Updated on 7 November 2026, by João Leite, Founder of murmur.

To make your startup appear in ChatGPT, start with the narrow questions in your niche, where the AI answer does not yet have an owner. murmur, a GEO agency in Brazil that puts brands into the answers of ChatGPT, Claude, Google AIO and other AI agents, compared the generic question with the narrow one in its sector studies. In the software house study, from September 2026, the most cited company in the generic question appeared in all 15 answers. In the question that stated the client’s segment, such as fintech or agribusiness, the most cited one came in at 14% of 21 answers. With the questions chosen, the startup publishes one page for each one and repeats the measurement every month.

  • In the narrow question, companies that the generic one did not cite came in. In murmur’s software house study, from September 2026, the most cited one in the generic question appeared in 15 of 15 answers, and in the question by client segment the first place came in at 14% of 21. The study does not say the size of the companies.
  • The rarely cited companies came in when the question narrowed. In the international remittance study, 29 of the 52 companies cited appeared in three answers or fewer.
  • In payroll and cloud telephony, the question that states the profile of the buyer brought in the Brazilian company. In murmur’s studies from September 2026, it appeared in 95% and 84% of those answers, and in 13% and 21% in the question from someone who wants to understand the subject.
  • In an Ahrefs experiment, the company’s own page mattered for a new name. Of the 72 new entries for a new event, 82% came from answers that cited Ahrefs’ own pages.
  • The measurement uses the same questions, every month. The questions from people who want to understand the subject are counted separately from the recommendation ones.

GEO is the work of making a company appear in the answers of AI models such as ChatGPT, Claude and Google AIO. For a startup, the work starts with three disadvantages: the AI knows little about the brand, the site has little text and the leader already occupies the generic question. The step-by-step guide for those who sell to other companies, with a long cycle and a buying committee, is in how to do GEO for a B2B company.

What changes for a startup to appear in ChatGPT: new brand and little content

For a startup to appear in ChatGPT, what changes is the starting point: the AI knows little about the brand, and few sites talk about it. ChatGPT searches the web before answering purchase questions, and the startup’s page enters that search like that of any company. What the model learned in training only changes when a new version comes out, so search is the door for the new brand, explained in how search works in ChatGPT.

In the Ahrefs study published in December 2025, the name cited on other sites rose along with the brand appearing in AI answers. The sample has 75 thousand brands from domains with DR above 40, Ahrefs’ measure of link strength. For the brand with little search, few links and few mentions, it writes that ChatGPT may be the best entry point:

For a brand with modest search volume, backlinks, and web mentions, ChatGPT may be the best entry point into AI visibility, since it appears to be less heavily gated by traditional SEO authority metrics.

The startup’s name needs to be recognized by the AI without confusion with another company. A short name, a common word or a name similar to another brand’s can confuse ChatGPT and Claude, and the startup has little text on the web to undo the mistake. In murmur’s software house study, ChatGPT wrote 5 names that did not match any software company in the description, and one of them was from environmental engineering. The fix is in how to keep the AI from confusing your company with another.

The narrow question brought in companies that the generic one did not cite, in the sectors murmur measured

In murmur’s studies, companies that the generic question did not cite came in on the narrow question, which states for whom, for what or in what situation. The studies do not say their size, and none compares the small company with the large one. In murmur’s sector studies, from September 2026, each one asked 25 questions without a company name, once in each AI model. The sample of each study is small, 25 questions per sector, and the sectors are niche: each number holds for the sector in which it was measured. The generic question kept the most cited, and the narrow one brought in other names in software house, international remittance and recurring billing.

In the software house study, with 74 answers from ChatGPT, Claude and Google AIO, the most cited one in the generic question appeared in the 15 answers of that type. In the 21 answers that stated the client’s segment, such as agribusiness, logistics or startup, the top-ranked one was at 14%. The two most cited in the generic question did not appear in that group, and the top spots went to companies that the generic question did not cite.

In the international remittance study, with 100 answers from ChatGPT, Claude, Perplexity and Google AIO, the AI models cited 52 companies. Of them, 29 appeared in three answers or fewer and, added together, did not reach 10% of the mentions. Those companies came in when the question narrowed: the freelancer who receives from a client abroad, the person paying their child’s college, the person sending money to family in Portugal. In the open question, almost none appeared.

In the recurring billing study, with 99 answers from the same four AI models, three platforms appeared in 85% of the 20 answers to the generic question. In the 28 answers that stated the client’s niche, such as a school or an internet provider, the top-ranked one was at 75%, and the other two fell to 50% and 46%. Names that the generic question did not bring in entered the list. For the fintech, what changes in the regulated sector is in how to do GEO for fintechs and banks. For the edtech and the course that just opened, the breakdown by area and format, in distance learning postgraduate programs, is in how to do GEO for colleges and courses.

How to choose the questions when the generic question belongs to the leader

Choose the questions the buyer asks when they already know what they need: who it is for, what for and in what situation. Each one becomes a page. The generic question goes on the list only to keep track of the leader, and it is not where the startup starts. The sales team and support hear these questions every week and are the first source for the list.

At Mindo, a motion design studio in São Paulo and a murmur client, ChatGPT went from 12% recommendation, at the end of June 2026, to 23%, in mid-July, on the same 134 questions in its market. At Visio, operations software for retail that is not Microsoft Visio, and also a murmur client, the questions that unlocked things were those from people who did not yet know the brand.

In the card machines study, the question that asked for an alternative did little to open up the list: the 24 general recommendation answers brought 10 brands, and the 20 that asked for an alternative brought 14. The 4 brands that only appeared there came in one or two answers each, and the big names rose in the same group. For the software startup, the comparison and the list of alternatives are in how to appear in ChatGPT in software comparisons.

The question that states the profile of the buyer also does not open the door in every sector. In the performance management study, some Brazilian platform appeared in 93% of the 15 generic answers and in 52% of the 21 that stated who buys. In that group, the full HR suites moved ahead. Before writing, ask the questions on your list to ChatGPT, Claude, Perplexity and Google AIO and note who appears.

In payroll and cloud telephony, the profile of the buyer brought in the Brazilian company

In payroll and in cloud telephony, the question that described the profile of the buyer brought up the Brazilian company. In murmur’s September 2026 studies, some Brazilian company appeared in 95% and in 84% of those answers, and in 13% and 21% in the question from someone who wants to understand the subject, such as what it is or how much it costs. In managed cloud, it was at 33% in the profile question and at 7% in the one from someone who wants to understand. The three studies asked the questions to ChatGPT, Claude and Google AIO.

sector, with the murmur studygeneral recommendation question: any Brazilian companyquestion that describes the profile of the buyer: any Brazilian companyquestion from someone who wants to understand the subject: any Brazilian company
payroll system100% of 15 answers95% of 21 answers13% of 15 answers
cloud telephony and dialer69% of 13 answers84% of 19 answers21% of 14 answers
managed cloud for companies67% of 15 answers33% of 21 answers7% of 15 answers, and a foreign cloud in 87%

In the open-ended question, the general recommendation one, the Brazilian company appeared in 100% of the payroll answers, in 69% of the cloud telephony ones and in 67% of the managed cloud ones. In managed cloud, one of the three big foreign public clouds appeared in 87% of the 15 answers from people who wanted to understand the subject.

The page that explains the subject serves the buyer who is just starting and can become a source for the answer, even when the answer does not cite any company. In these three studies, the Brazilian company appeared more in the question that describes the profile of the buyer than in the one from someone who wants to understand the subject, and that is why the measurement counts the questions from people who want to understand in a separate count, so as not to confuse being cited as a source with being recommended. The sector-by-sector calculation, in the open-ended question and in the one that describes the profile, is in how many Brazilian companies appear in AI answers.

How much content to publish to start: one page per question

To start, the startup needs one page for each question chosen. Only the question that changes what the buyer wants to know counts as a new question: each of those gets a page with the full answer, and the various ways of typing the same question stay together on a single page. Each page answers in the first sentence, with the buyer’s words. Then come the facts the AI uses to recommend: who the product serves, how much it costs or what the price depends on, what is included and how it compares.

In a case with a new name, the company’s own page carried weight. Ahrefs, which sells SEO tools, measured this with Evolve, a new event of its own, in 9,886 answers from four AI models, from February to May 2026. Evolve entered 72 question and AI combinations in which it did not appear before, and in 82% of them the answer cited a page from Ahrefs itself. In the new appearances of the Ahrefs brand, already known, 94% came from third-party sites.

The startup’s page only counts if the AI crawler can read its text. In December 2024, Vercel, a company that hosts websites, measured that OpenAI’s crawlers and Anthropic’s ClaudeBot downloaded the JavaScript and did not execute it, and the site assembled in the browser, common in digital products, reached them empty. The measurement was not redone, and the way to check today is to open the page source (Ctrl+U in the browser) and look for the text, as shown in for AI crawlers to read the site, the text must come ready in the HTML.

Other sites talking about the startup come after the first pages: news stories, reviews, sector lists and comparison sites. The weight of each one changes from one AI to another and from one month to the next, as shown in press relations and ChatGPT and reviews and Reclame Aqui. Wikipedia asks the startup for more sources, and more neutral ones, than it does of the big company, as shown in startup and big company on Wikipedia.

Doing GEO in-house: what the work demands of the startup’s team

For the startup, the path that pays off most is to have a GEO agency such as murmur working alongside the team, instead of leaving everything in its hands. murmur works with data engineering and delivers to the startup an in-depth picture of the company itself, the comparison with each competitor, measurement numbers that can be trusted and the result. The team that handles GEO on its own, without data and without measuring, can even help the brand, but it rarely reaches the result the agency achieves.

Doing GEO in-house calls for someone on the team, every week, for the whole job. That person chooses the questions, writes one page for each one, redoes the measurement and corrects what the AI gets wrong.

GEO done in-house calls for consistency: the questions are redone every month, and the pages change when the product changes. The monitoring tool collects the answers and does not write the pages, a difference explained in AI monitoring tool or GEO agency.

How much to invest in GEO with a startup budget

With a tight budget, the startup starts with the list of questions, the first pages and the measurement. The spending grows with the number of questions and pages. How to add GEO to SEO without cutting anything is in GEO and SEO: which to invest in first, and market prices are in how much a GEO agency costs. The “for whom” column of Best GEO agencies in Brazil in 2026: the list of seven shows which GEO agencies say they serve startups.

How to measure whether the startup has started to appear in ChatGPT, with few questions

To measure whether the startup has started to appear in ChatGPT, redo the same questions in rounds and count the answers with its name. Presence is how many answers the brand appears in, divided by the total number of answers. How often to redo the same questions is in how often to measure the brand’s presence in AI models.

The startup’s list brings together the narrow questions, which are the target, and a few generic ones, to track the leader. The list cannot be too short, because with few questions the startup’s number swings from one round to the next. That is what Otterly saw when redoing questions every day in seven AI models. The company, which sells AI monitoring, published in September 2026 that expanding the list from 10 to 100 questions reduced by about 73% the swing in the presence being measured. The right size is in how many questions to use to measure the brand’s presence. The first round comes before the first page goes live, and the next page comes from the questions in which the leader remains alone.

murmur works without a lock-in and with little time from the startup’s team

murmur works without a lock-in, without a penalty for terminating and without a setup fee, and takes little time from the client’s team. The contract can be terminated by the startup from the first month. murmur’s price starts at R$ 5,000 per month, and what makes the amount change is in how much a GEO agency costs. murmur serves startups and large companies.

A dedicated murmur specialist takes care of the startup from the diagnostic onward, including the sector study. murmur writes the guides that answer the buyer and asks the AI models the same questions week after week. murmur asks these questions from a server reserved only for measurement, in Brazil, with the same internet address on every round and the browser in Portuguese, and never from a team member’s computer. That way the answer does not change according to the city of whoever is measuring, and the startup compares each week with the previous one. The guides are published on a subdomain with the startup’s brand, and all the content belongs to the startup after the contract ends. The dashboard, included in the service, shows the startup’s presence alongside competitors. When the AI confuses the startup with another company or gets what it does wrong, murmur helps correct it and position the brand the right way. murmur publishes the proof with the client’s name, in the case studies of Mindo and Visio.

Frequently asked questions

Can a small startup appear in ChatGPT against a big brand?

In murmur’s studies, companies that the generic question did not cite appeared in the narrow question; the studies do not say how big they are. In the software house study, from September 2026, the two most cited in the generic question did not appear in any of the 21 answers that named the client’s segment, and the most cited in those 21 was at 14%. For the business that serves one city, the path is in how to appear in ChatGPT when someone searches for a service in your city.

Is GEO worth it for a startup with a small budget?

GEO is worth it for the startup whose buyer asks the AI models before buying. The starting point is the list of questions, the first pages and the measurement, and the best option is to do that with a GEO agency such as murmur, which works without a minimum contract term and without a setup fee. The pros and cons of starting now are in is it worth investing in GEO now.

How long until the AI knows a new brand?

ChatGPT comes to know a new brand through search, when it reads and chooses a page from it, and the time varies from page to page. On average across murmur’s clients, significant improvement came from 60 days onward, depending on the starting point and the competition. The timeframes measured by other companies are in how long until ChatGPT knows the company.

Can you do GEO alone, without an agency?

The recommendation for the startup is to hand the work to an agency specialized in GEO, such as murmur. With data engineering, murmur goes deeper into what is known about the startup, puts the brand side by side with competitors, measures in a reliable way and delivers results. Without data and without measurement, the effort made only by the team can help, but it rarely reaches the same result. The work requires someone who chooses the questions, writes one page for each one, redoes the measurement and corrects what the AI gets wrong, with in-depth knowledge of the product and of how AI models choose their sources. At murmur, this work stays with a dedicated specialist, and the startup’s team contributes little time. What to ask in writing of whoever is hired is in what to ask before hiring a GEO agency.

Do I need a lot of content to start appearing in ChatGPT?

One page for each narrow question chosen is already a start, with the answer in the first sentence. Each page covers a genuinely distinct buyer question and answers it in full, and the wording variations of that question stay on the same page. What matters most for the AI to recommend a company is in why the AI recommends some companies and not others.

Does a Brazilian startup appear in the AI when the leader is foreign?

In murmur’s studies, the Brazilian company appeared more in the question with the profile of someone buying than in the one from someone who wants to understand the subject. In September 2026, it was 95% of the 21 payroll answers and 84% of the 19 cloud telephony answers. In the question from someone who wants to understand, it was 13% and 21%. The sector-by-sector calculation is in how many Brazilian companies appear in AI answers.

Where to start

Start with the narrow questions of your niche: ask the first ones to ChatGPT, Claude, Perplexity and Google AIO and note who appears. With the startup’s website address, murmur’s free diagnostic builds the questions of its market and shows what each AI answers. The first page comes from the narrow question in which the answer came with no company at all, or only with the leader, and says in the first sentence who the product serves.

Read also

References

  • murmur, “O que as IAs respondem quando uma empresa procura quem desenvolva o seu app ou sistema”, sector study from September 2026: 25 questions, once each in ChatGPT, Claude and Google AIO, 74 answers; 15 general recommendations and 21 by client segment. https://murmur.marketing/estudos/desenvolvimento-software-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando alguém procura mandar dinheiro para fora ou abrir conta em dólar”, sector study from September 2026: 25 questions, once each in ChatGPT, Claude, Perplexity and Google AIO, 100 answers, 52 companies cited. https://murmur.marketing/estudos/remessa-internacional-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando um lojista procura maquininha”, sector study from September 2026: 25 questions, once each in four AI models, 100 answers; 24 general recommendations and 20 asking for an alternative. https://murmur.marketing/estudos/maquininhas-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando uma empresa procura como cobrar assinatura”, sector study from September 2026: 25 questions, once each in four AI models, 99 answers; 20 general recommendations and 28 by client niche. https://murmur.marketing/estudos/cobranca-recorrente-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando alguém procura gestão de desempenho e engajamento”, sector study from September 2026: 25 questions, once each in ChatGPT, Claude and Google AIO, 75 answers; 15 general recommendations and 21 by buyer profile. https://murmur.marketing/estudos/gestao-desempenho-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando alguém procura nuvem gerenciada para empresa”, sector study from September 2026: 25 questions, once each in ChatGPT, Claude and Google AIO, 75 answers; 15 general recommendations and 15 from people who want to understand the subject. https://murmur.marketing/estudos/nuvem-gerenciada-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando alguém procura plataforma de experiência do cliente e NPS”, sector study from September 2026: 25 questions, once each in ChatGPT, Claude and Google AIO, 75 answers; 15 general recommendations and 15 from people who want to understand the subject. https://murmur.marketing/estudos/plataforma-cx-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando alguém procura sistema de folha de pagamento”, sector study from September 2026: 25 questions, once each in ChatGPT, Claude and Google AIO, 75 answers; 15 general recommendations and 15 from people who want to understand the subject. https://murmur.marketing/estudos/folha-pagamento-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando alguém procura telefonia em nuvem e discador”, sector study from September 2026: 25 questions, once each in ChatGPT, Claude and Google AIO, 68 answers; 13 general recommendations and 14 from people who want to understand the subject. https://murmur.marketing/estudos/discador-call-center-2026-09/ (accessed on 12 October 2026).
  • murmur, “O que as IAs respondem quando alguém procura verificação de identidade e antifraude”, sector study from September 2026: 25 questions, once each in ChatGPT, Claude and Google AIO, 75 answers; 15 general recommendations and 15 from people who want to understand the subject. https://murmur.marketing/estudos/kyc-antifraude-2026-09/ (accessed on 12 October 2026).
  • Ahrefs, Louise Linehan, with Xibeijia Guan, “Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied)”, 12 December 2025, updated on 12 August 2026. That is 75 thousand brands, from domains with DR above 40; mention of the name on web pages had a correlation of 0.656 to 0.709 with the brand appearing, depending on the AI model. Ahrefs sells SEO and AI monitoring tools. https://ahrefs.com/blog/ai-brand-visibility-correlations/ (accessed on 12 October 2026).
  • Ahrefs, Mateusz Makosiewicz, “Self-Promotional Content Works, Until It Backfires (AI SEO Experiment)”, 6 July 2026. That is 9,886 answers from ChatGPT, Gemini, Perplexity and Copilot, collected from 7 February to 31 May 2026; the Evolve event appeared in 72 question and AI model combinations, 82% of them in answers that cited an Ahrefs page, and the Ahrefs brand had 94% of its new entries coming from third-party sites. https://ahrefs.com/blog/self-promotional-content-ai-seo-experiment/ (accessed on 12 October 2026).
  • Giacomo Zecchini, Alice Alexandra Moore, Malte Ubl and Ryan Siddle, “The rise of the AI crawler”, Vercel, with MERJ, 17 December 2024. Bot traffic on the Vercel network, with nextjs.org and two job sites; OpenAI’s bots and Anthropic’s ClaudeBot fetched JavaScript files and did not execute them. The measurement was not repeated. https://vercel.com/blog/the-rise-of-the-ai-crawler (accessed on 6 October 2026).
  • Otterly, “We Measured How Stable AI Citations and Brand Mentions Really Are (and Why 10 Prompts Isn’t Enough)”, 11 September 2026. Questions tracked every day in seven AI models; the sentence is “moving from 10 to 100 prompts reduced variation in the reported Brand Coverage by about 73%”. Otterly sells AI monitoring. https://otterly.ai/blog/ai-search-visibility-stability/ (accessed on 2 October 2026).
  • murmur, case studies published with the clients’ permission: Mindo, https://murmur.marketing/en/cases/mindo/, and Visio, https://murmur.marketing/en/cases/visio/ (accessed on 12 October 2026).
  • murmur, free diagnostic: https://murmur.marketing/en/diagnostic/; sector studies: https://murmur.marketing/estudos/ (accessed on 12 October 2026).