AI visibility: how to measure whether your brand shows up in ChatGPT, Gemini and Perplexity
You cannot improve what you do not measure. Learn how to build a prompt test, which AI visibility metrics to track, how to read analytics data and what to do when your brand is missing.
The sales director at an engineering consultancy heard something from a new client that made him curious: "I found you through ChatGPT." He decided to check. He opened the assistant, asked which engineering consultancies served manufacturers in his region and found his own firm in the answer, in third place, with a reasonably accurate description.
The next week, he asked the same question a different way, and his firm was gone. He asked in another language, and a competitor that no longer operated in the region showed up. Gemini gave a different answer. Perplexity gave yet another.
His question is the one many leaders are asking now: does my company show up in AI answers or not? And if it does, how do we track that reliably?
This guide answers that question. You will learn what AI visibility is, why it is hard to measure, which metrics to track, how to build a prompt test, how to use Google Analytics, Search Console and server logs, which tools exist and what to do when your brand is missing. If you have not read it yet, start with our guide to generative engine optimization, which explains how AI assistants decide whom to cite.
What is AI visibility?
AI visibility is how often, and how well, a brand appears in answers from AI assistants such as ChatGPT, Gemini, Perplexity, Copilot and Google's AI Overviews and AI Mode when people ask questions related to what the company sells.
It has four dimensions:
- Presence: whether the brand appears in the answer at all.
- Position: where it appears relative to competitors.
- Description: what the AI says about the brand, and whether that is accurate and aligned with its positioning.
- Source: which pages the AI relies on to talk about the brand, and whether any of them are the company's own.
All four matter. Appearing with the wrong description can be as damaging as not appearing. Always showing up last, after three competitors, has limited value too.
Why is measuring AI visibility different from measuring SEO?
Because on traditional Google, a search returns a relatively stable list of results, and tools measure each page's rank. With AI assistants, every answer is written on the fly and can change with each question, the phrasing, the user's history and the model version.
That creates three challenges:
- Answers vary. The same question asked twice can surface different brands. A single test proves nothing.
- Prompts are long and varied. People talk to AI assistants in full sentences, not two- or three-word keywords. There are thousands of ways to ask the same thing.
- Traffic data is incomplete. Often the AI answers and the person clicks nothing. They remember the company name and search for it later, or type the URL directly.
The practical consequence: AI visibility must be measured through sampling, repetition and trends. What matters is not the result of one test but the trajectory over months.
Why measure AI visibility now?
Because buyer behavior has already changed, and most companies still do not know how they appear in this channel. In a G2 survey, 51% of software buyers said they start their research with an AI chatbot. According to 6sense's Buyer Experience Report, the vendor already on the buyer's shortlist on day one wins 77% of the time.
Interest in the topic is mainstream in the US. On Google Trends, "AI visibility" ranks among the strongest marketing searches of the past year, and "generative engine optimization" peaked in June and July of 2026. Gartner, back in 2024, predicted traditional search volume would fall about 25% by 2026 because of AI assistants.
Measuring now creates a baseline. Six months from now, when the channel is more competitive, you will know exactly where you started and what worked.
Which AI visibility metrics should you track?
Track a small set of metrics, always calculated on the same prompt set and repeated at the same frequency. The most useful are mention rate, share of voice, average position, description quality and citations of your own site.
Mention rate
Of all the answers tested, how many include your brand. If you tested 30 prompts three times each and the brand appeared in 18 of the 90 answers, your mention rate is 20%.
Share of voice
Of all mentions of companies in your space across the answers, how many are yours. It shows whether you are gaining or losing ground against competitors.
Average position
When the brand appears in a list, where it tends to sit. Appearing first carries much more weight than appearing fifth.
Description quality
Is what the AI says about the company accurate? Does it describe you the way you want to be remembered? A simple approach is to tag each mention as accurate and on-message, accurate but generic, or inaccurate.
Own-site citations
In assistants that show sources, such as Perplexity and Google's AI Overviews, how many answers cite your website. It shows whether your content is actually being used.
AI referral traffic
How many visits reach your site from ChatGPT, Perplexity, Gemini and other assistants, and what those visitors do next: engagement time, pages viewed, conversions.
Branded search
Growth in searches for your company name on Google can indicate that more people have heard of you, including through AI answers. It is an indirect signal, but a useful one.
How do you build a prompt test to measure AI visibility?
Build a list of 20 to 40 prompts a real buyer would use, spread across different intent types, and run each one in the main AI assistants, several times, under the same conditions. Log results in a spreadsheet and repeat the test monthly.
Step 1: Choose prompt types
Include five types:
- Category: "What are the best B2B marketing consultancies for manufacturers?"
- Problem: "Why isn't my company generating qualified leads?" or "How do I reduce dependence on referrals?"
- Comparison: "What's the difference between a marketing consultancy and an agency?"
- Local: "Commercial roofing contractor in Austin" or "24-hour towing in Phoenix."
- Brand: "What does [your company] do?" and "Is [your company] reliable?"
Brand prompts show whether the AI knows and describes your company correctly. The others show whether it remembers you when the buyer does not know your name yet.
Step 2: Use the customer's language
Write prompts the way customers talk, not the way marketing writes. The best sources are sales calls, customer emails, inbound questions and the queries already visible in Search Console.
Step 3: Choose assistants and languages
Test at least ChatGPT, Gemini, Perplexity and Google's AI Overviews or AI Mode. If you serve multiple countries, test in each market's language. Answers in English and in other languages often differ significantly.
Step 4: Standardize conditions
Use private windows or accounts without history, with memory turned off, to avoid personalized answers. Record the date, the tool and the model version when available. Run each prompt at least three times, because answers vary.
Step 5: Log everything in a spreadsheet
One row per answer, with these columns:
- Date
- Tool
- Language
- Prompt
- Prompt type
- Brand appeared? (yes or no)
- Position
- Competitors mentioned
- Brand description (paste the excerpt)
- Description rating
- Sources cited
- Own site cited? (yes or no)
Step 6: Calculate and compare
With the data logged, calculate mention rate, share of voice, average position and description quality. Month one is your baseline. In later months, compare the trend.
A sample prompt set
To make the method concrete, here is an illustrative prompt set for a fictional industrial maintenance company based in Ohio that serves food and beverage plants.
Category
- What are the best industrial maintenance companies in the Midwest?
- Which companies provide preventive maintenance for food processing plants?
- Who are the leading service providers for bottling line maintenance in the US?
Problem
- How do I reduce unplanned downtime on a beverage production line?
- What should a plant do when in-house maintenance cannot keep up?
- Is it worth outsourcing industrial maintenance?
Comparison
- Preventive vs. predictive maintenance: which is better for a mid-size plant?
- In-house maintenance team vs. outsourced provider: pros and cons.
Local
- Industrial maintenance company in Columbus, Ohio.
- Conveyor and filler maintenance near Cincinnati.
Brand
- What does [company name] do?
- Does [company name] work with beverage plants?
- What do customers say about [company name]?
Notice that problem and comparison prompts make up the largest share. That is where buyers have not yet decided whom to hire, and where appearing makes the biggest difference.
How should you interpret the numbers?
There are no reliable industry benchmarks for AI visibility yet, and results vary widely by sector, region and prompt type. Compare your company with itself over time and with direct competitors, not with generic averages.
Some useful readings:
- On brand prompts, the goal is to appear every time, with an accurate description. Any error here is a priority.
- On local prompts, companies with strong reviews and complete Google Business Profiles usually advance first.
- On problem prompts, appearing in a portion of answers is already a good sign, because competition includes publishers, blogs and large companies.
- On category prompts, progress is usually slowest.
If your share of voice grows quarter over quarter, the strategy is working, even if absolute numbers still look small.
How do you get leadership to care about AI visibility?
Translate it into pipeline language. Executives respond to risk and opportunity, not to mention rates. Three framings tend to work.
First, show the shortlist. Run five category prompts in front of leadership and let them see which competitors the AI recommends and whether the company appears at all. It is a more persuasive moment than any chart.
Second, connect it to buying behavior. If half of software buyers start research with an AI assistant, and the vendor on the day-one shortlist wins most deals, absence from AI answers is a pipeline problem, not a marketing curiosity.
Third, show that the fixes are not wasted. Clearer positioning, better content, case studies with numbers and third-party mentions improve sales conversations and SEO at the same time. The investment pays off even if AI assistants change their rules.
How do you measure AI visibility in Google Analytics?
Google Analytics shows visits that arrive after someone clicks a link inside an AI answer. To see that group, create a custom channel group or segment for the main AI sources.
In practice:
- Create a custom channel group, for example "AI assistants."
- Include sources containing chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai, plus any others that appear in your reports.
- Track sessions, landing pages, engagement time and conversions for that group.
Two caveats. First, not every AI-driven visit is labeled; some show up as direct traffic. Second, volume is usually small at first. The value lies more in learning which pages AI assistants are recommending than in the absolute number of visits.
What about Search Console?
Google Search Console includes impressions and clicks from AI Overviews and AI Mode in its performance reports, alongside regular search. It does not yet separate them cleanly, but it offers clues.
- Watch for impression growth on long, conversational queries.
- Track pages that gain impressions without gaining clicks at the same rate. That can indicate content is being used in AI answers.
- Use queries that already appear as a source of new prompts for your visibility test.
How do server logs help?
Server access logs show which bots visit your site, when and which pages they read. For AI visibility, they answer a basic question: can AI assistants access my content?
Look for crawlers such as GPTBot, OAI-SearchBot and ChatGPT-User from OpenAI, PerplexityBot, ClaudeBot and Google's crawlers. If they are absent, check your robots.txt file and server security settings. If they are present, note which pages they visit most.
Also check that each page's main content is present in the HTML the server delivers, not only after JavaScript runs. Many AI crawlers do not execute JavaScript. A page that looks complete in a browser may be nearly empty to them.
Are there tools to measure AI visibility?
Yes, and the market is growing fast. Several well-known SEO platforms now offer modules that track brand mentions in AI answers, and dedicated tools have emerged that run hundreds of prompts automatically and chart the results.
They help most when the prompt volume is large, when you need to track several markets and languages or when you want to benchmark many competitors. Before buying, check:
- Which assistants the tool covers, and in which languages and countries.
- Whether you can use your own prompts, not just generic lists.
- How often prompts are run, and how many times each.
- Whether the tool shows cited sources.
- Pricing per prompt or per tracked brand.
For many small and mid-size companies, a well-structured monthly manual test delivers most of the value. A tool makes sense when the manual test starts consuming too much time.
What should you do when your brand is missing from AI answers?
First, find the cause. Absence almost always has an identifiable reason, and each reason calls for a different action.
The AI cannot read your site
Signs: AI crawlers are missing from your logs, content depends on JavaScript, key pages are not indexed. Action: fix technical access, move content into the HTML, review the sitemap and internal links.
The AI does not know what your company does
Signs: on brand prompts, the AI gives a vague, wrong or outdated description. Action: make positioning clear and consistent across every channel, with the same description on your website, LinkedIn, Google Business Profile and directories. See how in our guide to brand positioning.
The AI knows the company but does not connect it to the topic
Signs: the company appears when someone asks for it by name, but not in category or problem prompts. Action: create in-depth content on the topics buyers ask about, with data, case studies and direct answers.
Competitors have more off-site mentions
Signs: the cited sources are directories, trade publications and articles where competitors appear and you do not. Action: earn presence in those sources through industry directory listings such as Clutch or G2, bylined articles, interviews and events.
The description is wrong
Signs: the AI attributes services you do not offer, an old address or outdated facts. Action: fix the original sources, such as your website, profiles and listings, and publish current information clearly.
How often should you measure, and how should you report?
Measure monthly with the same prompt set, and do a deeper review each quarter. Content and reputation changes take time to show up in answers, and measuring too often produces more noise than insight.
A simple monthly report includes:
- Mention rate for the month and the previous three months.
- Share of voice against your three main competitors.
- Examples of brand descriptions, good and bad.
- Site pages cited as sources.
- AI referral visits from Google Analytics.
- Three actions for the next month.
The report should end with decisions, not charts.
How does AI visibility usually evolve?
AI visibility tends to grow in layers. First, the AI starts describing the company accurately on brand prompts. Then it begins citing the company in local and niche prompts. Finally, it shows up in broad category prompts, where competition is fiercest.
Local businesses often see faster progress on location-based prompts, because AI assistants draw on map data and reviews. RP Guincho 24h, for example, strengthened its Google Business Profile and surpassed 200 five-star reviews, the kind of signal that matters in both local search and AI answers that use map data.
Specialized companies advance through topical authority. At Royal American Group, consistent, specialized content on executive security, mobility and risk management, published on the blog and LinkedIn, built authority in a niche where few companies produce in-depth content.
How Comelato Company measures and improves AI visibility
At Comelato Company, AI visibility measurement is part of our strategic diagnosis. We build the prompt set from real customer conversations, test it across the main assistants in English and Portuguese where relevant and benchmark against competitors.
Then we address the causes: website technical structure, positioning clarity, in-depth content and off-site presence. We repeat the measurement monthly to show progress. Because we work in both Brazil and the United States, we often see the same brand described very differently in each language, which is one of the first gaps we help close. If you want to know how your company appears today when buyers ask ChatGPT, book a diagnosis. If you are evaluating partners for this work, see how to choose an AI marketing agency.
Frequently asked questions about AI visibility
How can I tell if my company appears in ChatGPT?
Ask. Build a list of questions your customers would ask, test them in a private window with memory off, repeat each a few times and record whether the company appears, in what position and with what description.
Why does the answer change every time I ask?
Because AI assistants generate each answer on the fly, with some variation, and may consult different sources on each search. That is why measurement needs repetition and should be read as a trend.
Can Google Analytics show how many people came from ChatGPT?
Partly. Visits that arrive via links in answers show up with sources such as chatgpt.com. Some visits arrive as direct traffic, and many people never click, they just remember the company name.
How many prompts should I test?
Twenty to forty prompts are usually enough for a small or mid-size company, spread across category, problem, comparison, local and brand questions.
Are paid AI visibility tools worth it?
They are when prompt volume, markets or competitors are numerous. To start, a well-structured monthly manual test delivers most of the value.
Should we track AI visibility in more than one language?
Yes, if you serve or want to serve customers in other markets. AI assistants often answer very differently in each language, and a brand that is well described in English may barely exist in Spanish or Portuguese answers, or the other way around.
Who should own AI visibility inside the company?
Usually whoever owns SEO and content, working closely with sales. Marketing runs the tests and the fixes, and sales contributes the real questions buyers ask. Without that link, the prompt set drifts away from how customers actually talk.
Can a company fix an incorrect AI description quickly?
Partly. For assistants that search the web in real time, correcting your website, profiles and listings can change answers within weeks. Descriptions that come from a model's training data change more slowly, usually with new model versions, which is why consistent, accurate information across many sources matters so much.
By Júlia Comelato, founder of Comelato Company, a positioning, communication and growth consultancy for B2B companies, based in São Paulo, Brazil. Meet the team.