What Is AI Visibility and How Do You Measure It? A 6-Metric Guide

Your buyers no longer ask only Google when they look for a vendor. They also ask ChatGPT, Gemini and Perplexity: 'Which agency should I use for a corporate website in Ankara?' or 'Which of these two tools is better for small businesses?' If your brand is not in the answer, you never make the shortlist. That is why marketing and sales teams now ask a new question: what is AI visibility, and how do you measure it?

Search results answer this in two ways. Tool pages run a technical 'readiness' scan and present the result as visibility. Definition articles explain the concept but give no measurement method. This guide covers the gap between them. It explains the difference between a readiness score and actual visibility, gives 6 metrics with formulas, and shows a sampling method that handles answer variability. You also get a copyable template and a worked local example with numbers.

What is AI visibility and how is it measured? The short answer

AI visibility is how often, in what position and in what tone AI assistants such as ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews mention your brand when they answer questions in your category. To measure it, you define the questions your buyers ask and run them regularly on each engine. You then record mention rate, list position, citations of your site, share of voice against competitors and accuracy.

What is AI visibility?

In one sentence, AI visibility means that an AI assistant mentions your brand and cites your site when it gives recommendations in your category.

Mention vs. citation

AI visibility has two layers:

• Mention: your brand name appears in the answer text.
• Citation: a page on your site is linked as a source the answer relies on.

Here is an example answer:

'Notable options for dental implants in Izmir include Clinic A, Clinic B and Clinic C. Clinic C stands out for offering a free initial consultation.' Sources: [a directory site], [clinicc.com.tr]

All three clinics are mentioned in this answer. Only Clinic C receives a citation to its own site. The names of Clinic A and B come from a directory page. A mention puts you in the buyer's mind, and a citation brings clicks and trust. You need to measure the two separately.

What is GEO and how does it relate to SEO?

GEO (Generative Engine Optimization) is the umbrella term for work that makes your brand more likely to be mentioned and cited in AI answers. It does not replace SEO. It builds on top of it. Crawlable, clearly written pages are the foundation for both. The difference is that in GEO, what others say about your brand off your site matters at least as much as your own pages.

Which engines matter?

B2B and local service buyers use five main surfaces: ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, which appears above Google search results. You do not have to measure all of them at once. Start with the three your buyers use most.

Is AI visibility the same as Google rankings?

No. A Google ranking is a position among ten blue links, so even at position 4 you are still on the page. An AI answer usually names only a few brands, most often 3 to 5 recommendations, and the rest get no second page.

That means you can rank first on Google and never be mentioned in an AI answer. The reverse also happens. A brand whose own site ranks poorly can still get mentioned because it appears often in forums, comparison lists and review platforms.

The reason is that the engines run a web search before they generate an answer. ChatGPT uses OpenAI's web search, Gemini uses Google Search grounding and Perplexity uses its own Sonar infrastructure. The model reads the pages it finds and combines them into one answer. So the number of source sites that mention your brand, and how consistently they describe it, has a large effect on the result.

Dimension	SEO ranking	AI visibility
Unit of measurement	Position for a query	Mention, position and citation in an answer
Number of winners	10 organic results per page	Usually a few brands
Deciding signals	Page content, links, technical health	Your own pages + off-site sources + consistent description
Variability	Relatively stable	The same question can return a different answer in another session

The difference between an 'AI visibility test' and real visibility measurement

Many free 'AI visibility tests' on the market are actually readiness scans. They usually check these areas:

• Crawling and access: Does robots.txt allow AI bots? Can pages be read without JavaScript?
• Structured data: Is there Organization, LocalBusiness, Product or FAQPage markup?
• Answer-focused content: Do your pages clearly say what you do, for whom, where and at what price?
• Brand signals: Are your name, address and social profiles consistent?

These checks tell you whether your site is readable. They do not tell you whether you are actually mentioned. A site with perfect structured data may never be mentioned if nobody writes about the brand. Real measurement means asking the engines your buyers' questions and recording the answers.

	Readiness score	Visibility measurement
What it measures	Technical readability of the site	Mentions, position, citations, tone and accuracy in answers
What it does not measure	Actual mentions, competitors	The root cause of technical errors
When to use it	At the start and after site changes	Regularly, to track KPIs

Bottom line: a readiness score is a starting check. Your real KPIs are mention rate and share of voice.

6 metrics to measure AI visibility (with formulas)

1. Mention rate = number of answers that mention your brand / total number of answers.

2. Average list position = sum of your positions in the answers that mention you / number of answers that mention you. Do not include answers where you are absent in this average, because mention rate already captures them. Report the two metrics together.

3. Citation rate = number of answers that link your site as a source / total number of answers.

4. Share of voice = your mention count / (your mentions + total mentions of the competitors you track).

5. Tone = mark each mention as positive (+1), neutral (0) or negative (-1). Tone score = (positive − negative) / number of answers that mention you.

6. Accuracy = mentions that match your brand facts / mentions that contain concrete information such as price, address, service or opening hours. Here are some examples of hallucination. The assistant shows your clinic at an old address you moved away from. It describes a service you discontinued as if you still offer it. It quotes a price you never published. Each of these means a lost customer or a customer who arrives with the wrong expectations.

Metric	Formula	Example value
Mention rate	Answers mentioning you / total answers	10 / 45 = 22%
Average position	Sum of positions / answers mentioning you	34 / 10 = 3.4
Citation rate	Answers citing your site / total answers	4 / 45 = 9%
Share of voice	Your mentions / all mentions	10 / 56 = 18%
Tone score	(Positive − negative) / mentions	(2 − 0) / 10 = 0.2
Accuracy	Correct info / mentions with info	9 / 10 = 90%

Step-by-step manual measurement: one spreadsheet and one afternoon

Keep your first measurement cycle small. Five to ten questions on 3 engines are enough for a solid baseline.

1. Split your question set by funnel stage

• Awareness ('what is X'): 'What is pre-accounting software and how does it help a small business?'
• Consideration ('best X'): 'What are the best digital marketing agencies for B2B companies in Istanbul?'
• Decision ('A or B, what does it cost'): 'Company A or Company B, and how much is the setup fee?'

Consideration and decision questions are closest to a sale, so give them more weight in your set.

2. Write questions in your customers' words and language

Buyers do not type 'enterprise solutions partner'. They type 'company that installs ERP for factories in Bursa'. Collect real phrasing from the first messages and form notes your sales team receives. Add city or industry context, because answers to local questions change completely.

3. Manage variability

Answers are probabilistic. Run each question at least 3 times on each engine, either logged out or in a private window. Chat memory and personalization change the result. If you have discussed your brand in your own account before, the answer may be biased. Your sample size is questions × engines × repetitions, so 5 questions × 3 engines × 3 repetitions = 45 answers.

4. Record the same fields for every answer

Record the date, engine, question, whether you were mentioned, position, citation link, competitors mentioned, tone and incorrect information. Also note the source sites the answer used. They are very useful for the next step.

5. Repeat weekly or every two weeks

Rerun the same question set without changes and track the trend. If you change the question set, you can no longer compare results.

Copyable measurement template

Date	Engine	Funnel stage	Question	Run	Mentioned (1/0)	Position	Citation link	Competitors mentioned	Tone (+1/0/-1)	Incorrect info	Source sites
Week 1	ChatGPT	Consideration	Good implant clinic in Izmir?	1	1	3	–	Clinic B, Clinic C	0	None	Directory site

Copy these columns into a spreadsheet. You can calculate every metric with simple COUNTIF and AVERAGEIF formulas.

Example: what does measurement look like for 'dental clinic in Izmir'?

The example below is a hypothetical scenario that shows how the method works. It is not data from a real clinic. We chose these 5 questions for 'Clinic A' in Izmir:

1. Can you recommend a good dental clinic for implants in Izmir?
2. What are the best dental clinics in Karşıyaka?
3. How much does teeth whitening cost in Izmir, and which clinic is affordable?
4. Implant or bridge, and who should I consult in Izmir?
5. Is there a dental clinic open on weekends in Izmir?

5 questions × 3 engines (ChatGPT, Gemini, Perplexity) × 3 runs = 45 answers.

Metric	Result	Interpretation
Mention rate	22% (10/45)	Present in about one in five answers
Average position	3.4	In the lower half of the list when mentioned
Citation rate	9% (4/45)	Own site rarely used as a source
Share of voice	18% (10/56)	Four competitors received 46 mentions in total
Tone	2 positive, 8 neutral	No negatives
Accuracy	90%	One answer gave an old (wrong) address

Why are competitors mentioned? The source lists in the answers show that competitors are mentioned mostly because of third-party pages, not their own sites. These include doctor and clinic directories, booking and review platforms, Google Maps reviews, complaint platforms such as Şikayetvar, forums such as Ekşi Sözlük, and 'best clinics' articles on local news sites. Sources like these are often cited in Turkey. Only the citation links in your own measurement tell you which ones dominate your category.

3 actions from these numbers:

1. Fix the address everywhere. Your Google Business Profile, directory listings and site footer must all show the same address. The wrong address caused the only accuracy error.
2. Get listed on the directories competitors are cited from. Complete your profiles on the two directories and review platforms that appear most often in the citation lists, and start collecting genuine patient reviews.
3. Write pages that answer decision questions. Create pages that clearly explain the price range, process and opening hours for implants and whitening, with an FAQ section and LocalBusiness markup. This is the most direct way to raise your citation rate.

Can you measure AI visibility with Search Console and GA4?

Only partly. Traffic tools show only mentions that people click.

Search Console: Impressions and clicks from Google AI Overviews are not reported separately. They are mixed into the overall 'Web' search type performance. So Search Console gives only indirect signals. For example, queries where impressions rise while clicks fall may mean that searchers get their answer on the results page.

GA4: Set up AI assistant traffic as its own channel group:

1. Go to Admin > Data display > Channel groups > Create new channel group.
2. Add a channel called 'AI Assistants' with the condition Source matches regex: chatgpt.com|perplexity.ai|gemini.google.com|claude.ai|copilot.microsoft.com
3. Move this channel above the Referral channel. Order matters, because otherwise the traffic stays under Referral.
4. In the landing page report, also check for the utm_source=chatgpt.com parameter, which ChatGPT adds to some of the links it shows users.

Server logs: These show whether AI bots visit your site.

Bot	Company	What does it come for?
GPTBot	OpenAI	Collects content for model training
OAI-SearchBot	OpenAI	Indexes pages to show in ChatGPT search results
ChatGPT-User	OpenAI	Reads a page on demand when a user asks
PerplexityBot	Perplexity	Indexes pages to cite in Perplexity answers
ClaudeBot	Anthropic	Collects web content for model training
Google-Extended	Google	Not a separate crawler. It is a robots.txt token that controls whether content crawled by Googlebot is used for Gemini models. It does not appear in logs and does not affect Google Search or AI Overviews.

Limitation: A bot visit shows that your site was read, not that you were mentioned. Traffic data also misses mentions that nobody clicks. That is why answer-level measurement is essential.

Why do I get a different answer to the same question every time?

• Probabilistic generation: The model generates the text again each time, so the brand list and its order can change.
• Model and version differences: Free and paid plans, and even different versions of the same product, may use different models.
• Changing web search results: If the engine searches before answering, the pages it finds that day shape the answer.
• Location, language, session and memory: Your location, interface language and previous chats personalize the result.

The fix is to avoid decisions based on a single screenshot. Look at rates based on repeated runs and at the weekly trend.

After measurement: 4 levers that increase visibility

1. Your own pages: State in the first paragraph what you do, for whom, where and in what price range. Use answer-focused paragraphs, FAQ sections and structured data.
2. Off-site presence: Build profiles on review and directory sites, and appear in industry lists, news and interviews. Engines often learn about your brand from what others say.
3. Comparison content: Get included in 'A vs. B' and 'best X' pages, and publish honest comparison pages on your own site.
4. Consistent brand description: Keep your name, category, location and pricing the same across your site, social profiles and directories. Conflicting information is the most common cause of hallucination.

Source site analysis tells you where to focus these levers. Count the 'Source sites' column in your template, find the sites the AI reads most often before answering, and prioritize those.

Manual or tool? Comparing measurement methods

	Manual spreadsheet	Free one-off test	Continuous tracking tool
Cost	Free, but takes team hours	Free	Monthly subscription
Engine coverage	Whatever you run by hand	A few engines, depending on the tool	Usually several engines
Repetition and trend	Depends on discipline	None, a snapshot	Automatic time series
Competitor tracking	Manual counting	Limited	Share of voice calculation
Alerts	None	None	Change alerts, depending on the tool

A free test answers 'where am I today?'. Continuous tracking answers 'is what I am doing working?'. Once you need more than 20 questions, several engines and competitor tracking, a manual spreadsheet takes hours each week. At that point, a tool makes sense.

For example, Jumpix AI Visibility's free AI visibility check with no sign-up runs 5 questions on 3 engines from your industry and city and gives you an instant snapshot. For continuous tracking, Jumpix AI Visibility repeats the measurement automatically every week on a 9.99 USD per month plan for 1 brand. It also calculates share of voice against up to 5 competitors and lists what to fix.

Conclusion

AI visibility is about getting onto a shortlist of recommendations, not about ranking. To measure it, you do not need an expensive tool. You need the right question set, repeated sampling and a consistent spreadsheet. Use the readiness score as a starting check and track mention rate and share of voice as your real KPIs. Fix accuracy errors first, and set priorities based on source sites. Run your first measurement with 5 questions this week. When you repeat the same questions two weeks later, you will start to see what works.

Frequently Asked Questions

What is the difference between AI visibility and SEO, and does AI visibility affect my site traffic?

SEO measures your position in search results. AI visibility measures whether assistants mention you in their answers and cite you as a source. It affects traffic in two ways: citations bring referral traffic, but when the answer appears directly on the results page, search clicks may drop. Even without a click, a mention puts you on the buyer's shortlist.

Can I see how many times ChatGPT recommended my brand?

ChatGPT does not share this data with brands. You can only estimate it by sampling. Run buyer questions logged out and repeatedly, then calculate your mention rate. chatgpt.com traffic in GA4 shows only the recommendations people clicked.

Can Google AI Overviews visibility be measured in Search Console?

Not directly. Impressions and clicks in AI Overviews have no separate filter in Search Console and are included in overall Web performance. To see whether AI Overviews mentions you, search the relevant queries in a private window and record the answer.

Why do I get different results when I ask ChatGPT the same question myself?

Answers are generated probabilistically. They change with session, memory, location, model version and the web search results at that moment. If you have discussed your brand in your own account before, the result may be personalized. Look at rates from at least 3 runs per question, not at a single try.

How often should AI visibility be measured, and can I measure competitor brands too?

Measuring weekly or every two weeks with the same question set is enough to see the trend. You can add an extra measurement after a major content or PR push. You can measure competitors too. Because you already record the competitors mentioned in the same answers, calculating share of voice needs no extra queries.

Related reading

<!-- rank-agent-related -->

• AI Search Visibility for Clinics: A 2026 Guide for Turkey <!-- rank-agent-link:e8fd01f0-0667-4a09-9ded-58074aeaca39 -->