Updated on 15 November 2026, by João Leite, Founder of murmur.

The case study page that AI uses states in text the client, the sector, the result with what it measures and when. In the test by Aggarwal and colleagues, published on arXiv in 2023, the three best-performing techniques, citing sources, including quotations and including numbers, increased the page’s space in the AI answer by 30% to 40%, in one of the study’s measures. A well-written case study brings all three. murmur, a GEO agency in Brazil that places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents, tracked three questions about GEO case studies across four AI models. In the rounds from 12 October to 2 November 2026, more than half of the 48 answers used a case study page as a source.

  • Each case study brings the client with name and sector, the starting point, what was done and the result. The result goes in percentage, with what it measures, in which AI or place and when, and the case study carries the date of the last update.
  • Numbers, quotations and sources help the page gain space in the answer. In the test by Aggarwal and colleagues, the three techniques increased that space by 30% to 40%, in one of the study’s measures.
  • In the questions about case studies, the case study page appeared as a source. From 12 October to 2 November 2026, more than half of the 48 answers to the three questions about GEO case studies in murmur’s measurement used a case study page.
  • The law calls for truth in results and testimonials. Article 37 of the Consumer Protection Code calls false advertising information misleading, and the CONAR Code calls for a genuine and verifiable testimonial.
  • A URL for each case study, linked to the case study page. The client’s name and the number go in text, and not only in the logo or in an image.

The case study page that AI uses states who the client is, what changed and when

The case study page that AI uses answers, in text, three things: who the client is, what changed for them and when. Anyone who asks the AI for an example of a result wants a name they can check, a number that says the size of the change and a date that says whether the result is recent. The AI, when it searches the internet before answering, writes the answer with what it finds on the pages. A case study page with logos, phrases like “we increased sales” and no number adds little to that answer.

A case study is the short story of a client: where they were, what was done and what happened afterwards, with numbers and dates. The case study page gathers these stories and links each one to its URL. It serves two readings at the same time. The buyer reads it to decide whether the company solves a problem similar to theirs, and the AI reads it to answer who has results in a market.

The way of writing each part follows the rules of any page that AI cites: the answer at the beginning and the numbers with what they measure. The step-by-step for writing is in the guide on how to write a text to appear in ChatGPT, with numbers and sources. On the case study page, the rule applies to each client, one by one.

What the studies show about numbers and sources on the page that AI cites

The studies show that numbers, quotations and sources on the page increase its space in the AI answer. The study by Aggarwal and colleagues tested several ways of rewriting a page and measured how much of its text entered the answer. Citing sources, including quotations and including numbers were the three best-performing techniques, with a gain of 30% to 40% in one of the measures. A case study with the client’s number and the source of the data already uses two of the three techniques; the testimonial comes in when the client authorizes it.

Martinez’s review, which gathered 45 studies published up to July 2026, reaches a similar conclusion by another route. In the table that summarizes the degree of support for each factor, the evidence that can be extracted from the page has moderate to strong support. The recommendation with the most support in the review is to publish a relevant, complete, verifiable, well-organized page that the crawler can read. A case study with name, number and date is verifiable by definition: anyone who reads it can check.

The third study says where the number needs to be. In a test by Otterly, a company that sells AI monitoring, a piece of data written only inside images came back right in 3 of 120 attempts, and the three came from ChatGPT, when the file name and the alt text carried the data. More than 55% of the answers in the test invented a number. The test used a single domain, and Otterly writes that the AI models, outside deep reasoning mode, did not read the text inside the images. That is why the case study result is written on the page, and not only in a chart or a logo.

In the questions about case studies, the AI used the case study page as a source

In murmur’s measurement, from 12 October to 2 November 2026, more than half of the 48 answers to the three questions about GEO case studies used a case study page as a source. The measurement tracks the same questions in ChatGPT, Claude, Perplexity and Google AIO, the AI-written summary that Google shows at the top of many searches. There were four rounds, with the same three questions in each one. The pages used were agency case study pages.

Three questions show what happened in them, and not a rule for every question about results. The questions were about GEO case studies, murmur’s own market. In your company’s market, the case study question is different, and the way to check is the same: ask the AI the question in the buyer’s words and look at the sources the answer shows.

When the sources of the answer are case study pages of other companies, your case study page is out of the conversation. The way to get into it is to publish the case study with what the AI needs to find written out: the client, the number and the date.

What each case study on the case study page needs to have

Each case study on the case study page needs seven parts, all in text and with the words the buyer uses:

  1. The client, with name and sector.
  2. The starting point, with the date.
  3. What was done, with the dates of each step.
  4. The result in percentage, with what it measures, in which place and when.
  5. The basis of the number: how many questions, answers or requests the percentage comes from.
  6. The client’s testimonial, with name and role, when they authorize it.
  7. The date of the last update of the case study.
partwhat to writewhy it serves the AI answer
client and sectorthe company’s name and what it does, such as “B2B motion design studio, in São Paulo”the question about case studies asks for a company that can be checked
starting pointthe situation before the work, with the dateit gives the result a before, and the before shows the change
what was donethe main actions, in a few sentences, with datesit links the result to the work
resultthe percentage, what it measures, where and whena number on the page was one of the best-performing techniques in Aggarwal’s test
basishow many questions or answers the number comes fromit lets the reader check the size of the measurement
testimonialthe client’s quote, with name and role, when they authorize itincluding quotations was another of the best-performing techniques in the same test
datethe day, the month and the year of the last updateit tells when the result dates from

The part most often missing from case study pages is the number with what it measures. “200% growth” without saying of what, where and between which dates answers nothing. On the page of a company that sells to other companies, the case study is one of the items in the step-by-step guide to GEO for B2B companies, alongside the price and the service pages.

The case study number: what it measures, in which place and when, in percentage with the basis

The case study number says what it measures, in which AI or place and on which dates, and comes in percentage with the basis alongside. Two numbers from the same client can measure different things, and the case study needs to separate the two. In the case study for Mindo, a B2B motion design studio, in São Paulo, ChatGPT recommended the brand in 23% of the questions in its market on 13 and 14 July 2026, against 12% on 25 and 26 June. In the same period, ChatGPT put Mindo in first place in 16% of the questions, against 3%.

Recommending and putting in first place are different measures, and each one comes written with its own name. The same goes for the AI: Mindo’s number in Google AIO is a different one, and measures something else. In July 2026, Google AIO used Mindo’s guide as a source in 36% of the answers. Anyone reading the case study knows, in each sentence, which AI the number is from and what it counts.

The percentage with the base makes the result comparable. The percentage gives the proportion, and the base gives the size: Mindo’s questions are 134, the same in June and in July. Without the base, the reader does not know whether the number came from ten questions or a thousand. Without the dates, the reader does not know whether the result is from this year.

A case study shows what happened with that client, on those questions, in that period. It does not become a rule for another company, and the case studies page does not promise the same result to any future client.

The client’s authorization and what the law asks of anyone who publishes a result and a testimonial

A named case study starts with the client’s authorization to publish the name, the sector, the numbers and the testimonial, agreed in writing. Without authorization, the case study goes out without a name, with the sector described, and loses part of what the AI can use. With authorization, each number has to be true and verifiable.

The Consumer Protection Code, Law No. 8,078/1990, deals with false advertising in article 37. Paragraph 1, published on Planalto, says that “É enganosa qualquer modalidade de informação ou comunicação de caráter publicitário, inteira ou parcialmente falsa” (“Any form of information or advertising communication that is wholly or partly false is misleading”). That is why each number in the case study is what was measured, with the sample and the date.

The Brazilian Code of Advertising Self-Regulation, from CONAR, deals with testimonials in article 27, paragraph 9, in the 2026 edition. The text of the CONAR Code says that the ad will contain only “depoimentos personalizados e genuínos, ligados à experiência passada ou presente de quem presta o depoimento” (“personalized and genuine testimonials, tied to the past or present experience of the person giving the testimonial”), and that “o testemunho utilizado deve ser sempre comprovável” (“the testimonial used must always be verifiable”). How the law and the Code apply to your company’s case studies page is a question for legal. The safe way to write is the same in both: what the client said, with their name, and what was measured, with the sample and the date.

A screenshot of an AI answer sometimes goes into the case study as proof. It proves one answer, on one date, and not the brand’s presence over time, as the guide on what a screenshot of an AI answer proves shows.

One URL per case study, and the case studies page linking all of them

Each case study gets its own URL, with the client’s name, and the case studies page links all of them. The dedicated URL keeps the whole case study on a single page, with the client, the number and the date together. The URL carries the words that matter, such as the client’s name, and has no parameter, the part after the ”?” that some sites add. Mindo’s case study, for example, sits at a URL with the client’s name.

The case studies page works as an index. For each client, it shows the name, the sector, the main number with what it measures and the date, and leads to the full case study. That way the case studies page already answers who has results in which market, and each case study answers how the result happened.

The service pages also link to the case study from the same sector. Anyone who reads a service page gets to the result of a client who bought that service, and the AI finds the two linked. To check a GEO agency’s case study before hiring, the path is in the guide on what to ask before hiring a GEO agency, and what each agency publishes is in the guide on GEO agencies in Brazil and what each one publishes.

murmur publishes case studies with the client’s name and what each percentage measures

murmur publishes case studies with the client’s name, and each percentage comes with what it measures, in which AI and when. The FotoRegistro case study, a photo printing and personalization e-commerce company with its own factory in Joinville, in Santa Catarina, reports that ChatGPT cited the brand’s guide for the first time on 24 July 2026, 30 days after the guide launched. The same case study shows that, on 2 September 2026, 84% of the new questions that appeared in FotoRegistro’s market already had an article from the brand answering them.

Mindo’s case study starts with the starting point. On 29 May 2026, Mindo did not appear in any of the answers murmur measured across three AI models. The site was an application that built the page in the browser, and the AI crawler saw an empty page. murmur published the guide on a subdomain of the brand, guia.mindo.com.br, and on 22 June 2026 it was fully live. In July 2026, ChatGPT recommended Mindo in 23% of the questions in its market, against 12% in June.

These case studies show the work murmur does to place the brand inside AI answers. Each brand has the same person at murmur from the diagnostic onward, who knows the company and the sector, sees what the AI models answer to the buyer and who takes its place, and takes care of the content that changes that answer. The content goes out on a subdomain of the client’s brand, like Mindo’s, and stays with the client after the contract ends. The measurement tracks the same questions every week, with the dashboard included in the service.

Frequently asked questions

Can I publish a case study without the client’s name?

You can, but an unnamed case study answers less: anyone who asks the AI for an example gets a sector, and not a company that can be checked. When the client does not authorize the name, describe the sector, the size and the region precisely. The number still comes with what it measures, where, when and the sample. The named case study goes in as soon as the client authorizes it.

How often should the case studies page be updated?

Update the case studies page whenever a client has a new measurement, and write the date on each case study. The old number can stay, with its date, as the starting point for the new number. The case studies page also gets a new case study as soon as a client authorizes it. A case study without a date looks old to the reader, even when the number is recent.

Does a case study need a number?

A case study with a number answers more than a case study without one, because anyone asking about results wants to know what changed and by how much. When the result does not yet have a measured number, the case study describes the change with verifiable facts and dates. The number goes in as soon as there is a measurement, with what it measures, where and when. A percentage with no sample and no date does not count as a case study number.

Does a client testimonial without a number work as a case study?

It works as a testimonial, and not as a complete case study: the measured result that the question about a case study is looking for is missing. The testimonial gains strength next to the number, with the name and the role of the person who said it. The CONAR Code, in article 27, paragraph 9, calls for a personalized, genuine and verifiable testimonial. Keep the authorization and the client’s original wording.

Can I put only the client’s logo on the case studies page?

The logo can stay, but the client’s name also goes in text, next to the sector and the result. In Otterly’s test with data written inside images, AI models almost never got the data right, and more than 55% of the answers invented a number. A strip of logos without text shows anyone viewing the page who is a client. For the AI, the written name is what counts.

Where to start

Start with the clients who already authorize the name and with one number from each, with what it measures, where and when. To find out whether your company already appears when the buyer asks for examples of results, murmur’s free diagnostic reads the AI answers about your market in less than two minutes, with no sign-up, and sends the full report by e-mail. murmur’s sector studies give a picture of who already appears in each market: 25 questions per sector, asked in a single round, with the brands the AI models listed.

Read also

References

  • Aggarwal and colleagues, “GEO: Generative Engine Optimization”, arXiv 2311.09735, 2023. The three best-performing techniques (citing sources, adding quotations and adding statistics) showed gains of 30% to 40% on one of the study’s measures. https://arxiv.org/abs/2311.09735 (accessed on 3 October 2026).
  • Martinez, “Optimizing Visibility in Generative Engines: A Critical Survey of GEO (2023-2026)”, arXiv 2607.14035v1, 15 July 2026. A review of 45 studies; in table 4, the evidence drawn from the page has moderate to strong support. https://arxiv.org/abs/2607.14035 (accessed on 8 October 2026).
  • Otterly, “GEO Experiment: Can AI Search Read Image Metadata?”, undated page. Of the 120 attempts, 3 got the data right, all in ChatGPT; more than 55% made up the answer; one test domain. https://otterly.ai/blog/geo-experiment-image-metadata/ (accessed on 1 October 2026). Otterly sells AI monitoring.
  • Planalto, Law No. 8,078/1990 (Consumer Protection Code), consolidated text, article 37, paragraph 1: the definition of misleading advertising. https://www.planalto.gov.br/ccivil_03/leis/l8078compilado.htm (accessed on 7 November 2026).
  • CONAR, “Código Brasileiro de Autorregulamentação Publicitária”, 2026 edition, article 27, paragraph 9: personalized testimonials, genuine and verifiable. https://conar.wpenginepowered.com/wp-content/uploads/2026/08/Codigo_CONAR_2026.pdf (accessed on 7 November 2026).
  • murmur, weekly measurement in ChatGPT, Claude, Perplexity and Google AIO: three questions about GEO case studies, in four rounds from 12 October to 2 November 2026, with 48 answers; no page published.
  • murmur, published case studies: Mindo, https://murmur.marketing/en/cases/mindo/; FotoRegistro, https://murmur.marketing/en/cases/fotoregistro/ (accessed on 6 November 2026).