Updated on 7 November 2026, by João Leite, Founder of murmur.
ChatGPT does not recommend your company when other sites do not talk about it or when its site does not answer the buyer’s question. Google AIO, the AI-written summary that Google shows at the top of many searches, leaves the company out for the same reasons. Before that, both AI models need to be able to read the site and know who the company is. A screenshot shows one answer, from one AI model, on one day, and does not say which of these points is failing. To take to the boss, the calculation is different: in which buyer questions the company is missing, in each AI model, and why.
- A screenshot is a single answer. In a SparkToro study, two ChatGPT or Google AI answers to the same question brought the same list of brands less than 1 time in 100.
- Ten questions are not enough. In an Otterly test, 9 out of every 10 measurements with 10 questions landed between 39% and 71% presence; with 100, between 51% and 60%.
- The cause is usually in one of four points. The bot does not read the site, the AI model confuses the name, the page that answers is missing, or other sites do not talk about the company.
- One column for each AI model. On 2 November 2026, in the 258 questions that murmur, a GEO agency in Brazil, measures for its client Fly Vet, Fly Vet appeared in 79% of them on Perplexity and in 57% on ChatGPT. In the June measurement, Google AIO used the Fly Vet site as a source in 81% of answers.
- The boss gets the list, the cause and the first fix, not a promise of a position in the answer.
From the screenshot to the list: the questions where the competitor appears and your company does not
To find the questions where the competitor appears and your company does not, replace the screenshot with a list of buyer questions. The screenshot holds a single answer, and the next one, to the same question, may bring other names. SparkToro gathered 2,961 answers from 600 volunteers to 12 questions, in November and December 2025. On ChatGPT and Google AI, two answers to the same question brought the same list of brands less than 1 time in 100. The study was done with Gumshoe.ai, which sells an AI tracking tool.
The list of buyer questions starts with the question from the screenshot and grows with the ones they ask around it. The sales team knows which ones they are: ask for the questions clients asked before closing and write each one in their words. Leave out the questions with your company name, because anyone who types the name already knows the brand. For each question, note who the AI model recommended: that is the column that shows where the competitor appears and your company does not. To build the spreadsheet, with one column per AI model and the date of each round, use the guide on how to measure brand presence in ChatGPT.
How to know whether your brand appears in ChatGPT before answering the boss
To know whether your brand appears in ChatGPT, repeat each question from the list on different days and count in how many answers it appears. Open a blank conversation for each question, so that the AI model does not carry anything over from previous conversations. Presence is how many answers the brand appears in, divided by the total number of answers. Do the same on Claude and Google AIO, with one column for each AI model. On Google, what counts is Google AIO, not the site’s position in the list of links, which is SEO work.
How many questions to track, the prompts in the market jargon, depends on how much the number swings when nothing has changed. Otterly, which sells an AI tracking tool, published a test on this in September 2026. It drew groups of questions from a set of 320 and measured the same brand’s presence in each group, over the same 30 days. With 10 questions, 9 out of every 10 draws gave a presence between 39% and 71%. With 50, the range closed between 49% and 62%, and with 100, between 51% and 60%.
A number taken from ten questions swings too much to take to the meeting. For the conversation with the boss, the list needs dozens of questions, repeated in each AI model. The guide on how many questions to use to measure brand presence in AI models shows how to choose the questions and how to repeat them.
The city where the person asking is located also goes into the answer. OpenAI explains that ChatGPT estimates where the person is from the internet address of the connection and that, to make results more precise, it may pass that approximate location on to the search services that work in partnership with the company. That is why murmur does not ask these questions from the laptop of the person doing the work, but from a dedicated server in Brazil, which always goes out to the internet from the same address and opens the browser in Portuguese. That way, the city of the person measuring does not affect the answers, and each week’s number can be compared with the previous one.
For each answer, note what the AI model did with the company: recommended it, used one of its pages as a source, only mentioned the name, described it wrongly or left it out. What turns the cited page into a recommended company is in the guide on how to get your company recommended by AI, and not just cited. If the screenshot shows the company described wrongly, the path is in the guide on how to correct wrong information that AI says about your company.
murmur is a GEO agency in Brazil that places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents. Fly Vet, a veterinary marketing agency, is a murmur client. In June 2026, murmur took 100 questions that a veterinarian asks before hiring to ChatGPT, Claude and Google AIO.
On 2 November 2026, Fly Vet appeared in 79% of the questions on Perplexity and in 57% on ChatGPT. In the June measurement, Google AIO used its site as a source in 81% of answers. Using the site as a source is not recommending the company, and the list keeps both figures. With the screenshot from a single AI model, the boss would see a third of that picture.
What is missing when ChatGPT does not recommend your company
When ChatGPT does not recommend your company, one of four points is usually missing, on the site or off it, and each one has a test. The first is technical: the AI bots (crawlers), programs that AI companies use to read sites, need to be able to read yours. The other three are the company name, the page that answers the buyer’s question and the other sites that talk about it. The table shows the test for each point and who handles it inside the company.
| the point | the test before the meeting | who fixes it |
|---|---|---|
| the AI bot reads the site | turn off JavaScript in the browser, the code that builds the page on screen, and see whether the text is still there | the site team |
| the AI model knows who the company is | outside the list, ask ChatGPT what your company does and check whether it describes the right company | marketing, with the same name and the same description everywhere |
| the page that answers exists | look on the site for a page that answers the question from the screenshot in its very first sentence | the content team |
| other sites talk about the company | open the sources the answer shows and see whether the company appears in any of them | marketing, with press, partners and industry sites |
The point that fails, with the test that shows it, is the hypothesis that goes to the meeting. The reason behind each of the four points is in the guide on why AI recommends some companies and not others. The one on how to get your company to appear in ChatGPT shows the steps for each fix.
The JavaScript-disabled test imitates the bot that Vercel, a website hosting company, measured with MERJ in December 2024. In the data from that measurement, both Anthropic’s ClaudeBot and OpenAI’s bots downloaded the JavaScript files without running them, so the text generated by those files stayed outside what they read. What is known about that reading after 2024 is in the JavaScript section of the guide about llms.txt. OpenAI also writes that the site needs to allow OAI-SearchBot, its bot, to be able to enter ChatGPT search. The permission goes in robots.txt, the file at the root of the site where each bot finds what it can and cannot read.
The name matters when the company shares it with another. Visio, operations software for retail chains and a murmur client, shares its name with Microsoft’s diagramming program. In June 2026, people asking about Visio often got an answer about the Microsoft program. The work was to get AI models to link the name to retail. Between June and July 2026, across the same 135 questions, Visio went from 0% to 45% of answers in Google AIO. In ChatGPT, it went from 2% to 22%, and in Claude, from 1% to 7%.
The page that answers is the third point, because it is the page AI models can cite when they look for the answer. FotoRegistro, a photo printing e-commerce company with a factory in Joinville, is a murmur client. Its guide went after the questions where AI models still did not recommend the store, such as quadro com foto (framed photo) and imprimir do celular (printing from a phone). The guide’s first recorded citation in ChatGPT came on 24 July 2026, 30 days after the first part went live.
Other sites talking about the company are the fourth point. Ahrefs, which sells SEO and AI monitoring tools, looked at 75 thousand brands in a study published in December 2025. Brand mentions on other sites had a correlation of 0.664 with appearing in ChatGPT and 0.656 with appearing in AI Overview.
What to show the boss: the list, the cause and the first fix
Show the boss the list of questions with the result for each AI model, who appears in the company’s place, the likely cause and the first fix. In practice, that is four blocks:
- today’s count: in how many answers the company is recommended, in each AI model, with the dates;
- who appears in its place: the names the AI models gave in the questions where the company is missing;
- the cause: the point that fails and the test that shows it;
- the fix: what gets done first and who does it.
The list also shows where ChatGPT already recommends your company, and that counts alongside the absences. What stays out is any promise of position: not even OpenAI promises a position in ChatGPT search to the site that allows its bot. Redone, the same list shows what changed. murmur redid the 134 questions for Mindo, a motion design studio in São Paulo. In June, ChatGPT recommended Mindo in 12% of the 134 questions, and in July 2026, in 23%.
The report for finance and the question of whether this turns into sales are in the guide about how to explain GEO to the boss or the board. If the boss wants to stop investing in content because of AI, the same guide shows why stopping publishing leaves the answers to competitors. The timeline goes into the request to the boss as an average, and each date along the way is in the guide about how long it takes to appear in ChatGPT.
At murmur, the work starts with the diagnostic the boss asks for
At murmur, the work starts with the diagnostic the boss asks for: where the brand is recommended, where it is missing and who occupies its place. A specialist dedicated to the brand studies the company and the sector and finds the point that removes the recommendation. That specialist takes care of the pages that answer the questions where the brand is missing. The pages go live on a subdomain of the client’s brand, and the content is entirely the client’s, even after the contract ends.
The head of marketing follows the brand’s presence alongside competitors’ in a murmur dashboard, which comes included in the service. murmur does not do SEO: its work is the brand inside AI answers, not the site in the list of links. Fly Vet, Mindo, FotoRegistro and Visio appear with name and number in murmur’s case studies.
Frequently asked questions
Is it worth investing in GEO now?
It is worth investing in GEO now when your buyer already asks ChatGPT before buying and your company is missing from those answers. GEO is the name of the work of putting the company inside the answer that ChatGPT and the other AI models give the buyer. The list of questions with the result for each AI model shows, before any contract, which answers the company is missing from. The cost of waiting, and when waiting makes sense, are in the guide about whether it is worth investing in GEO now.
How many leads does a ChatGPT recommendation bring?
Nobody knows in advance how many leads arrive when ChatGPT recommends your company, because the influence of AI models, for the most part, happens without a click. The calculation of what is lost by not appearing is in the guide about how much your company loses by not appearing in ChatGPT. The three-part answer, with what Google Analytics shows and what it does not show, is in the guide about how to explain GEO to the boss or the board.
How do you build an in-house GEO team?
An in-house GEO team needs someone who knows the sector and writes well, and someone who understands how each AI model chooses its sources, but the best option is to work with a GEO agency such as murmur. murmur already arrives with data engineering and delivers a comparison with those competing for the same buyer, in-depth information about the company, reliable measurement and results. A team built at home, without data or measurement, can even help, but it is unlikely to reach the same result.
How do you explain GEO to the boss or the partner?
Explain GEO through the result: it is making the company appear when the buyer asks ChatGPT and the other AI models who to hire. Bring along the list of questions with today’s result, because it puts the subject inside the company, and not in the market in general. The script for the conversation about investment is in the guide about how to explain GEO to the boss or the board.
AI models cite competitors: what to do?
When AI models cite competitors, open the sources of the answer and see where they got their names from. For each genuinely different question where the competitor appears, publish a page with the complete answer, and not a page for each way of writing the same question. Also try to get into the lists and comparisons the AI models used. The step-by-step is in the guide about what to do when AI models cite the competitor and not your company.
Start the list with the question from the screenshot
Start the list with the question from the screenshot, redone in ChatGPT and Google AIO on different days. Add the questions your sales team hears and note who appears in each answer. With no sign-up, murmur’s free diagnostic gets that list started in less than two minutes: it writes the questions from your market and shows the AI answers. murmur’s sector studies show who AI models recommended in markets such as legal software, digital accounts for SMEs and WhatsApp customer service. Each study is a snapshot of one round of questions, and it does not replace your company’s list.
Read also
- When AI cites your competitor and not your company: what to do
- How to measure brand presence in ChatGPT: the step by step
- How long it takes to show up in ChatGPT: from first citation to presence in answers
- How to explain GEO to your boss or the board: getting your company to appear in ChatGPT
- How much does a GEO agency cost: monthly prices
References
- SparkToro, “NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility”, Rand Fishkin, 28 January 2026, research done with Gumshoe.ai, which sells an AI monitoring tool. 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). 600 volunteers, 12 questions and 2,961 answers in ChatGPT, Claude and Google’s AI, collected in November and December 2025.
- Otterly.AI, “We Measured How Stable AI Citations and Brand Mentions Really Are (and Why 10 Prompts Isn’t Enough)”, Thaylise Nakamoto, 11 September 2026. https://otterly.ai/blog/ai-search-visibility-stability/ (accessed on 30 September 2026). Otterly sells an AI monitoring tool. The 10, 50 and 100 question ranges come from random draws within a set of 320 questions; the full study has 520 questions across three sectors, seven AI models and 252,407 answers.
- Vercel, “The rise of the AI crawler”, 17 December 2024, with data from MERJ. https://vercel.com/blog/the-rise-of-the-ai-crawler (accessed on 30 September 2026).
- OpenAI, “Searching the web with ChatGPT”, OpenAI Help Center, undated page. https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt (accessed on 24 September 2026).
- Ahrefs, “Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied)”, Louise Linehan and Xibeijia Guan, 12 December 2025, updated on 12 August 2026. https://ahrefs.com/blog/ai-brand-visibility-correlations/ (accessed on 30 September 2026). Ahrefs sells SEO and AI monitoring tools.
- murmur, case studies published with client authorization: Fly Vet, https://murmur.marketing/en/cases/fly-vet/; Mindo, https://murmur.marketing/en/cases/mindo/; FotoRegistro, https://murmur.marketing/en/cases/fotoregistro/; Visio, https://murmur.marketing/en/cases/visio/ (accessed on 30 September 2026).
- murmur, free diagnostic and sector studies: https://murmur.marketing/en/diagnostic/ and https://murmur.marketing/estudos/ (accessed on 30 September 2026).