What is AI visibility, and what is a good AI visibility score?
What AI visibility measures, the four numbers that describe it, how to judge your own score, and why it moves from one week to the next.
AI visibility is how often and how favorably AI assistants mention your brand when people ask ChatGPT, Gemini, Perplexity or Google's AI Overviews and AI Mode about your category. There is no universal good AI visibility score: a citation rate of 30% means little until you set it against your competitors on the same questions and engines, over several weeks. Most AI visibility tools express it with four numbers: citation rate, average position, share of voice and sentiment.
Hypervisible, the software we publish, reads those numbers engine by engine on ChatGPT, Gemini, Perplexity and both of Google's AI features, on the same questions every week, which is the setup a score needs before you can judge it.
AI visibility, defined
AI visibility is the presence of a brand in the answers AI assistants write. You are visible when an answer names you, links to your site, or both, on a question your buyers actually ask.
A Google results page lists ten links and lets the reader choose. An AI answer reads like advice from someone who has already chosen: it names three or five companies, often in some order. A brand left out of that short list gets no second chance on page two.
Two distinctions make the term useful. Start with mention versus citation. A mention is your name in the text; a citation, in the narrow sense, is a link to one of your pages among the sources. The two often come apart. Google tells site owners its AI features may show what people say about products and services on blogs, in videos and in forum threads, so a glowing write-up on someone else's site can earn you a recommendation without a single link to your pages. The reverse happens too: your page gets used as a source and a competitor gets the recommendation.
Then there's branded versus unbranded. Type your company name followed by “reviews” and you learn how the AI describes you. Type “best roofer in Denver for hail damage” and you find out whether it puts you forward to a stranger. New customers come from that kind of question, so give it most of the weight.
The search demand is still small: about 880 searches a month in the US for “ai visibility” and 210 for “ai visibility score” (Google Ads estimates, October 2026). Researchers got there first. The paper that named generative engine optimization, by Pranjal Aggarwal and colleagues (KDD 2024), scored a source's visibility by how many words of the AI answer were tied to it, weighted down the later it was cited.
The four numbers that describe AI visibility
Citation rate, average position, share of voice and sentiment describe AI visibility between them. Any one of them, read alone, can mislead you.
| Metric | The question it answers | How it's calculated | Example |
|---|---|---|---|
| Citation rate | Does the AI name you at all? | Answers that name you ÷ all answers | Named in 12 of 40 answers: 30% |
| Average position | How early are you named? | Your rank among the brands in each answer that names you, averaged | 1st to 4th in those 12 answers: 2.1 on average |
| Share of voice | How much room do you get next to competitors? | Your mentions ÷ mentions of every tracked brand | 12 of 60 brand mentions: 20% |
| Sentiment | What does the AI say about you, and how? | Tone of each answer that names you | 9 positive, 3 neutral, 0 negative |
Citation rate
The citation rate, also called mention rate or visibility rate, is the share of answers that name your brand. It's the headline number and the easiest one to explain to a boss: out of 100 answers to your buyers' questions, how many mention you.
Read it engine by engine before you read the total. A brand named in 60% of ChatGPT answers and 5% of Google AI Mode answers has a Google problem, and a blended 30% hides it. Our guide to AI rank trackers explains how tools collect these answers.
Average position
Average position is your rank among the brands an answer names, averaged over the answers that name you. Named first out of five beats named fifth, much as on a results page.
The trap is that it only counts the answers where you appear. A brand named once, in first place, out of 40 answers has a perfect average position and a 2.5% citation rate. Read the two numbers together. Position is also the least stable of the four (more on that below), so it only means something as an average over many answers.
Share of voice
Share of voice compares your mentions with your competitors' on the same answers. If the 40 answers above name brands 60 times and 12 of those mentions are yours, your share is 20%. The formula and its variants are in our guide to AI share of voice.
The number depends on who you count as a competitor. Add a fifth rival to your tracking and your share drops without a single answer changing. Fix the list early and keep it.
Sentiment
Sentiment records what the answer says about you and in what tone. A mention isn't always good news: “known for slow support” is a mention too. Check the facts as well as the tone, because AI answers do repeat old prices or send people to a location that has since closed. When an answer gets something wrong, the pages it links to are the first place to look, as our article on brand mentions in AI answers explains.
One more column belongs next to these four: the sites the AI cites. They show which review platforms and roundup articles shape the answers in your category. Most of the work to raise your score happens on those pages.
What is a good AI visibility score?
A good AI visibility score is one that beats your direct competitors on the same questions and holds, or rises, over several weeks. No public benchmark sets a threshold that applies to every industry, and a score from one tool can't be compared with a score from another.
A composite figure out of 100 blends rate, position and sometimes sentiment, with weights each vendor chooses, so a 62 in one tool and a 62 in another aren't the same measurement. Compare such a score with itself, over time, and look at the citation rate underneath.
Whether 30% is good depends on what sits behind it:
- Which questions. A question that names your brand will nearly always get an answer that names you, since the question is about you. Mix ten of those into twenty unbranded questions and a weak brand can post a strong total. Judge the unbranded rate on its own.
- Against whom. A 25% citation rate is strong if your closest competitor sits at 10%, and weak if three of them are at 50%. The comparison only holds on the same questions, asked to the same engines, in the same week.
- On which engine. Each AI assistant searches and writes in its own way. Even inside Google, AI Overviews and AI Mode can run on different models and techniques, by Google's own account, and one question may get different links in each. A brand can do well in one and badly in the other.
- How wide the category is. In a study SparkToro published in January 2026, answers about cloud computing kept returning the same top brands, while answers about science fiction novels scattered across many titles. When 142 people each wrote their own question about headphones, Bose, Sony, Sennheiser and Apple appeared in 55% to 77% of the 994 answers. If names that famous miss at least 23% of the answers, a smaller company shouldn't expect to appear in all of them.
The shape of the results then points to the fix:
- Named on branded questions, absent on unbranded ones. The AI knows who you are but doesn't think of you when someone asks for a recommendation.
- Named often, but low in the lists. You're in the running. The sources cited for the brands above you show what they have that you lack.
- Named with wrong facts or a lukewarm tone. Fix that before chasing a higher rate. More mentions of a wrong description won't help.
- Named without a link to your site. Other pages speak for you. Find them among the cited sources and make sure they are accurate.
Our view: pick your comparison set before you judge your score. For a local business, that means the three or four firms in your city that win the jobs you want. A dental practice in Austin compares itself with the other practices in Austin.
How to check your AI visibility score
You can check your AI visibility by hand with ten questions and an hour of work, or with a free tool that asks the same questions to several engines at once. Either way, one check shows where you stand today, and only repeated checks show a trend.
By hand:
- Write ten questions your buyers ask, at least six of them without your name. A roofing company in Denver might use “best roofer in Denver for hail damage” and “metal or asphalt roof for a Colorado house”, plus its own name followed by “reviews”.
- Ask each one in ChatGPT, Gemini, Perplexity and Google AI Mode (google.com/ai), in a private window and signed out. Google personalizes AI Mode answers for signed-in users who allow it.
- Log every answer: named or not, at what rank, which competitors sit beside you, which sites are linked, and what the text says about you.
- Divide the answers that name you by the total, engine by engine.
Forty answers by hand is manageable once. Every week, on five engines, it turns into a part-time job.
For a deeper one-off review, an AI visibility audit goes question by question. Small businesses can start with how customers find you in ChatGPT.
Reading your AI visibility score in Hypervisible
Hypervisible is the tracker we publish, and we point readers of this guide to it first because it holds still what a score needs held still: the questions, the engines and the competitor list.
What that gives you, in practice:
- Each engine on its own line. Every plan covers ChatGPT, Gemini and Perplexity plus Google's two AI features, AI Overviews and AI Mode, so a strong ChatGPT rate can't mask a weak Google one.
- The numbers from this guide, and the sources behind them. Your citation rate and average position in the answer, your rank among competitors in identical answers, the websites the AIs draw on, and the tone they take about you.
- A Monday reading. Paid plans re-run the same questions each Monday and show the gap with the last reading in points, which is how you separate a trend from a flip.
- More than one market. It measures France, the United States, the United Kingdom and Germany, with questions written in English, French or German, for brands that sell abroad.
NEXUS GEO, an agency that works only on GEO, builds it from data from Google and the AI engines. It has two limits. Readings come once a week, so a one-day spike can fall between two Mondays, and Claude and Copilot aren't measured.
To see where you stand first, the free AI visibility checker puts 10 questions to the five AIs, five about your industry that never mention you and five that use your name. Within 5 to 10 minutes it has read the 50 answers; no card is asked for, and your account keeps the results. You launch it from the scan page.
Why your AI visibility score moves from week to week
AI answers are written fresh for each request, and the models and web pages behind them keep changing. On a small question set, each of those changes is worth several points, so most week-to-week movement is noise, and a change that lasts over several readings is a signal.
Answers vary from one run to the next
Ask the same question twice and you'll often get two different lists. In November and December 2025, Rand Fishkin of SparkToro and Patrick O'Donnell of Gumshoe.ai, an AI tracking startup, had volunteers send 12 recommendation prompts to ChatGPT, Claude and Google's AI answers, 2,961 runs in all. On ChatGPT and Google, an identical list of brands turned up in fewer than 1 run out of 100; the same list in the same order, closer to 1 in 1,000. The study wasn't peer-reviewed, and Fishkin called for larger follow-ups. Its most useful finding for anyone reading a dashboard: how often a brand appeared across many runs was steadier than where it ranked.
Small samples magnify every flip
With 20 questions on one engine, each answer is worth 5 points. Two answers that flip between one week and the next move your citation rate by 10 points with nothing real behind it. The more answers a reading includes, the less a random flip weighs.
Models and sources change
AI assistants switch models on their own schedule. On May 19, 2026, Google made Gemini 3.5 Flash the default model for everyone in AI Mode, and a new model can reorder recommendations overnight. The web moves too. AI Mode splits each question into several searches on Google's index and writes from the pages that come back, so a new comparison article or a batch of fresh reviews can change who gets named. Personalization adds a layer, which is one reason your own phone and a tracking tool can disagree.
How to tell a real change from noise
Keep the same questions, engines and competitor list from one reading to the next, because changing any of them breaks the comparison. Then look at which questions moved. A gain on four related questions means more than a jump on one. We act on a change once it has held over several readings in a row.
On Hypervisible's paid plans, the same questions are re-run every Monday and the gap with the previous reading appears in points (see the pricing page). A spike that lasts one day can slip between two Mondays; daily tracking would catch it.
Raising the score is a separate job, covered in our guide to generative engine optimization. Be wary of anyone who promises you a set spot in AI answers: they change between sessions, as Google results do. Getting cited is something you can work on and then check.
The free analysis asks 10 questions to five AI engines (ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode), reads the 50 answers in 5 to 10 minutes and keeps your dashboard in your account.
Frequently asked questions
What is a good AI visibility score?
There is no universal threshold. A good AI visibility score is one that beats your direct competitors on the same questions and engines, and holds or rises over several weeks. Judge the citation rate on questions that don't name your brand, since questions that do name you inflate the total. The category matters too: in a January 2026 SparkToro study, even Bose, Sony, Sennheiser and Apple appeared in 55% to 77% of answers about headphones.
How can I check my AI visibility score?
Hypervisible's free checker, which we publish, is the quickest route: 10 questions to five AI engines (ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode), 50 answers read in 5 to 10 minutes, no credit card, results kept in your account. By hand, write ten questions your buyers ask, most without your brand name, ask each one in those engines from a signed-out browser, and divide the answers that name you by the total, engine by engine.
How do you increase AI visibility?
Start with the sources. See which review sites, directories and articles the AI cites for your category, then make sure you're present there and described accurately. On your own site, Google advises pages it can crawl and index, content that adds first-hand information instead of repeating what's already everywhere, and up-to-date Business Profile and Merchant Center data. Then ask the same questions every week to see what worked.
Is AI visibility the same on every engine?
No. ChatGPT, Gemini, Perplexity and Google's AI answers use different models and draw on different sources, so the same question can bring up different brands in each. Even within Google, AI Overviews and AI Mode may use different models and techniques, according to Google's documentation. A brand can be named often in one and rarely in another, which is why a citation rate should be read engine by engine before it is averaged.
How often should you measure AI visibility?
Often enough to see a trend, on the same questions each time. Hypervisible, the tool we publish, re-runs a fixed question set every Monday and shows the change in points, which is enough to follow a trend and report it once a month. Daily tracking catches a sudden drop sooner, which helps during a launch or a bad news cycle. Because individual answers vary from run to run, a single check, however detailed, only shows where you stood on the day it ran.
Sources
- NEW Research: AIs are highly inconsistent when recommending brands or products, SparkToro, January 28, 2026, read October 22, 2026
- AI Recommendations Change With Nearly Every Query: SparkToro, Search Engine Journal, January 30, 2026, read October 22, 2026
- GEO: Generative Engine Optimization, Aggarwal et al., arXiv, accepted at KDD 2024, read October 22, 2026
- Get AI-powered responses with AI Mode in Google Search, Google Search Help, read October 22, 2026
- AI features and your website, Google Search Central, read October 22, 2026
- Optimizing your website for generative AI features on Google Search, Google Search Central, updated July 10, 2026, read October 22, 2026
- Google Search's I/O 2026 updates, Google blog, May 19, 2026, read October 22, 2026
- About, NEXUS GEO, read October 22, 2026