Updated on 7 November 2026, by João Leite, Founder of murmur.
Analyze brand presence in ChatGPT and the other AI models at least once a month, always with the same list of questions. Presence is how many answers the brand appears in, divided by the total number of answers. A round is each time the whole list is run again, and within it each question is asked more than once, on different days. The repetition exists because the answer changes from one conversation to another, and reading the result every day shows that back-and-forth, not a trend. Outside the calendar, an extra round goes in when the AI changes model or when the way of measuring changes.
- At least one round per month, with the same list, in the same AI models and with each question repeated on different days.
- Collecting every day is not reading every day: the decision comes from the sum of the answers for the period.
- Extra round when the model changes: in the switch to GPT-5.5, the distribution of sites cited by ChatGPT in German changed 47% in 48 hours (SISTRIX).
- After publishing, weeks: the first recorded citation of the FotoRegistro guide, a photo printing e-commerce, in ChatGPT came 30 days after launch.
- In writing: the frequency of the rounds, the date of each one and who reads the answers.
The frequency for measuring brand presence in each situation
In the routine, brand presence is measured at least once a month, with the same list of questions. Four other situations call for their own care. The round always goes through the same AI models, such as ChatGPT, Claude and Google AIO. Google AIO is the summary written by AI that appears above the links in many searches, and that Google calls AI Overview.
| situation | when to measure | what the round shows |
|---|---|---|
| before publishing any page | one full round, right at the start | where the brand stands today and how much the answer varies on its own |
| routine | at least one round per month, with the same list | whether the brand’s presence went up, went down or stayed the same |
| after publishing a page | the routine round continues, and the effect is read over weeks | whether the new page started bringing the brand into the answers |
| AI model change | an extra round, right after the change | whether the brand is still in the new model’s answers |
| change in the list, the tool or the account used for the questions | a round right after the change | the new starting point for comparisons |
Each month’s round is the basis for comparison, and the extra rounds protect that basis. They separate what changed in the AI or in the measurement from what changed in the brand’s presence. Without them, a drop caused by a new model looks like the content’s fault.
Collecting the answers more than once a month adds repetitions, as long as the reading brings together the answers for the period. Peec AI, a monitoring tool, collects the answers every day on the Starter, Pro and Advanced plans. naia, a Brazilian tool, sells four analyses per month, of 220 queries each, on the Autoridade plan. The comparison table on the same page says 20 per analysis.
murmur is a GEO agency in Brazil that puts brands into the answers of ChatGPT, Claude, Google AIO and other AI agents. It runs the same set of questions week after week. Every answer is stored, with the date.
Each question is repeated within the round, on different days
Each question is repeated within the round, on different days, because the same question asked again to ChatGPT can come back with other brands. In a SparkToro study, 600 volunteers asked the same questions 2,961 times, adding up ChatGPT, Claude and Google’s AI, in November and December 2025, and the chance of ChatGPT or Google’s AI repeating the same list of brands in two answers came out below 1 in 100. One answer shows that conversation, and several answers show how often the brand appears. The reason for the change, from the rewording of the question to the sources the search brings, is in the guide on how ChatGPT search works.
Otterly.AI, which sells AI monitoring, asked the same 520 questions every day to seven AI models, for 40 to 91 days, in the United States. Depending on the AI, only 12% to 19% of the brands mentioned for a question appeared on a single day in 30 days of monitoring. Among the cited sources, that share reached 67%: brands change less than sources, but they do change. For Otterly, the goal is to know the normal variation in your own data, so you can recognize when something really changes. The study measures only the change between days, and Otterly warns that the numbers serve as context, not as a universal standard.
murmur measured how much the brand’s result for each question repeats when nothing changes. Of the 387 questions in a Brazilian e-commerce category, 98% repeated the result when they were asked again three days later, with no intervention. In ChatGPT, 7 of 387 questions changed result; in Claude, 6 of 386. The rounds were on 30 June and 3 July 2026. Anyone who reads the number every day sees a few questions change on their own and can mistake chance for a trend.
The evolution of brand presence is compared on the same questions
The evolution of brand presence appears when two rounds ask the same questions, to the same AI models and in the same way. The reading goes question by question: the ones that started bringing the brand, the ones that stopped bringing it and the ones that stayed the same. The spreadsheet for this reading, with one column per AI, comes in the step-by-step guide on how to measure brand presence in ChatGPT.
Doing it the same way means each question goes in a new conversation, without the memory of previous ones, and is always asked from the same place, because location changes the answer. OpenAI writes that ChatGPT takes the approximate location from the IP address of the person asking and may pass it to partner search engines to improve the accuracy of results. That is why murmur does not measure from a team member’s computer: each round comes from a dedicated server in Brazil, which keeps the same IP address from one week to the next and uses the browser set to Portuguese. The city of the person taking the measurement no longer interferes with the answer, and this week’s round can be compared with last week’s.
Mindo, a motion design studio in São Paulo and a murmur client, was measured in two rounds less than three weeks apart. On the same 134 questions in its market, ChatGPT recommended Mindo in 12% of them in the first round and in 23% in the second. The rounds were on 25 and 26 June and on 13 and 14 July 2026. The comparison holds because the list of questions and the counting criterion were the same on both dates. A change of that size shows up in the comparison between rounds, without daily reading.
If the AI changes model, measure again before the month ends
When the AI changes model, do an extra round right after the switch, without waiting for the month’s round. The model is the version of the program behind ChatGPT, such as GPT-5.5. With a new model, the brand can gain or lose ground without anything having changed on its site. The extra round shows what happened before the change is attributed to the content.
SISTRIX, a German SEO and AI monitoring tools company, analyzed 38 daily samples of 100 thousand ChatGPT answers in German. On 23 May 2026, with the switch from GPT-5 mini to GPT-5.5, the distribution of cited sites changed 47% in 48 hours. On an ordinary weekday in May, the distribution of cited sites changed between 1% and 2%. After the switch, almost half of the citations started pointing to other sites.
SISTRIX warns that it measures a correlation: from the outside, you cannot separate the new model from changes in search or in the instructions. The study counts cited sites, not brands, and covers only answers in German. Every measurement holds for the model at the moment it was made.
The study by SparkToro, an American audience research company, on the brands AI models recommend used answers from November and December 2025. The text warns that newer models may vary more or less. In your measurement, the round right after the switch becomes the new starting point.
How long it takes for a new page to appear in ChatGPT
A new page usually takes weeks to show up in ChatGPT, so reading the effect waits while the routine round continues. In the first weeks, the lack of change may just mean the AI has not used the page yet. The timelines in detail are in the guide on how long it takes to appear in ChatGPT.
FotoRegistro, a photo printing e-commerce company and a murmur client, launched its guide on 24 June 2026. The first citation of that guide recorded in ChatGPT came on 24 July, 30 days later. On average across murmur’s clients, significant improvement came from 60 days onward, depending on the starting point and the competition.
Adapting the strategy at every round: what to keep, publish and fix
Every round ends in a decision that adapts the strategy: what to keep, which page to publish or update and which information to fix. The first questions to look at are the ones that stopped bringing the brand and the ones that started bringing a competitor. In those answers, the sources the AI cited point to the page that has yet to be published or the one that has gone out of date. What to do about the competitor appearing in place of the brand is in the guide on when the AI cites the competitor and not your company.
Wrong information about the company, such as an old price or a service it does not offer, is fixed before any new page. As long as the sources the AI uses carry the old information, the error comes back in the following rounds. The step-by-step is in the guide on how to fix wrong information ChatGPT says about your company.
A change that appears in one round and disappears in the next may be fluctuation, and goes under observation before it becomes a new page. One that holds across the following rounds calls for action. A new buyer question goes on the list with its date and only counts in the comparison from its second round onward.
What to ask for in writing from the people who monitor the brand’s presence
Ask in writing how often the same list of questions runs and the date of each round. Also ask who reads the answers and where you see the result. Without the date of each round, a rise or fall from one month to the next is left unexplained.
Each vendor counts a round its own way, so ask what goes into each one. naia, which sells monitoring, writes on the page where it compares tools that “Prompts diários e consultas por análise não são a mesma unidade” (“Daily prompts and queries per analysis are not the same unit”). Prompt is the name the market gives to each question asked of the AI. Before the meeting with each vendor, bring the full list from the guide on what to ask before hiring a GEO agency.
At murmur, the specialist reads each week’s answers and the measurement sets the next step
At murmur, the specialist dedicated to the brand reads each week’s answers, and the measurement sets the next step of the work. It is the same person from the diagnostic onward, who studies the company and the sector and monitors the brand’s presence closely. The same list runs week after week. A competitor that rises on a question, a seasonal campaign or a topic that has not taken off yet changes the next week’s work queue.
murmur can respond with a new page for a sector question or with the revision of a page that lost ground. If the round catches the AI getting something wrong about the company, murmur helps fix the error and put the brand in the right place. The client sees the result in the dashboard included in the service, where the brand’s presence appears side by side with the competitors’.
Frequently asked questions
Which part of AI citation monitoring can be automated?
Collection is automated with a monitoring tool, which runs the questions, stores the answers and counts the mentions. The part that calls for a person is reading the total for the period and a sample of the answers. Automated counting goes wrong with a company that has a similar name and with the brand cited only as a source.
In Peec AI, each question generates one answer per AI, per day: 25 questions across 3 AI models for 30 days gives 2,250 answers in the month. That volume is enough to read the month, not to react to every day. In the Otterly.AI study, ChatGPT reused only 25.8% to 28.9% of sources from one day to the next, depending on the question category.
How do you run an AI presence audit?
Ask the buyer’s questions to ChatGPT, Claude and Google AIO, and note in which answers the brand appears. An audit done once shows where the brand stands today, and the same list repeated every month shows presence over time. For the audit to be valid, the list cannot be short: in the Otterly.AI study, the variation in measured presence was about 73% smaller with 100 questions than with 10. After 2,961 answers, Rand Fishkin, of SparkToro, accepted the percentage of visibility across dozens to hundreds of repeated questions as a reasonable metric.
Which AI models should be monitored, and how?
In every round, monitor the AI models your buyer uses, starting with ChatGPT, Claude and Google AIO. Each AI model changes at its own pace, and measuring one does not give a picture of the others. SISTRIX tracked 82,619 questions in six countries, with a snapshot per week, from December 2025 to April 2026.
In the SISTRIX study, 53% of AI Overview questions did not change a single source in 17 weeks. In AI Mode, Google’s search mode in which the person converses with the AI, sources changed 56% per week. In ChatGPT Search, source turnover was 74% per week in Germany, 60% in the United Kingdom and 42% in France. The ChatGPT data counts only the questions in which it cited sources on a regular basis. The study counts the sources cited, not the brands.
How do you track the brand’s presence every month in a spreadsheet?
Build a spreadsheet with one row per answer: the date of the round, the AI, the question and whether the brand appeared. At the end of the month, those rows give the brand’s presence in each AI. That is the KPI, the number the company chooses to track to know whether the work is delivering results. The spreadsheet columns, one by one, are in the step-by-step guide on how to measure the brand’s presence in ChatGPT. To choose the questions on the list, use the guide how to build the list of questions to measure the brand’s presence in ChatGPT.
Where to start
The starting point of any frequency is knowing where the brand stands today. That first reading comes from murmur’s free diagnostic, which does not ask for registration. In less than two minutes, it shows what the AI models answer in your market. After that, rounds with the same questions show the brand’s presence over the months. murmur’s sector studies bring, market by market, the snapshot of a single round of the brands recommended by the AI models.
Read also
- What to ask before hiring a GEO agency
- How many questions to use to measure brand presence in AI models: from dozens to hundreds
- How long it takes to show up in ChatGPT: from first citation to presence in answers
- How to build the question list to measure brand presence in ChatGPT
- Brand presence in AI models: the reliable metric and how to check any number
References
- Otterly.AI, Thaylise Nakamoto, “We Measured How Stable AI Citations and Brand Mentions Really Are (and Why 10 Prompts Isn’t Enough)”, updated on 11 September 2026. https://otterly.ai/blog/ai-search-visibility-stability/ (accessed on 30 September 2026).
- SISTRIX, Johannes Beus, “ChatGPT Core Updates: Chatbots Have Algorithm Changes Too”, 1 June 2026. https://www.sistrix.com/blog/chatgpt-core-update/ (accessed on 30 September 2026). The original in German, from 29 May 2026 and updated on 21 August 2026, is at https://www.sistrix.de/news/chatgpt-core-update/.
- SISTRIX, Johannes Beus, “AI Citation drift: How stable are sources in AI search results?”, 1 May 2026, modified on 12 May 2026. https://www.sistrix.com/blog/ai-citation-drift-how-stable-are-sources-in-ai-search-results/ (accessed on 30 September 2026). The original in German, from 28 April 2026 and updated on 21 August 2026, is at https://www.sistrix.de/news/ai-citation-drift-wie-stabil-sind-quellen-in-ai-suchergebnissen/.
- 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, produced with Patrick O’Donnell, of Gumshoe.ai, which sells brand monitoring in AI. 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 30 September 2026).
- OpenAI, “Searching the web with ChatGPT”, OpenAI Help Center, no exact date on the page: “ChatGPT also collects general location information based on your IP address and may share that general location with third-party search providers to improve the accuracy of your results.” https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt (accessed on 2 October 2026).
- Peec AI, “Pricing for Peec AI”. https://peec.ai/pricing (accessed on 30 September 2026).
- naia, “Planos naia: Autoridade e Enterprise para visibilidade em IA”. https://naia.today/pricing (accessed on 30 September 2026).
- naia, “Preços de ferramentas GEO: planos mensais e anuais em 2026”, editorial research verified on 26 September 2026. https://naia.today/ferramentas-geo/precos (accessed on 30 September 2026).
- murmur, home page, frequently asked questions and ongoing monitoring: https://murmur.marketing/en/; Mindo and FotoRegistro case studies, published with the clients’ authorization: https://murmur.marketing/en/cases/mindo/ and https://murmur.marketing/en/cases/fotoregistro/; free diagnostic and sector studies: https://murmur.marketing/en/diagnostic/ and https://murmur.marketing/estudos/ (accessed on 30 September 2026).