Updated on 19 October 2026, by João Leite, Founder of murmur.

Brand presence, counted across many repeated questions, is the reliable metric for knowing whether the company appears in ChatGPT answers. Presence is how many answers the brand appears in, divided by the total number of answers; the market calls this share of voice. Measuring stops being done in the dark when the number brings its base alongside it: how many questions, which AI models, what dates and how many rounds. A position in an AI ranking and a screenshot of one answer do not support a decision, and traffic numbers measure the click, not the presence.

  • Brand presence is the metric that holds up: it counts how many answers the brand appears in, across a fixed list of repeated questions.
  • One answer proves nothing: in a SparkToro study, the chance of ChatGPT or Google’s AI repeating the same list of brands came in below 1 in 100.
  • More questions, less fluctuation: in an Otterly.AI study, going from 10 to 100 questions, in a set of 320, reduced the variation in presence by about 73%.
  • Every number needs its base: who measured, what was counted, how many questions, which AI models, what dates and how many rounds.
  • AI traffic is a different calculation: it measures the visit that reached the site, not how many answers the brand appeared in.

Brand presence is the reliable metric because it adds up many answers

Brand presence is reliable because it adds up many answers, and a single ChatGPT answer changes from one conversation to the next. SparkToro, an American audience research company, published a study in January 2026 with 600 volunteers and 12 questions. There were 2,961 answers from ChatGPT, Claude and Google’s AI. The chance of ChatGPT or Google’s AI giving the same list of brands in two answers came in below 1 in 100.

Rand Fishkin, co-founder of SparkToro, changed his mind because of the study and wrote in the conclusions:

“I’ve been swayed from my initial position and now believe visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric.”

The presence percentage across dozens to hundreds of questions, asked several times, is the part of measurement that is useful. Position in an AI ranking is left out. In the same conclusions, Fishkin writes that “any tool that gives a ‘ranking position in AI’ is full of baloney”.

Otterly.AI, which sells an AI monitoring tool, reached the same conclusion in a study with 520 questions, seven AI models and 252,407 answers. The company recommends brand coverage as the main number: how many of the tracked questions the brand is mentioned in. In the set of 320 questions from the same study, going from 10 to 100 questions reduced the variation in that number by about 73%.

Brand presence only becomes a reliable number under three conditions. The questions are the ones the buyer asks, without the brand name, in a list that does not change from one round to the next. Each question is asked more than once, on different days, and each time in a new conversation, without the memory of previous ones. And each AI model has its own number: the brand can appear a lot in ChatGPT and not at all in AI Overview, Google’s AI summary.

Where the person asking is located also changes the answer. According to OpenAI’s help page on searching the web, ChatGPT infers the user’s approximate region from their internet address and may send it to the search providers it works with. The stated reason is “to improve the accuracy of your results”, that is, to make the results more accurate. So that this effect does not contaminate the number, murmur uses a dedicated server, installed in Brazil, instead of a team member’s computer. On that server, the internet address is always the same and the browser is in Portuguese. That way, the city of the person measuring stays out of the answer, and one week’s number can be compared with the next.

murmur, a GEO agency in Brazil that places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents, publishes sector studies on what AI models answer in each market. Each study states how many questions were asked, to which AI models and in which month, and it is a snapshot of one round, not ongoing monitoring. The method of each study is written on the page: ask the questions a buyer would ask, in the tools where they ask them, and record which brands appear in the answer.

What to ask before trusting a brand presence number

Before trusting a brand presence number, ask for the basis: who measured it, what was counted, how many questions, which AI models and when. In Brazil, the advertising self-regulation code from CONAR requires proof for any objective data used in an ad. Article 27 says that such claims “devem ser comprobatórias, cabendo aos Anunciantes e Agências fornecer as comprovações, quando solicitadas” (“must be substantiated, and Advertisers and Agencies are responsible for providing the substantiation when requested”).

what to askwhy ask
who measured, and whether they sell what they measureda study by someone who sells a tool or service is valid, as long as the method is open
what was countedmention, recommendation, link to the site and click are four different numbers
how many questions, and whether they include the brand namea small list fluctuates a lot, and a question with the brand name inflates presence
which AI models, each with its own numberthe average across several AI models hides the AI model where the brand is missing
the dates and how many roundsthe difference between two months only counts if it is larger than the fluctuation between rounds
how the questions were askedon the ChatGPT screen, by program or by tool, always the same way from one round to the next

A presence number that answers these six questions can be compared from one month to the next. Without the answers, it serves as a hint, not as a basis for deciding where to invest. The same six answers are what a report needs to bring to convince a skeptical finance director. To compare the method of those who publish their own research on AI answers, the guide on own research to appear in ChatGPT applies similar criteria to six companies.

In monitoring tools, the list of questions behind the number depends on the plan. Otterly.AI tracks 15 questions on the US$ 29 per month plan and 400 on the US$ 489 one. The Enterprise plan starts at 1,000 questions. Before reading the percentage on the screen, check how many questions are behind it.

The AI numbers that do not hold up: position, screenshot and a question with the brand

Three numbers do not hold up: position in an AI ranking, a screenshot of one answer and presence in questions with the brand name. Position changes from one answer to the next, because ChatGPT’s list of brands almost never repeats. The brand that opens the list in one conversation can disappear in the next, with the same question. A report that shows “average position in AI” summarizes a list that the AI itself does not repeat.

A screenshot of one answer shows that the brand appeared once, for one person, on one day. It illustrates the report, but it does not prove presence, because the same question asked again almost always brings a different list. The guide on AI answer screenshots as proof goes deeper into the case.

Presence counted in questions with the brand name inflates the result. Whoever asks “vale a pena contratar a empresa X?” (“is it worth hiring company X?”) gets an answer about company X, and the brand appears in almost all of them. These questions show what the AI knows about the brand, and their count is kept separate from presence.

AI traffic numbers: what they measure and what they leave out

AI traffic numbers count visits that arrived from an AI model, not how many answers the brand appeared in. Most of the influence of AI models happens without a click. The answer cites the brand without a link, or the person reads the recommendation and looks up the company by name afterwards. That visit enters the report as a search by name or a direct visit, and the AI gets no credit.

The click that does exist appears in the company’s own Google Analytics. The step-by-step instructions for separating these visits are in the guide on how to measure brand presence in ChatGPT. Brand presence shows the part the click does not see: how many answers include the brand and how many include each competitor, over time. In the conversation with the board, the two numbers go side by side, each with its own base.

The AI traffic number also changes with who is counting and with how they count. Search specialist Duane Forrester gathered seven numbers published over 13 months for the same figure, all attributed to Cloudflare. The calculation is how many pages the crawlers of Anthropic, the owner of Claude, read for each visit that Claude sends to websites. The numbers ranged from 2,237 to 1 to 70,900 to 1. They changed with the period, with the crawlers included and with the network that did the collection. Forrester notes in the text that he also makes an AI measurement product. His rule: “A figure without its window, its grouping, and its collection boundary is not a figure.”

An AI traffic number cited by an agency or a tool calls for four answers before it enters planning. What is the period? How many sites were included, from which sector and from which country? Is it a mean or a median? And how was the visit recognized as coming from AI? A number from another country or another sector does not become a forecast for your site.

How to compare brand presence with competitors without inflating the number

The comparison with competitors is only valid when all brands are counted on the same questions, in the same AI models and on the same dates. The comparison number inflates in three ways. The first is choosing the questions in which your own brand already appears. The second is counting the brand on questions with its name and competitors on general questions. The third is putting side by side numbers from different tools, with different lists and dates.

The comparison that becomes an ad has a rule in Brazil. The CONAR code accepts comparative advertising as long as it is objective and “passível de comprovação” (“open to verification”), in article 32. In practice, a presence comparison without the list of questions and without the dates does not let anyone check it.

murmur works on brand presence with no cap on questions

murmur works on brand presence with no cap on questions: it makes the brand appear in the answers and tracks the number against competitors. With no cap, the plan you sign does not shrink the sample behind the percentage. In plans that go up in price with each tier of questions, it is the plan that decides the size of the list.

At murmur, the number comes with someone who answers for it. A specialist dedicated to the client’s brand studies the company and the sector. They analyze what AI models answer in that market and show who appears in place of the brand. It is the same person from the diagnosis onward, so the person who explains the presence is the same one who handles the content to change it.

murmur works with its own method and technology. The technology is the dashboard that shows brand presence alongside competitors, included in the service and not sold separately. The production method follows studies in the field published by universities and conferences, and the content answers the real questions of the client’s sector.

Frequently asked questions

How many prompts should I monitor for the number to be reliable?

Monitor from dozens to hundreds of prompts, and the more prompts, the smaller the fluctuation in presence. A prompt is the question a person writes to the AI. Otterly.AI measured the same brand repeatedly, in the same AI models and over the same 30 days, with prompts drawn from a list of 320. With 10 prompts, 90% of the measurements fell between 39% and 71% presence; with 100, between 51% and 60%. Where the fluctuation comes from and why ten questions are not enough is in the guide on how many questions to use to measure brand presence in AI models.

Does good SEO on Google mean the brand appears in ChatGPT?

No: position on Google and presence in ChatGPT are different numbers, and one does not replace the other. Position on Google measures where a page appears in a list of links. ChatGPT runs its own searches and often cites comparisons and pages from other sites in place of the company’s site; since August 2026, it cites fewer forums and news items and more official pages. That is why brand presence in ChatGPT is measured by asking it the questions, and not through the SEO report.

Can you identify the traffic that comes from AI Mode and AI Overview?

Partly, because the click from AI Overview and AI Mode, Google search’s conversational mode, arrives mixed in with organic search. In Google Analytics, the AI Assistant channel groups visits from ChatGPT and other AI models, without AI Overview and AI Mode. Search Console has a generative AI report that counts how many times the site’s pages appeared in AI Overview and AI Mode. The setup is in the guide on how to measure brand presence in ChatGPT.

Do consumers trust AI recommendations?

The buyer trusts AI, but checks the answer before deciding. TrustRadius surveyed 1,862 technology buyers from several countries in January 2026. Of these, 63% used AI in the buying journey, and 94% of those who used it check the answers at least sometimes. The company that ran the survey sells customer review services. For the brand, appearing in the answer opens the conversation, and what the buyer finds when they check influences the choice.

The next step to get out of the dark

murmur’s free diagnostic shows in less than two minutes, with no sign-up, what AI models answer today in your market. It is the first portrait of the brand in the answers, before any ongoing monitoring. murmur’s sector studies show which brands appear in AI answers in each market studied. Each study is the portrait of one round, with the sample published.

Read also

References