Updated on 7 November 2026, by João Leite, Founder of murmur.
To measure brand presence in ChatGPT, ask the same buyer questions several times and count how many answers the brand appears in. Presence is the number of answers in which the brand appears, divided by the total number of answers. The questions are the ones your customer asks before buying, without the brand name, and they stay the same from one round to the next. The same calculation applies to Claude, Gemini, Perplexity and the AI Overview, the AI summary at the top of Google search. The clicks that come from those answers are measured separately, in Google Analytics.
- Presence is a calculation: if the brand appears in 30 of 100 answers, presence is 30%. In English, the market calls this share of voice.
- The questions are the buyer’s, without the brand name, in a fixed list that repeats in every round.
- One round is not enough: in a SparkToro study with 2,961 answers, the chance of ChatGPT or Google’s AI repeating the same list of brands came out below 1 in 100.
- Each AI has its own column, because each one searches and answers in its own way.
- Clicks live in Google Analytics, in the AI Assistant channel, and AI Overview impressions live in Search Console.
What brand presence in ChatGPT is and how to calculate it
Brand presence in ChatGPT is the percentage of answers in which the brand appears. If the brand appears in 30 of 100 answers, presence is 30%. The market calls this calculation share of voice, and monitoring tools call the whole set AI visibility, or visibility in AI. Anyone doing GEO, the work of making the company appear in the answers of ChatGPT and the other AI models, uses presence as the main number. The number is only worth something with the sample size next to it: 30% of 10 questions says much less than 30% of 300. Otterly.AI, which sells AI monitoring, reran questions daily across seven AI models and published in September 2026 that raising the sample from 10 to 100 questions reduced the variation in measured presence by about 73%.
Presence holds up as a metric because it adds up many answers, and a single answer changes all the time. SparkToro asked 600 volunteers to ask 12 questions to ChatGPT, Claude and Google’s AI, for a total of 2,961 answers. In that study, the chance of ChatGPT or Google’s AI repeating the same list of brands in two answers came out below 1 in 100. Rand Fishkin, cofounder of SparkToro, concluded that the percentage of presence across dozens or hundreds of questions, repeated several times, is a reasonable metric. A tool that gives the brand’s position in an AI ranking, he wrote, does not hold up.
In each answer, note how the brand appeared: recommended, only mentioned, cited as a source with a link to your site, or absent. A recommendation counts for more than a mention, and a source with a link is the only one that can turn into a click. Presence adds up the three ways of appearing, and the spreadsheet keeps the difference between them. The full calculation is in the guide on what share of voice in AI is.
A mention only counts when it is about your brand. Visio, operations software for retail, is a murmur client. murmur is a GEO agency in Brazil that places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents. In questions containing the name Visio, the AI models often answered about Microsoft’s diagramming program. murmur’s measurement then started counting the brand only when the answer was about retail or included the visio.ai domain.
Build the list of buyer questions and the spreadsheet
The list of questions comes from what the buyer asks before buying, in the buyer’s words and without your brand name. Gather the questions that reach the sales and customer service teams, the searches Google autocompletes on your topic, the questions in forums and the questions ChatGPT itself suggests after an answer. Mix category questions, such as “qual o melhor sistema para restaurante” (“what is the best system for a restaurant”), with problem, comparison and price questions.
Questions with the brand name go in a separate list. That list shows what the AI knows about the company, not whether the AI remembers the company on its own, which is what presence measures. Adding the two inflates the number, because almost every answer to “a empresa X é boa?” (“is company X good?”) talks about company X.
Each question on the list is a prompt, the question or request a person writes to the AI. The list stays the same from one round to the next, and a new question enters marked with its date, without changing the old calculation. The main KPI, the number the company chooses to track to know whether the work is delivering results, is presence. The full step by step is in the guide on how to build the list of questions to measure brand presence in ChatGPT.
| spreadsheet column | what to note |
|---|---|
| date and AI | the day of the round and the AI model asked |
| question | the exact text, the same in every round |
| the brand | recommended, mentioned, cited as a source or absent |
| competitors | the other brands that appeared in the answer |
| sources | the URLs the answer cited |
| what it says about the brand | whether the information about the company is correct |
With these six columns, the same spreadsheet gives the brand’s presence, each competitor’s presence and the sources the AI uses in your market.
Which AI models to track besides ChatGPT
Track ChatGPT and, alongside it, Google’s AI Overview and AI Mode, Gemini, Claude and Perplexity. AI Mode is Google’s search mode in which the person talks with the AI instead of getting a list of links. The sources Perplexity shows are in the guide on how to appear in Perplexity.
Each AI searches the web in its own way and cites different sources, which is why the presence of one is not added to the presence of another. A brand with 30% in ChatGPT and nothing in Google AIO has two different problems, and adding them up hides both. Choose the AI models based on your buyer: ask customers where they research before buying and start with the ones they mention.
Measuring through the ChatGPT interface, by API or with a monitoring tool
The ChatGPT interface shows the answer the buyer sees; the API and the tools repeat hundreds of questions without manual work. In the interface, ask each question in a new conversation, without logging in or in an account with no history, because memory and previous conversations change the answer. Save the text of each answer and the sources, not just the screenshot. It is the most faithful way and the slowest: 50 questions across three AI models give 150 answers to read and note down.
Location also changes the answer. In the help page about ChatGPT search, OpenAI explains that it may pass on to partner search engines the approximate location of the person asking, which it estimates from each person’s internet address. According to the company, this makes results more precise. Because of that location, murmur does not measure from a team member’s computer: the questions go out from a server dedicated to measurement, in Brazil, which keeps the same internet address from one round to the next and opens the browser in Portuguese. That way, the answer does not depend on the city of whoever is measuring, and each week can be placed side by side with the previous one.
The API is the paid access OpenAI sells so that programs can use its models. Through it, a program asks the same questions as many times as needed and records the answers in the spreadsheet. The API answer is not the ChatGPT screen: web search has to be turned on in the request, and the model and instructions are the ones the program chooses. That is why anyone measuring by API checks a sample in the interface every round.
Monitoring tools do this work on a subscription basis: they run the questions, store the answers and count the mentions. In September 2026, Otterly.AI charges from US$ 29 per month, with 15 questions, to US$ 489, with 400. Its Enterprise plan starts at 1,000 questions, with pricing on request. Peec AI charges from US$ 95 per month, with 50 questions, to US$ 495, with 350, with three AI models to choose from and a daily run. naia, a Brazilian company, charges R$ 1,490 per month on the Autoridade plan, with four analyses per month on ChatGPT, Gemini and Perplexity. The plan card says 220 questions per analysis, and the comparison table on the same page says 20. Profound offers a free 7-day trial, with 50 questions per day, and the paid plan is on request.
The four tools, in alphabetical order:
| tool | price per month, in September 2026 | questions | AI models |
|---|---|---|---|
| naia | R$ 1,490, on the Autoridade plan | 220 per analysis on the plan card and 20 in the comparison table on the same page, with 4 analyses per month | ChatGPT, Gemini and Perplexity, with AI Overview as a complementary collection |
| Otterly.AI | from US$ 29 to US$ 489; Enterprise on request | from 15 to 400; Enterprise from 1,000 | ChatGPT, AI Overview, Perplexity and Copilot; Claude, AI Mode and Gemini separately |
| Peec AI | from US$ 95 to US$ 495 | from 50 to 350 | 3 AI models to choose from, with a daily run |
| Profound | free 7-day trial; paid plan on request | 50 per day in the trial | ChatGPT, Gemini and AI Overview in the trial |
At Otterly.AI and Peec AI, the price rises with the number of questions tracked. The tool delivers the data, and deciding what to publish to change the answer stays with the client. Start with the interface, with the ready-made list and two AI models, and move to the API or to a tool when the list goes beyond a few dozen questions.
How often to measure brand presence and how to see progress
Measure brand presence at least once a month, with the same list, and ask each question more than once in each round. AI answers change from one conversation to another, and measuring every day shows more oscillation than trend. Repeating the question within the round, on different days, separates the brand that appears every time from the one that appears by chance.
A Brazilian e-commerce category, measured by murmur, shows the size of that oscillation. The same 387 questions, asked to ChatGPT again three days later with nothing changed, had the same result in 98% of them: only 7 changed. The rounds were on 30 June and 3 July 2026. That is the natural oscillation of that set of questions. A change smaller than the measured oscillation means nothing, and that is why the first round is repeated before any comparison.
Progress is read question by question, not only in the total. At Mindo, a motion design studio in São Paulo and a murmur client, ChatGPT recommended the brand in 12% of the 134 questions in its market on 25 and 26 June 2026. On 13 and 14 July, it recommended the brand in 23% of the same 134, and among the new ones are the general questions, such as best animation studios, that the buyer asks before having any name in mind. In your spreadsheet, the total shows the balance, and the questions that came in and went out show where the brand gained ground and where it still oscillates.
The comparison is only valid with the same way of measuring: the same questions, the same AI models, the same account and the same type of access. If the tool, the account or the way of asking changes, note the change in the spreadsheet and compare from there. The frequency for each situation is in the guide on how often to measure presence in AI models.
Compare brand presence with that of competitors
To compare brand presence with that of competitors, note in each answer which competitors appear and do the same calculation for each one. Each competitor’s presence uses the same total of answers, and that is why the percentages sit side by side in the same spreadsheet. The comparison is valid within the same measurement: presence measured in one tool is not compared with presence measured in another.
The most instructive list is the questions where the competitor appears and your brand does not. Filter the spreadsheet by those questions and open the sources the AI cited in them. They are the pages the AI used to build the answer, and they show what is missing from your site and which sites your brand still does not appear on.
The visits that come from AI in Google Analytics
In Google Analytics 4, visits that come from ChatGPT, Gemini and other AI models appear in the AI Assistant channel. Google Analytics groups in that channel the visits that arrive from sources such as ChatGPT, Gemini, Deepseek, Copilot or Grok. To see the channel, open Reports, Acquisition and Traffic acquisition. The session source shows the site of each visit, such as chatgpt.com or perplexity.ai.
The AI Assistant channel leaves out AI Overview and AI Mode, and Google says so on the same help page. The click that comes from those summaries arrives as Google organic search, mixed in with the rest. Search Console counts impressions from both in the generative AI performance report, without separating one from the other.
The AI Assistant channel shows the minimum of visits that come from AI, not the total. Anyone who copies the URL of the answer, or searches for the brand on Google after reading about it, arrives as direct traffic or as a search for the company name. The dashboards that count the site inside the AI answer, with or without a click, are in the guide on Search Console and Bing Webmaster Tools.
Brand presence, leads and sales: what can be linked
Brand presence can be linked to leads and sales only in part, because most of the influence of AI models happens without a click. According to Similarweb, fewer than 7% of the questions asked to ChatGPT carried any source citation in May 2026. The effect that remains arrives as a search for the company name, without the AI in the report.
When there is a click, it is worth measuring. In the Leadster panorama with 2,425 Brazilian sites and 2025 data, the visitor who arrived from AI models converted at a median of 7.80%, against 2.39% for organic search. Mark the AI Assistant channel as the lead source and carry that field into the CRM, the system where the sales team records each negotiation. That way the pipeline, the list of deals in progress, shows how many started in an AI model.
For the part without a click, ask. A “how did you hear about the company” field on the form, with the option “ChatGPT or another AI”, and the same question in the first sales conversation show what Google Analytics does not see. Brand presence completes the picture: in how many answers the brand appears, against each competitor, over time.
The GEO KPIs beyond brand presence
Beyond brand presence, track competitor presence, recommendation, citations of your site, correct information and visits from AI models. Each KPI is read in one place:
| KPI | what it shows | where it is read |
|---|---|---|
| brand presence | in how many answers the brand appears, out of the total answers | in the spreadsheet or in the tool |
| competitor presence | who appears in place of your brand | in the same spreadsheet |
| recommendation | in how many answers the AI recommends the brand, not just cites it | in the brand column |
| citations of your site | in how many answers a page of yours appears as a source | in the sources column |
| correct information | whether the AI gets right what the company does, how much it charges and where it serves | in reading the answers |
| visits and leads from AI models | visits and contacts that arrived from an AI model | in Google Analytics and in the CRM |
| impressions in AI Overview and AI Mode | how many times your pages appeared in Google summaries | in Search Console |
Brand presence is the main KPI, and the others explain why it rises or falls. Presence that falls together with citations of your site points to a page the AI stopped using as a source. Wrong information in an answer calls for a fix before any new content, and the guide on how to correct wrong information the AI says about your company has the step-by-step instructions.
murmur makes the brand appear in the answers and monitors presence closely
murmur makes the client’s brand appear in the answers of ChatGPT, Claude, Google AIO and other AI agents. It tracks the brand’s presence alongside competitors, continuously. At murmur, a specialist dedicated to the client’s brand studies the company and the sector and reviews the measurement every round. They point out the questions where the brand is missing, take care of the content that changes that number and track progress from the diagnostic onward.
murmur’s measurement sits in its own software dashboard, included in the service, with the brand’s presence in AI models alongside competitors. The contract is monthly with no minimum term, with no cap on questions, hours or content pieces, and the price starts at R$ 5,000 per month. When reading the answers finds wrong information about the brand, murmur helps correct it and position the brand the right way in AI models.
Frequently asked questions
How many prompts should I monitor?
Monitor from a few dozen to a few hundred questions, depending on the size of your market. With 20 questions, one answer more or less changes presence by 5 percentage points, and natural fluctuation swallows the real change. On Otterly.AI and Peec AI, the entry plan tracks 15 and 50 questions. At murmur, there is no cap on questions: Mindo was measured on 134 questions and FotoRegistro on 361. Where the fluctuation comes from and how many questions the measurement calls for are in the guide on how many questions to use to measure brand presence in AI models.
Is there a reliable metric to know whether the brand appears in the answers, or is it still guesswork?
There is: brand presence, counted over a fixed list of questions and repeated across rounds, is the metric that holds up today. What does not hold up is the position in an AI ranking and the traffic number without the sample alongside it. Always ask for the sample behind any number: how many questions, in which AI models, on which dates and how many times each one. The guide on the reliable metric for brand presence in AI models shows how to check each one.
What should go in the report to defend the result before a skeptical finance director?
In the report, put the brand’s presence alongside that of each competitor, with the basis: how many questions, which AI models and which dates. Add the list of questions, the answers saved as text and the comparison with the previous round on the same questions. Visits and leads from the AI Assistant channel go on a separate line, with the Google Analytics number. A screenshot of an AI answer serves as an example, not as proof, because the answer changes in the next conversation. The argument for the board is in the guide on how to justify GEO to the board.
How do I automate the monitoring of AI citations?
Automate it with a monitoring tool or with a program that asks the questions through the API and saves the answers to a spreadsheet. The part that can be automated well is running the questions, storing the answers and counting the mentions. The part that needs a person is reading a sample every round, because automatic counting gets it wrong with companies that have similar names and with the brand cited only as a source.
Does Search Console show performance in AI Overview?
It does: since June 2026, Search Console has generative AI performance reports. The Search report counts the impressions of your pages in AI Overview and in AI Mode, and Discover has its own separate report. Google launched the reports on 3 June 2026 and released them to all sites from 31 August. The breakdown is by page, country, device and date. In the general performance report, traffic from AI Overview and AI Mode is still counted under the Web search type. A mention of the brand without a link to your site stays out of the report, and that is what counts for presence. Bing Webmaster Tools counts the site’s citations in Copilot and in Bing, and the guide on AI performance in Search Console and Bing Webmaster Tools compares the two dashboards.
Where to start
To measure on your own, build the buyer’s list of questions, open the spreadsheet and run the first round in ChatGPT and in Google AIO. To see now what AI models answer in your market, murmur’s free diagnostic reads your site, writes the questions for your market and shows the AI answers in less than two minutes, with no sign-up. murmur’s sector studies show which brands AI models recommend today in each market studied.
Read also
- How often to measure brand presence in AI models: every month, at minimum
- How to build the question list to measure brand presence in ChatGPT
- How many questions to use to measure brand presence in AI models: from dozens to hundreds
- How to get your company to appear in ChatGPT: the step by step, with no registration or ads
References
- SparkToro, Rand Fishkin, “NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility”, 28 January 2026. https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/ (accessed on 25 September 2026).
- Otterly.AI, “We Measured How Stable AI Citations and Brand Mentions Really Are (and Why 10 Prompts Isn’t Enough)”, updated on 11 September 2026, with 520 questions, seven AI models and 252,407 answers. https://otterly.ai/blog/ai-search-visibility-stability/ (accessed on 30 September 2026). The test of 10 versus 100 questions draws groups from a set of 320. Otterly.AI sells AI monitoring.
- OpenAI, “Searching the web with ChatGPT”, OpenAI Help Center, marked as updated in the month before access, undated page. https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt (accessed on 2 October 2026).
- Similarweb, Maayan Zohar Basteker, “Zero-Click Searches in 2026: What They’re Costing You, and How to Win Anyway”, 10 August 2026. https://www.similarweb.com/blog/marketing/seo/zero-click-searches/ (accessed on 25 September 2026).
- Leadster, Gustavo Luby, “Geração de Leads no Brasil: Panorama 2026 em dados”, updated on 6 July 2026. https://leadster.com.br/blog/panorama-geracao-de-leads-2026/ (accessed on 25 September 2026).
- Google, “Grupo de canais padrão”, Google Analytics Help, undated page. https://support.google.com/analytics/answer/9756891?hl=pt-BR (accessed on 25 September 2026).
- Google, “Introducing Search Generative AI performance reports in Search Console”, Google Search Central Blog, 3 June 2026, with a note from 31 August 2026. https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports (accessed on 25 September 2026).
- Google, “Relatório de performance da IA generativa (Pesquisa)”, Search Console Help. https://support.google.com/webmasters/answer/16984139?hl=pt-BR (accessed on 8 October 2026).
- Google, “AI features and your website”, Google Search Central, updated on 10 December 2025. https://developers.google.com/search/docs/appearance/ai-features (accessed on 25 September 2026).
- naia, pricing page. https://naia.today/pricing (accessed on 25 September 2026).
- Otterly.AI, pricing page. https://otterly.ai/pricing (accessed on 25 September 2026).
- Peec AI, “Pricing for Brands”. https://peec.ai/pricing (accessed on 25 September 2026).
- Profound, pricing page, brands tab. https://www.tryprofound.com/pricing (accessed on 25 September 2026).
- murmur, case studies for Mindo, FotoRegistro and Visio, published with the clients’ authorization: https://murmur.marketing/en/cases/mindo/, https://murmur.marketing/en/cases/fotoregistro/ and https://murmur.marketing/en/cases/visio/; free diagnostic and sector studies: https://murmur.marketing/en/diagnostic/ and https://murmur.marketing/estudos/ (accessed on 25 September 2026).