.5% and first in 24.3%.

Each Engine Has Its Own Scale

Read a score against the benchmark for its own engine. ChatGPT's and Perplexity's medians sat within 5 points of each other at every rank. Gemini's were the highest at every rank, though never more than 10 points above the next engine's, the difference of 2 answers in a typical set: its third-placed brands typically scored 40%, against 30% on Perplexity. Of these three engines, Perplexity's answers named the fewest brands, 4.5 on average in a typical set, against 5.5 on ChatGPT and 5.8 on Gemini.

Your overall score, across every engine, moves less between runs than a single engine's, which makes it the better number to follow over time. Compare yourself with other brands one engine at a time, though: a company strong on one engine and absent from another can average out to a number that describes neither.

Is Your Lead Bigger Than the Noise?

A rank is only as real as the gap behind it. Most answers hold from one run to the next: in our earlier measurement, whether one engine named the company for one question changed between runs only 6.4% of the time, and in Ranqo's research paper, 77.5% of a brand's question and engine pairs named it every time or never. But a score averages only about 20 answers, so the few that change still move it: a single engine's score moved 20 points or more between two runs one time in 20, and the overall score 14.5 points or more. That was measured on one brand's score rather than on the gap between two, but it gives the scale. Most gaps at the top are smaller. The leader's typical margin over second place was 10 points, and in 79.6% of cases the leader was tied with second place or ahead by under 20 points.

Research that compares whole lists of recommended products finds more movement, since any product joining or leaving a list counts. In a September 2026 preprint, researchers at Maastricht University, Utrecht University and the University of Zurich found that repeated runs of the same query on ChatGPT had a mean overlap of only 0.178: on average, the products both runs named made up 17.8% of all the products either run named. By the authors' account that is a lower bound, since one product named two ways counts as two. In another preprint, researchers at Northwestern and Boston University argue that recommendations should be treated as a probability distribution rather than a single list.

So treat a gap under 20 points on one engine as a tie until it holds over two runs in a row, and three before you call it settled. In our earlier measurement, a mention that had just appeared, absent the run before, was still there in the next run 54.3% of the time, and one seen in two runs in a row 96.1% of the time.

Your Questions Set the Scale

Every benchmark here comes from unbranded questions, the ones that don't name the company asking. A question that names you is answered with your name in it almost every time: in Ranqo's research paper, each engine named the brand in 94% to 100% of its answers to such questions on a brand's first tracking run. So a set that mixes branded questions in reads higher without telling you anything about how buyers find you.

The kind of unbranded question matters too. In the same paper, broad discovery questions named the brand asking them in 22.9% of answers, averaged across brands, against 10.8% for problem-and-solution questions and 11.4% for use-case questions, so a set weighted toward broad questions reads higher. And “best X” questions make answers name more vendors, which tends to raise how often brands appear without raising their share of the names. So compare like with like: the same questions, on the same engine, over the same window. Our guide to choosing prompts to track covers building a set that can show a real change.

Your Size Sets the Bar

What counts as good also depends on how well known you are. Ranqo's research paper measured each brand's own score on its first tracking run, across the engines it tracked, and found a ladder: global household names such as Stripe and Nike were named in 73% of unbranded answers, mid-market brands such as Olipop and Klaviyo in 44%, and niche brands in 11%, each step about 30 points lower.

Those are a brand's own score across engines, not the score of whoever holds a place on one engine, so they are not benchmarks for the charts above. They do show how to read yours: 11% is the average start for a niche brand, while a household name scoring like a niche one is behind its peers. For a brand AI already names most of the time, the paper suggests watching share of voice against rivals rather than visibility. Our post on why big brands win AI search covers how smaller brands catch up.

How to Read Your Own Score

1. Name the number. Say which metric it is and what it divides by: visibility over every answer, share of voice over the mentions of you and the brands a tool counts as competitors, or a tool's composite.
2. Read it per engine. Compare a ChatGPT score with ChatGPT benchmarks, not with a figure taken across engines.
3. Find your rank among every brand named. Count the brands the answers name, not only the ones on your watchlist.
4. Check the gaps either side of you. A gap under 20 points on one engine, or 15 points overall, is a tie for now. In a set of 20 answers each answer is worth 5 points, so 20 points is 4 answers.
5. Judge change over runs, not one check. Read a move only once it holds over two runs in a row on the same questions, and call it settled after three.

In our data, a score of 30% or more usually put a company in the top three on that engine, 60% or more always put it first or second, and no company scoring under 10% reached the top five. Between 10% and 30%, the score alone can't tell you, and your rank and the gaps around it can.

If your score is low

Check first whether you are named at all. In 42.1% of score sets no answer named the company asking the questions, so a zero is common, and getting named comes before climbing the ranks. Then allow for your size: a niche brand starts far below a household name, so compare yourself with brands like you as well as with the leader. Then look at who the answers name instead and which pages they cite. Our guides to fixing an AI visibility gap and why a competitor shows up in AI answers cover what to change.

How Ranqo Fits

Every week, Ranqo asks your buyer questions on each engine your plan covers and detects the brands the answers name. The Visibility page shows your score overall and on each engine. The Competitors page shows Your Rank among every qualifying brand, your Share of Voice and your #1 Rank Rate, and ranks those brands by visibility with each engine's score beside it, so you can see the gaps either side of you.

To read your rank the way this post counts it, filter the Competitors page to one engine, to unbranded prompts and to the last 7 days, which usually holds only your latest weekly run. By default it reads every engine you track, branded prompts included, over 30 days. Two differences remain: where you tie with a brand, Your Rank puts the one with the better average position first, while this post gives you the higher place; and its #1 Rank Rate counts how often an answer names you first, not how often you lead the ranking.

When you act on a recommendation in the Action Center, Ranqo reports the change in its metric in points only when that change is bigger than the noise band for the metric. Otherwise it reports the answers you won, or no detectable change.