An AI visibility score is a number, usually from 0 to 100, that sums up how often a brand appears in answers from ChatGPT, Gemini, Perplexity and Google's AI features across a fixed set of questions. There is no standard formula: on 27 October 2026 we calculated six different scores from the same 10 ChatGPT answers, and the same brand came out anywhere between 8.5 and 67.
A marketing manager opens two reports in the same week.
In the first, the brand scores 42. In the second, 61.
Which one is right?
Both. And neither, until you know how each was calculated.
The mistakes we see most often with these scores:
- comparing the number from two different tools, as if they were two thermometers;
- reading 45 as a fail, because out of 100 it looks like one;
- celebrating a jump from 32 to 61 that comes from a change of denominator, not from a single extra answer.
Take Sunscape, which designs solar plants across Italy. In the six months from March 2026 its organic impressions went from 329 to 5,809 a month, as we showed in how long SEO takes. That number has one definition: Google's, and you read it in Search Console.
An AI visibility score doesn't. Every tool has its own, and the formula moves the result more than real visibility does.
In this guide:
- the two families of score that share the same name;
- the formulas published by Semrush, Peec AI, Profound, Ahrefs and ScryRank;
- our test: 10 Italian questions on ChatGPT, six formulas, six results;
- what a "good" score really means, with four vendors' scales side by side.
Let's start.
What is an AI visibility score? (two families, one name)
An AI visibility score is a single number that summarises a brand's presence in AI-generated answers. Two different measures travel under that name, though, and the first thing to check is which one you're looking at.
A presence score counts what the AI actually answers. You ask ChatGPT, Gemini or AI Overviews the same questions, record how many answers name the brand and compare it with competitors. This is the score in Semrush, Peec AI, Profound, Ahrefs Brand Radar and ScryRank.
A readiness score asks the AI nothing. It checks the website: structured data, robots.txt, llms.txt, titles. Krawfly, for example, gives 25 points to structured data, 20 to content, 20 to citation signals, 20 to crawler access and 15 to locale and freshness. The Italian consultant Lorenzo Magliani says it plainly on his own checker: the score assesses the site's signals, not actual citations.
So which one do you need?
The first tells you whether you appear. The second tells you whether your site is readable. A site scoring 90 on readiness can still never be named, because AI tools recommend brands they find cited elsewhere (our test below shows it).
From here on we mean the presence score: it's the one that ends up in monthly reports and drives budget decisions. The wider picture, engine by engine, is in our AI visibility guide. If you want its close relative from classic SEO, we took apart the SEO score in the same way.
How is it calculated? (the formulas vendors publish)
A presence score is calculated by dividing the answers, or the mentions, that contain your brand by a total. Everything depends on that total. Here is what vendors say in their documentation, checked on 27 October 2026.
| Tool | Metric | What it counts | Divided by |
|---|---|---|---|
| Peec AI | Visibility | answers naming the brand | all answers |
| Profound | Visibility Score | answers naming the brand | answers naming at least one brand |
| Profound | Share of Voice | brand mentions | all brand mentions |
| Semrush | AI Visibility Score | brand presence, 0 to 100 | other brands in the industry, formula not published |
| Ahrefs Brand Radar | AI Share of Voice | brand mentions and citations | the brands compared; estimated impressions weighted by Google volume |
| ScryRank (Visilay) | Weighted visibility score | appearances multiplied by search volume | 0-100 scale, with an unweighted version alongside |
Three things stand out.
Peec and Profound use "visibility" for two different ratios. If your brand appears in 50 answers out of 100 and only 60 answers name any brand at all, Peec gives you 50% and Profound 83%. The example comes from AuthorityTech's audit of 10 September 2026, which reviewed the documentation of nine platforms.
Counting answers or counting mentions changes the order of magnitude. In the same audit, Similarweb reports Brand Visibility of 24.23% (1,346 answers out of 5,556) and a mention share of 2.86% (1,346 out of 47,020). Same data, a number almost ten times smaller.
Weighting by volume pulls the score towards the most-searched questions. Ahrefs writes that weighting mentions by Google volume is "a modeling choice, not a measured relationship". For the same reason, in ScryRank, the monitoring tool we built at Visilay, we show the unweighted score next to the weighted one.
The metrics underneath every score (mentions, citations, position, sentiment) are covered one by one in our guide to AI visibility metrics; share of voice has a page of its own.
Same brand, six different scores (our test on 10 Italian questions)
To see how much the formula matters, we took real answers and ran them through six different calculators.
The method. On 27 October 2026 we asked ChatGPT 10 questions a small Italian business would ask about accounting and e-invoicing software, through the DataForSEO API with the location set to Italy. Two examples: "what is the best e-invoicing software for a small business?" and "best ERP systems for Italian manufacturers" (both asked in Italian). We then counted six brands: Fatture in Cloud, Aruba, Fattura24, Danea Easyfatt, TeamSystem and Zucchetti. None of them is a client: we needed a market where answers name names.
The 10 answers listed 47 products, 28 of them from the six brands we tracked.
The formulas:
- A. Mention rate: answers naming the brand out of all answers, as in Peec AI;
- B. Share of voice in the set: brand mentions out of mentions of the six tracked brands;
- C. Total share of voice: brand mentions out of all 47 products named;
- D. Position score: 5 points for first in the list, 3 for second or third, 1 from fourth down, divided by the maximum possible (the 5-3-1 scheme is the one proposed by Campaign Creators);
- E. Relative to the leader: mention rate rescaled so that the top brand scores 100.
| Brand | A. Mention rate | B. SoV in set | C. Total SoV | D. Position | E. Relative to leader |
|---|---|---|---|---|---|
| Fatture in Cloud | 60 | 21.4 | 12.8 | 36 | 100 |
| Aruba | 50 | 17.9 | 10.6 | 34 | 83 |
| Danea Easyfatt | 50 | 17.9 | 10.6 | 26 | 83 |
| TeamSystem | 50 | 17.9 | 10.6 | 26 | 83 |
| Zucchetti | 40 | 14.3 | 8.5 | 20 | 67 |
| Fattura24 | 30 | 10.7 | 6.4 | 14 | 50 |
Zucchetti scores 8.5 or 67. Only the column changes. The answers are the same.
And TeamSystem?
This is the sixth formula, the one nobody writes in the report. Fatture in Cloud has been part of the TeamSystem group since 2015 and Danea since 2012, as TeamSystem itself states. If the tool counts the group instead of the brand name, TeamSystem appears in 10 answers out of 10: a score of 100. If it only counts the word "TeamSystem" in the answer, 50.
Tools call this setting aliases, or brand variants. It's usually chosen by whoever sets up the project on day one, and then forgotten.
One thing does hold, though.
The order between competitors barely moves. Fatture in Cloud is first in all five columns, Fattura24 last in all five. The only shift is in column D: Aruba, which ChatGPT often puts at the top as the "cheapest" option, pulls ahead of Danea and TeamSystem.
A limit, before you point it out: one run per question is a snapshot, not a measurement. The SparkToro and Gumshoe study of January 2026 ran 12 prompts 60 to 100 times each on ChatGPT, Claude and Google AI: the same list of brands came back less than 1% of the time, the same order less than once in 1,000. How much a score moves from one day to the next is what we measure in our study on AI visibility data accuracy.
A wrong score leads to a wrong budget.
You've just seen the same brand go from 8.5 to 67 by changing nothing but the formula. If the report you read every month uses a denominator you don't know, you may be rewarding or cutting work because of a calculation effect. We start from the questions your customers ask, fix the formula and the competitors on day one, and show you the raw data next to the score.
Book a strategy callWhat is a good AI visibility score? (the same number gets four different names)
A good AI visibility score is one higher than your direct competitors' score, measured with the same tool, on the same questions and the same engines. There is no absolute value that works for everyone: Semrush, on its own tool page, gives no thresholds and suggests using the score "as a baseline to track progress month over month".
HubSpot, which sells its own grader, says the same on its blog: "there's no single universal benchmark".
The thresholds you find online are each vendor's own convention. Line them up and the problem is obvious.
| Vendor | What it calls a score of 45 | Its bands |
|---|---|---|
| AmICited | average visibility | under 40 poor, 40-69 average, 70-89 good, 90+ excellent |
| OptimizeGEO | competitive | under 20 structural gaps, 30-50 competitive, over 50 strong; leaders rarely exceed 70 |
| Campaign Creators | category presence | 0-8 pre-visibility, 8-25 traction, 25-50 presence, 50-75 authority, 75-100 dominance |
| Krawfly (readiness) | partially visible | 0-30 not ready, 31-60 partial, 61-80 nearly, 81-100 ready |
For AmICited 45 is a borderline pass, for OptimizeGEO it's already a good result. For Krawfly it isn't even the same measure.
The method we use has three steps.
1. Work out your fair share. With six tracked brands, the neutral share of voice is 100 divided by 6, or 16.7%. It's the same idea as the "equal split" baseline used by DiagnoSEO. In our test Fatture in Cloud sits at 21.4, above its share; Zucchetti at 14.3 and Fattura24 at 10.7, below.
2. Look at the gap, not the number. "Eight points behind the top competitor" still means something if you change tool. "45" doesn't.
3. Read the trend over at least three months. With answers that change on every request, a shift of a few points in one month is noise until it repeats. How many questions to ask and how often is covered in our guide to prompt tracking.
How to read your tool's score (five questions for the vendor)
Before an AI visibility score goes into a report, ask the vendor these five questions. If they can't answer the first two, the number shouldn't be compared with anything.
- What is the denominator? All answers, answers naming at least one brand, or all mentions.
- Does it count answers or mentions? A brand named three times in one answer: does it count once or three times?
- How are brands and competitors defined? Brand name, products or group: in our test this choice alone takes TeamSystem from 50 to 100.
- How many questions and how many runs? Ten questions run once are a snapshot; for stable data the questions need repeating.
- Which engines, and how often? A single score averaging ChatGPT and AI Overviews hides the engine where you're missing. Ahrefs, for example, refreshes ChatGPT, Perplexity, Gemini and Copilot monthly over a 90-day window, and AI Overviews and AI Mode continuously.
Which tools answer these questions best we checked on eight tools in our guide on how to choose an AI visibility tool; the full comparison is in the best AI visibility tools, and running costs are on our page about AI visibility tool pricing.
And in the monthly report?
The score never stands alone. Next to it we put the mention rate per engine, the sources cited and the gap to the top competitor. The full template, including one for agencies with several clients, is in our guide to the AI visibility report. If you'd rather not set it up yourself, in our AI Monitoring service we fix the questions, engines, competitors and formula for you.
The score and citations (what actually moves the number)
The score tends to rise when your brand appears in the sources AI tools read to answer, not when the website improves in the abstract. That's why a high readiness score guarantees no mentions at all.
Our test shows it. In 7 answers out of 10, alongside the products' official sites, ChatGPT cited comparison guides published by others: Shopify Italia, Qonto, Agenda Digitale, partitadigitale.it and Talete365, the last one in 3 answers. A brand missing from those guides has one source fewer from which the answer can pick it up.
The same applies to a page like this one. Over the last 90 days Search Console shows questions reaching visilay.com written out in full, such as "good ai visibility score" (7 impressions) and "what is a good visibility score on ai platforms?": long, conversational questions of the kind people type into AI Mode. AI Mode and ChatGPT extract passages, not whole pages, which is why every section here opens with the answer in the first two sentences. Nobody can guarantee a citation: you can make the passage easy to lift and work on the sources your industry relies on. How AI brand mentions and citations come about we've explained separately, and GEO is the work we do on clients' sites.
Frequently asked questions
No. Each tool uses a different denominator and weighting: in our test on the same 10 ChatGPT answers, the same brand ranged from 8.5 to 67. Only compare scores from the same tool, on the same questions and against the same competitors.
Yes, for a first snapshot: Semrush and Ahrefs both offer a free check on their websites. They use questions they choose, though, so they tell you whether you appear, not how your presence changes month by month.
No, not on its own. It measures how often you appear in answers, not how many people then visit your site or get in touch. Traffic from ChatGPT and other assistants needs reading separately in GA4, as we explain in our guide to ChatGPT traffic in Google Analytics.
It depends: for most businesses a monthly reading is enough, as long as the questions are run several times during the month. Semrush updates its data daily, Ahrefs monthly for ChatGPT, Gemini, Perplexity and Copilot. A consistent sample matters more than frequency.
Yes, as a technical checklist: AI crawlers not blocked, structured data, readable pages. It doesn't tell you whether AI tools name you, though: for that you need a presence score based on real answers.
Summary
An AI visibility score is presence in AI answers reduced to one number.
There are two families: presence, based on real answers, and readiness, based on site signals.
There is no standard formula: on the same data, the same brand scores anywhere from 8.5 to 67.
A good score is one above your competitors, on the same data.
The order between competitors holds better than the number.
The one piece of advice we give clients: in the report, write down the gap, not the score. "Eight points behind the top competitor on ChatGPT, three ahead on AI Overviews" survives a change of tool, formula and even vendor. The 42 or the 61 doesn't.
