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

Monitor from dozens to hundreds of questions (prompts) to measure brand presence in ChatGPT and the other AI models: ten is not enough. A prompt is the question or request a person writes to the AI. Presence is how many answers the brand appears in, divided by the total number of answers. In a short list, each question weighs heavily in that calculation, and swapping one question for another already changes the result. The size of each list comes from the questions buyers ask in your market, not from a fixed number.

  • Dozens to hundreds of questions, repeated: this is the range in which Rand Fishkin, of SparkToro, came to accept presence as a reasonable metric.
  • Presence fluctuates less with a larger list: in the Otterly.AI test, the range fell from 32 to 9 points between 10 and 100 questions.
  • With one answer per question, each answer is worth 100 divided by the size of the list: 10 points in a list of 10, 1 in a list of 100.
  • The right number changes from brand to brand: Otterly.AI warns that 100 questions are not the right number for every brand.
  • A prompt is not the same unit in every plan: compare distinct questions, AI models and times per month.

How many prompts to monitor: from dozens to hundreds of questions

The number of prompts to monitor ranges from dozens to hundreds of distinct questions. In the size of the list, each question counts once, however often it is repeated. Rand Fishkin, cofounder of SparkToro, an American audience research company, reached the same range in a January 2026 study. He considers presence measured across dozens to hundreds of questions, each asked several times, to be reasonable. For him, the result comes out scattered like darts on a target, but it already allows the brand to be compared with the others.

murmur is a GEO agency in Brazil that places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents. It publishes the size of the list in each case study. Mindo, a motion design studio in São Paulo, was measured in ChatGPT on the same 134 questions in June and July 2026. Visio, operations software for retail, had the same 135 questions in both months, in ChatGPT, Claude and Google AIO. Google AIO is the short name for AI Overview, the AI-written summary Google puts before the links. These are lists in the hundreds, built from the questions of each client’s market.

Ten questions are not enough: brand presence changes with the list

Ten questions are not enough because, in a short list, brand presence changes depending on which questions made it in, even with nothing changing in the AI models. Otterly.AI, an AI monitoring tool that charges by number of questions, measured this effect in a study published in September 2026. The test started from 320 questions in the AI search monitoring sector, asked once a day to seven AI models, in the United States. Otterly calculated the presence of the same brand many times across random groups drawn from those 320, without changing the AI models or the 30 days.

In the Otterly.AI test, presence is measured as brand coverage, which counts the questions in which the brand is mentioned. In groups of ten questions drawn from the 320, it was between 39% and 71% in 90% of the draws, a range of 32 points. With 50 questions, the range narrowed to 13 points, from 49% to 62%, and with 100, to 9 points, from 51% to 60%. Across the set of 320 questions, the variation in presence fell about 73% between the list of 10 and the list of 100, according to Otterly. The biggest gains came from the first questions added: from 10 to 50, the range lost 19 points, and from 50 to 100, only 4.

The range of 9 points with 100 questions does not make 100 the right number for every brand, Otterly writes. The company recommends that each brand measure its own starting point in each AI model and in each category. With ten questions, two different lists can give the same brand, in the same month, presences more than 30 points apart. A number like that does not say whether the brand improved, got worse or simply changed lists.

How much each answer weighs in brand presence, depending on the size of the list

In a measurement with one answer per question, each answer is worth 100 divided by the number of questions, in points of brand presence. That is 10 points in a list of 10 questions, 5 in a list of 20, 2 in a list of 50 and 1 in a list of 100. In a list of 10, a single answer that cites the brand by chance raises presence by 10 points. In a list of 100, the same increase calls for 10 more answers. The calculation, done AI model by AI model in a spreadsheet, is in the step-by-step guide to how to measure brand presence in ChatGPT.

The size of the list comes from the questions the buyer asks in your market

The size of the list comes from the questions the buyer asks in your market, in each business line and at each stage of the purchase. A company with one product and one type of customer covers the market with fewer questions than a store with five product lines. Ten ways of asking the same question increase the list without covering more of the market. The final number of the list balances two forces: fluctuation, which falls as the list grows, and coverage of the buyer’s real doubts.

Repeating the question and adding questions measure different things

Repetition shows whether the brand comes back on the same question, and the larger list shows whether it appears in the rest of the market. A question asked ten times still counts as one in the size of the list. Reliable measurement of brand presence combines the two things: the list sized to the market and each question repeated on different days.

At Peec AI, a monitoring tool, the answers of the month are the questions tracked, times the active AI models, times the frequency. In the entry plan, with 50 questions across 3 AI models every day, that gives 4,500 answers in a 30-day month. That is a lot of answers about the same 50 questions. A round is each time the whole list is redone, with each question in a new conversation, without the memory of the previous ones. The same question changes its answer from one time to the next, as SparkToro showed in ChatGPT, Claude and Google’s AI, and that is why presence is measured by repeating the same questions over time. The guide on how often to measure brand presence in AI models shows how to repeat each question and how to compare two rounds.

What each tool plan calls a prompt

Each monitoring tool counts prompts its own way, and the number in a plan can only be compared when the unit is the same. In October 2026, Otterly.AI sells plans of 15, 100 and 400 questions, asked once a day across four AI models. Its Enterprise plan starts at 1,000 questions. Peec AI sells 50, 150 and 350 questions, also every day, across 3 AI models of your choice.

naia, a Brazilian tool that also sells monitoring, counts the Autoridade plan in analyses: 4 per month, for 1 brand. On the same page, the plan card says “220 prompts por análise” (“220 prompts per analysis”), and the comparison table says “20 por análise” (“20 per analysis”) in the prompts row. On a page that compares tools, including its own, naia asks for the full unit: distinct questions, AI models per question, runs per month and resulting answers. With those four numbers written down, two plans become comparable, whether from a tool or from an agency. What else to ask before signing is in the guide on what to ask before hiring a GEO agency to appear in ChatGPT.

At murmur, the size of the list does not come from a plan, and there is no cap on questions. murmur’s monthly fee depends on the business lines, the brands, the complexity of the sector and the languages. A second language, for example, brings into the measurement the questions asked in that language.

Questions with the brand name are counted separately

Questions that include your brand name sit in a separate account, and the range of dozens to hundreds does not include them. In a plan with a fixed number of questions, each question with the brand name takes the slot of a market question. In a list of 100 with 20 brand questions, market presence comes from 80 questions. Questions in another language or from another country also form a separate list, with their own presence. The guide on the reliable metric for brand presence in AI models explains why a question with the brand inflates the number.

At murmur, the list comes from the sector study and repeats every week

At murmur, the list of questions comes from the client’s sector study, and the same set runs week after week. The study ranges from the questions of those just starting to research to those already choosing a supplier, in each business line. The size of the list is what that study finds in each client’s market. At FotoRegistro, a photo printing e-commerce, the ChatGPT measurement used the same 361 questions on both dates. The dates were 20 July and 1 September 2026. At Fly Vet, a veterinary marketing agency, the ChatGPT measurement on 2 September 2026 had 228 questions.

Where the person asking is located also changes the answer. In OpenAI’s explanation of ChatGPT search, the company says it estimates each user’s approximate location from their internet address and may share it with partner search engines, and gives the reason “to improve the accuracy of your results”, which in Portuguese means making the results more accurate. murmur asks the list’s questions on a dedicated server located in Brazil, not on a team member’s computer, with an internet address that does not change and the browser set to Portuguese. As a result, the city of the person doing the measurement does not interfere with the answers, and the same list generates numbers that can be compared from one week to the next.

murmur’s list of questions also guides content production, based on the questions it identified in the sector. Each answer is stored with its date. In the dashboard included in the service, the client sees each question on the list, one by one, and how many competitors compete for that answer.

Frequently asked questions

How do I choose the questions that go on the list?

Choose the questions your buyer has before buying, written the way they write and without your brand name. Cover each business line and each stage, from the problem to the comparison of suppliers. To find these questions and note each one down, see how to build the list of questions to measure brand presence in ChatGPT.

Can you trust a percentage of brand presence?

You can, when the percentage comes with the number of questions, the AI models, the dates and who measured. A presence taken from ten questions changes a lot just by swapping the questions, and one taken from a hundred changes much less. What to ask of any number is in the guide on the reliable metric for brand presence in AI models.

Does the same list work for ChatGPT, Claude and Google AIO?

It does, and each AI model needs the whole list: dividing 100 questions among three AI models leaves each with a little over 30. Each AI model has its own presence, calculated over the answers to the whole list, in its own column of the spreadsheet. Adding the AI models into a single number hides the one where the brand is missing.

Does repeating the same question replace having more questions?

No, because repeating does not put a new question on the list, and the result stays tied to the questions chosen. Repeating on different days is covered in the guide on how often to measure brand presence in AI models.

Can you know what your buyer actually asks the AI?

You can get to know their questions, and the list needs them: the exact words change from person to person. In a SparkToro test, among 142 requests written by people about headphones for a relative who travels, barely two were alike. Even so, across the 994 answers, brands such as Bose, Sony, Sennheiser and Apple appeared 55% to 77% of the time. The AI models recognized the same intent behind different texts.

Where to start the list of questions

Start with what AI models answer today when someone looks for what you sell. With no sign-up and in under two minutes, the free murmur diagnostic shows those answers, as a first snapshot before the whole list exists. To see this kind of snapshot in other markets, see murmur’s sector studies. Each one gives the number of questions, the AI models and the month, and covers a single round.

Read also

References