AI visibility metrics are the numbers that tell you whether, how often and how a brand appears in answers from ChatGPT, Gemini, Perplexity and Google's AI features: mention rate, citation rate, position in the answer, share of voice and sentiment, plus the impressions and traffic you read in Search Console and GA4. In our test on 27 October 2026, eight Italian questions to ChatGPT about CRM software, the most mentioned brand and the most cited website were not the same: HubSpot and Zoho appeared in 7 answers out of 8, but the source cited most often was migliorcrm.it, an Italian comparison site, also in 7.
The marketing director asks one question: "Are we on ChatGPT or not?"
The honest answer is another question: by which metric?
On Google the question has one number. Take Sunscape, which develops solar plants across Italy: in six months from March 2026 its organic impressions went from 329 to 5,809 a month and clicks from 46 to 195, as we explain in how long SEO takes. Impressions and clicks have one definition, and it is Google's.
In AI answers there are five definitions, sometimes seven. And they produce different rankings.
The mistakes we see most often:
- counting mentions only, and finding out late that every cited source belongs to someone else;
- reporting a share of voice without saying how many competitors it is calculated on;
- tracking questions that already contain the brand name, where showing up is a given;
- ignoring Search Console's generative AI report, available to every site since August 2026.
What you will find here:
- the seven metrics, with definition and formula;
- our test on 8 Italian questions: four metrics, three different leaders;
- the two metrics already sitting in Search Console and GA4;
- which one to look at first, depending on your goal.
Let's start.
Which AI visibility metrics should you track? (seven, in two families)
There are seven metrics to track: five are calculated on a sample of AI answers (mention, citation, position, share of voice, sentiment) and two are read in your own accounts (impressions in Google's AI features in Search Console and traffic from AI assistants in GA4). The first five tell you whether AI talks about you; the last two whether that presence turns into visits.
| Metric | What it measures | Formula | Where it comes from |
|---|---|---|---|
| Mention rate | How many answers include the brand name | answers with the brand / total answers | A fixed set of prompts, by hand or with a tool |
| Citation rate | How many answers link to your site as a source | answers linking your domain / total answers | Same sample |
| Position in the answer | Where you appear in the list of recommended brands | average position when you appear | Same sample |
| Share of voice | Your share versus competitors | your appearances / appearances of all brands counted | Same sample, with competitors |
| Sentiment | The tone the answer uses about you | share of positive, neutral, negative mentions | Same sample, read or classified |
| Impressions in Google's AI features | How often links to your site are shown in AI Overviews and AI Mode | provided by Google | Search Console, generative AI report |
| Traffic from AI assistants | Sessions from ChatGPT, Perplexity, Gemini, Copilot | GA4 data | GA4, source or dedicated channel |
The formulas are the ones vendors publish: Peec AI defines visibility as answers mentioning the brand over total answers, Semrush defines share of voice as the percentage of mentions versus competitors in your market. Many tools then blend them into a single score, and why two scores cannot be compared is covered in what an AI visibility score is. Here we go one step earlier: the metrics one by one, on the same answers.
The test: 8 questions on CRM software, four metrics, three different leaders
On the same 8 ChatGPT answers the leading brand changes with the metric: Zoho wins share of voice, HubSpot wins first place, Flusia wins citation as a source.
Method. On 27 October 2026 we asked ChatGPT 8 questions an Italian small business might ask when choosing a CRM, with web search on and location set to Italy, through DataForSEO. Two examples: "what is the best CRM for a small Italian business?" and "which CRM integrates with Italian electronic invoicing?". One run per question. None of the brands is a client of ours: we needed a category where answers name products and cite sources.
The 8 answers listed 42 products. For each brand we counted mentions, presence in the list, position, share of listed items and links to its domain. Zoho CRM and Zoho Bigin are counted as one brand.
| Brand | Mentioned (of 8) | First in list | Average position | Share of items (of 42) | Site linked | Site used as source |
|---|---|---|---|---|---|---|
| HubSpot | 7 (88%) | 3 | 2.7 | 14% | 5 | 1 |
| Zoho | 7 (88%) | 0 | 2.6 | 19% | 5 | 0 |
| TeamSystem | 6 (75%) | 1 | 2.8 | 14% | 5 | 0 |
| Pipedrive | 5 (63%) | 0 | 3.8 | 10% | 4 | 1 |
| Mokapen | 4 (50%) | 2 | 2.8 | 10% | 3 | 0 |
| Flusia | 3 (38%) | 1 | 3.0 | 7% | 3 | 3 |
Read the table column by column and you get three winners.
- By mention rate HubSpot and Zoho tie at 7 out of 8.
- By share of voice Zoho wins with 8 items out of 42, because in one answer it appears twice (Zoho CRM and Bigin).
- By first place HubSpot wins, first in 3 answers. Zoho is never first.
- By citation as a source Flusia wins: an Italian CRM mentioned only 3 times, used as a source all 3 times.
And the other sources?
Almost all third parties. migliorcrm.it, a comparison site, is cited in 7 answers out of 8. b2bcompare.net in 3. TeamSystem's website is linked in 5 answers, but always as the "official site" at the end of the entry: ChatGPT took prices and integrations from the comparison sites.
One limit, before you point it out: one run per question is a snapshot. The SparkToro and Gumshoe study of January 2026 repeated the same prompts 60 to 100 times: the same list of brands came back in fewer than 1% of cases. How much the numbers move from day to day is what we measure in our study on AI visibility data accuracy.
Counting mentions is not enough when the sources belong to others.
You have just seen a brand mentioned in 6 answers out of 8 that ChatGPT never uses as a source: prices and integrations come from comparison sites. If your report only counts mentions, that gap stays invisible. We measure mentions and citations on your customers' questions and show you which pages, yours or third parties', are building the answer.
Book a strategy callMention rate (the baseline, and why on its own it misleads)
Mention rate is the percentage of answers in which the brand name appears in the text. It is the simplest metric and the one most tools put at the centre.
Its limit shows in the test: HubSpot and Zoho are both at 88%, yet HubSpot is first in 3 answers and Zoho in none. The same mention rate can hide two very different presences.
Two rules:
- separate prompts with and without the brand name. Of the nine keywords that bring the most enquiries to Macropix, an Italian LED screen manufacturer, none contains the company name. Prompts work the same way: the visibility that matters is on generic questions;
- decide up front what counts as a mention. In the test HubSpot appears in one answer only in a closing sentence ("the choice widens to Zoho, HubSpot and Pipedrive"). We counted it; a stricter tool would not.
Which prompts to include, how many and how often to rerun them is covered in our guide to prompt tracking.
Citation rate (when the source is you)
Citation rate is the percentage of answers that link your domain as a source. A mention says the AI knows you; a citation says it is using your pages to answer, and it is the only one that comes with a clickable link.
In the test the gap is stark. TeamSystem has 6 mentions and zero citations as a source. Flusia has 3 mentions and 3 citations: its pages on "CRM for SMEs" and on the Fatture in Cloud integration answer the exact question, and ChatGPT uses them.
International data points the same way. AirOps, analysing 800 queries and more than 45,000 citations in 2025, found that only 28% of answers include brands that are both mentioned and cited, and that brands with both are 40% more likely to resurface in later runs than brands that are only cited. Those queries were in English, not Italian.
Not every link counts as a citation: many "official site" links support no statement at all, and we only count links attached to a sentence in the answer. The difference between citations and mentions, and how to earn both, is in AI brand mentions and citations.
Position in the answer (first name or fifth?)
Position in the answer is where the brand appears in the list of recommendations, usually averaged over the answers it appears in. It matters because someone reading a list of five CRMs rarely gives the fifth the attention of the first.
In the test Mokapen has half of HubSpot's mentions but two first places: when the question is about budget or Italian-made software, ChatGPT puts it at the top. Pipedrive's average position is 3.8: it appears, but almost always after its competitors.
Check how your tool calculates it. Semrush measures the position of your domain's citation among the sources, not of your name in the list. Two different numbers under the same label.
How do you measure AI search visibility through share of voice? (the denominator decides)
Share of voice in AI search is your brand's appearances divided by the appearances of all brands counted, on the same answers. The result depends on who sits in the denominator: only the competitors you choose, or every name the AI comes up with.
In the test we used all 42 listed products. Zoho takes 19%, HubSpot and TeamSystem 14%. Had we counted only the six brands in the table, every share would rise, and Flusia would go from 7% to 10%.
Which denominator is right?
The one you keep fixed over time. The full calculation, comparing classic share of voice with share of voice in AI answers, is in our share of voice guide.
Sentiment (what AI says about you, not just whether it names you)
Sentiment is the tone the answer uses about the brand: positive, neutral or negative, often on a scale from -1 to +1. In product answers it is rarely openly negative; it hides in the caveats.
In the test the caveats repeat. HubSpot gets the same one in 3 answers out of 8: costs rise once you need automation. Zoho, in 2 answers out of 8, carries Italian-language support only through a partner. For a company selling to Italian SMEs that sentence weighs more than a point of share of voice.
How to read these caveats and where they come from is covered in how AI describes your brand.
The metrics already in your accounts (Search Console and GA4)
Two metrics need no prompt sample: impressions in Google's AI features, in Search Console, and traffic from AI assistants, in GA4. They are the only ones with a single definition, set by the platform.
Search Console. The generative AI performance report counts how often links to your site were shown in AI Overviews and AI Mode. Google announced it in June 2026 and since 31 August 2026 it has been rolled out to all websites. It shows impressions only, split by page, country, date and device, and excludes Search Labs experiments. If the report is missing, the site usually does not yet have enough impressions in AI features.
GA4. In the test every ChatGPT link used as a source carried the parameter utm_source=chatgpt.com, so in GA4 those visits can be read by source. Perplexity, Gemini and Copilot need a referral filter, and part of the traffic still lands in Direct. The full setup is in our guide to ChatGPT traffic in Google Analytics.
These two metrics tell you how much presence turns into visits. The five sample-based ones tell you why.
Which metric to look at first (it depends on the goal)
The first metric to watch depends on what you want: awareness, traffic or enquiries. None of the seven works alone, but one has to lead.
| Goal | Lead metric | Control metric |
|---|---|---|
| Become known in a category | Mention rate on prompts without the brand | Share of voice |
| Bring visits to the site | Citation rate | AI assistant traffic in GA4 |
| Beat a specific competitor | Position in the answer | Share of voice on the same set |
| Protect reputation | Sentiment | Sources cited when you are named |
| Understand Google's weight | AI feature impressions in Search Console | Organic clicks |
For a typical client selling to SMEs, we start from the pair mention plus citation: it tells you whether you are named and whether your pages are the source. Every month we put the two columns side by side for the same competitors. What goes in the rest of the report is covered in the AI visibility report guide.
ScryRank, the monitoring tool developed by Visilay, calculates mention rate, citation rate, share of voice and sentiment from -1 to +1 on ChatGPT, Claude, Gemini and Perplexity, plus presence among AI Overview sources, with plans from 39 euros a month (price list checked on 27 October 2026). How to compare it with other tools on the same criteria is in how to choose an AI visibility tool; if you would rather hand the work over, see our AI monitoring service.
What makes a brand more citable (what the test shows)
A brand becomes more citable when it has pages that answer the exact question with verifiable data, and when the third-party sites AI consults talk about it. The test shows both sides.
On the comparison side: migliorcrm.it is cited in 7 answers out of 8 because its pages carry prices in euros and the update date in the title ("Ottobre 2026"). ChatGPT took TeamSystem's and Mokapen's prices from there. Being present and accurate on those sites matters as much as your own.
On the vendor side: Flusia has one page for the Fatture in Cloud integration and one for CRMs for SMEs. On the e-invoicing question it is first in the list and the source of the answer. One page per real question, as with Google.
None of this guarantees a citation. The metrics are there to show whether it works: it is the work we do in GEO projects, and the starting point of our whole cluster on AI visibility.
Frequently asked questions
No. A mention is the brand name in the text of the answer; a citation is a link to your domain used as a source. In our test TeamSystem had 6 mentions out of 8 and no citation as a source.
No. Search Console's generative AI report only counts impressions of your links in Google's AI Overviews and AI Mode. For ChatGPT, Gemini and Perplexity you need a sample of prompts rerun over time, by hand or with a tool.
It depends on the denominator: the brand's appearances divided by the appearances of all brands counted on the same answers. In our test Zoho was at 19% of all 42 products named; counting only six brands, every share goes up. What matters is keeping the set fixed over time.
It depends: for most companies a monthly report is enough, as long as each prompt is rerun several times during the month. Answers change from one run to the next, so a single monthly measurement is a snapshot, not a trend.
Summary
There are seven AI visibility metrics: five on answers, two in your accounts.
Each metric can crown a different leader: in the test Zoho, HubSpot and Flusia.
A mention is not a citation: TeamSystem, 6 mentions and no source.
Share of voice depends on the denominator: keep it fixed.
Search Console measures Google only, and impressions only.
The most cited sources were third parties: migliorcrm.it in 7 answers out of 8.
One last thing the test taught us: our AI visibility reports now carry a column listing the domains cited when the client is not. It tells you where to work next month, and it is almost always a page that is not yours.
Until next time!
