Updated on 7 November 2026, by João Leite, Founder of murmur.
To know how many pages to publish, count the questions your buyer asks that the site does not yet answer: each genuinely different question calls for a page, with the complete answer. There is no right number of articles or pages in GEO, the work of making the company appear in the answers of ChatGPT and the other AI models. Variations of the same question stay on the same page, and a question answered halfway calls for a revision, not a new page.
- One page per genuinely different question, with the complete answer: for Google, creating a page for every possible search variation, especially to manipulate its results, is spam, and Bing, ChatGPT’s search partner, warns that content made to manipulate its models can lose visibility or drop out of search.
- Revisions count too: at FotoRegistro, 84% of the new questions already had an article, and 5 new articles and 16 revisions were enough.
- First, the questions where the AI does not yet recommend your company, such as those from people who do not have a name in mind.
murmur, a GEO agency in Brazil, places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents. The AI Overview, or Google AIO, answers at the top of many Google searches with a summary written by AI. murmur studies each client’s sector before writing any page. The study gathers the questions the buyer asks AI models and decides which pages to produce and in what order. Then the same questions go back to the AI models week after week.
How to know how many pages to publish: the calculation in four steps
The calculation of how many pages to publish has four steps: list the questions, group the identical ones, check the site and choose the first group.
- List the questions the buyer asks before buying, each one with its source written down: a sales conversation, a Google suggestion, a sector forum or the question the AI suggests after answering. Search volume does not decide, because niche questions almost never have volume in keyword tools.
- Group the questions that call for the same answer. Two ways of asking the same thing fit on one page.
- Check what the site already answers. A missing question calls for a new page, and an incomplete one calls for a revision.
- Choose the first group and publish. Then ask the AI models the same questions again and decide the next group with the result in hand.
A round is each time the whole list is redone, and within it each question is asked more than once, on different days. The list changes with every round, with new questions in the market and pages that age. The guide on how to write a text to appear in ChatGPT shows how to write each one.
What matters most: useful content and the complete answer on each page
Google writes, in the guide on Google Search’s AI features, that unique, interesting and useful content should count more toward the site’s presence in these features, in the long term, than any other suggestion in the guide. In ChatGPT, the reference is Bing, its search partner: in the webmaster guidelines, the page has a better chance of being chosen to support answers and citations when its content stands on its own. The other reasons for a recommendation are in the guide on why AI recommends some companies and not others.
One page per different question, with variations on the same page
Each genuinely different buyer question gets a page, with the complete answer, and the variations of how people write the same question stay on that page. Google’s guide on search AI features goes in the same direction. There, Google writes that creating separate content for every possible search variation, with the main purpose of manipulating ranking or AI answers, violates its spam policy. Its examples are other queries people have already made and the smaller searches the AI makes from the question. For Google, this strategy is also ineffective in the long term, because many pages made this way do not make the site better or more relevant for the reader.
In the same guide, Google writes that AI systems can understand synonyms and the general meaning of what the person is looking for. For Google, this means the site does not have to cover every variation of how someone searches. Both guidelines are for Google Search, and do not talk about ChatGPT or Claude. On the ChatGPT side, there is the Bing rule, maintained by Microsoft and named by OpenAI as one of the assistant’s search partners. This rule, which is in the Webmaster Guidelines that Bing keeps online, says that content created to manipulate the language models of Bing and Copilot can make the site lose visibility or be removed from search.
Two questions are the same when the right answer is the same. “Quanto custa um sistema para clínica” (“How much does a system for a clinic cost”) and “preço de software para clínica médica” (“price of software for a medical clinic”) share the page, and “qual sistema escolher para uma clínica pequena” (“which system to choose for a small clinic”) gets its own, with size as the criterion. What separates one question from another is what the buyer wants to decide with the answer, not the change of words.
Answered, incomplete or missing: what the site already answers
The site already answers part of the questions on the list, and the gap calculation marks each one as answered, incomplete or missing.
| how the site answers the question | how to recognize it | what to do |
|---|---|---|
| answered | a page answers right in the first sentence, with the facts the buyer is looking for, and the AI bot can read the text | keep it, and check it every round |
| incomplete | the answer exists, but only in part, is outdated or is repeated across several pages | revise the page, or merge the ones that repeat the same answer |
| missing | no page on the site answers it | write a new page |
Only missing questions add up to a new page, and that is why the calculation can give far fewer pages than the list of questions. A page that is live but that the AI bot cannot read counts as missing. For AI bots to read the site, the text needs to come ready in the HTML, the code the server delivers.
At FotoRegistro, a photo printing e-commerce company with a factory in Joinville, in Santa Catarina, murmur mapped 395 questions across five business lines of the store. On 2 September 2026, 84% of the 44 new questions that had appeared in the market already had a FotoRegistro article answering them. That is why the next round needed only 5 new articles and 16 revisions, instead of 44 articles. Anyone who does the calculation this way arrives at the number of new pages through the unanswered questions, not through a target of articles per month.
First, the questions where the AI does not yet recommend your company
The first pages go to the questions where the AI does not yet recommend your company. To find them, ask ChatGPT, Claude and Google AIO the questions on the list and note who appears, as the guide on how to measure whether your brand appears in ChatGPT teaches.
murmur’s case studies bring three types of question from people who do not yet know the company:
- the questions where the AI recommends no company at all: the guide for Mindo, a motion design studio in São Paulo, started with them, such as training video and presentation for a corporate event;
- general questions, from people who do not yet have a name in mind: such as best animation studios, one of the questions where ChatGPT began recommending Mindo in July 2026;
- searches for an alternative to another system: at Visio, operations software for retail chains, they unlocked these searches and the general questions about management for retail.
Review the page that already exists before publishing another
Review the pages the site already has before publishing a new one. In a review of 45 studies on AI answers, published in July 2026, Olivier Martinez concludes that content that AI search has already brought in can come to be cited or used differently when it changes.
Three situations call for a review before a new page. The page with an outdated price, timeframe or product comes first, because AI repeats what it finds written. Two pages with the same answer become one, with the URL of the one that goes redirected to the one that stays. The page that answers halfway gets what is missing, such as the number, the price or the case study. Changing only the date, without new text, is not an update, as shown in the frequently asked question about changing the date of the text.
Publishing a lot of content with AI harms the site in Google search when the page does not help the reader
Publishing a lot of content harms the site in Google search when the pages exist to manipulate ranking, and not to help the reader. Google’s spam policy calls this scaled content abuse: generating many pages mainly to manipulate search ranking, with little or no value for the reader. The rule does not set a number of pages: what it judges is the purpose and the value of each page.
Writing with AI is not what the rule condemns, and how to do it is in the frequently asked question about writing with the help of an AI. Anyone who published many similar pages and saw Google search traffic fall starts with the Manual actions report in Search Console. It shows whether the site received a penalty after a review by a person at Google. The other signs are in the guide on when organic traffic fell because of your site, and not because of AI.
Questions to measure and pages to publish are two separate calculations
The questions tracked in AI models and the pages to publish are two different calculations. The first measures the brand’s presence. Presence is how many answers the brand appears in, divided by the total number of answers. The size of that list is in the guide on how many questions to use to measure brand presence in AI models. The second answers the buyer’s question, and one well-made page answers several questions on the list when they all ask for the same answer.
At murmur, the number of pages comes from each client’s sector study
At murmur, the number of pages for each client comes from the study of that client’s sector, and not from a package with a fixed quantity. murmur produces the content that fills the client’s gaps. There is no cap on questions, hours or pieces.
At murmur, a specialist dedicated to the brand follows each round closely, and its result decides the next page or the next review. The pages go live on a subdomain, a URL within the brand’s domain. The dashboard, which comes included in the service, shows the brand’s presence and each competitor’s side by side. The content is all the client’s, even after the contract ends.
Frequently asked questions
Can I do GEO alone?
Doing it alone is not the recommended path: the best option is to have a GEO agency leading the work, as is the case with murmur. It starts from data engineering and gives the client in-depth information about their business, their presence compared with competitors, reliable measurement at each round and results. Those who publish alone, without data and without measuring, may even gain some ground, but it is hard to reach the same result. At murmur, the plan comes from the client’s sector study, and the pages it calls for are part of the service. To compare the two paths, see the guide on how to explain GEO to your boss or the board.
How much does it cost to hire GEO, and does the price change with the number of pages?
At murmur, the work costs from R$ 5,000 per month, and there is no package based on the number of pages. The plan comes from the study of each brand. The amount changes with the business lines, the complexity, the languages, the number of brands and other factors. The guide on how much a GEO agency costs puts that amount next to the published prices of other agencies and tools.
How long does it take to appear in ChatGPT?
The first citation of a new page in ChatGPT can come the next day, or not come in the first month. The company’s presence across many questions takes longer. On average across murmur’s clients, significant improvement came from 60 days onward, depending on the starting point and the competition. It is an average across clients, and not another company’s timeline. The dates from publication to first citation are in the guide on how long it takes to appear in ChatGPT.
How often should AI answers be analyzed after publishing?
Analyze AI answers at least once a month, always with the same questions. The effect of each group of pages is read in the sum over the period, and not day by day, because the answer changes from one conversation to another. The guide on how often to measure brand presence in AI models has the full schedule.
Before ordering pages, see what AI models answer in your market
Before ordering pages, ask ChatGPT and Google AIO your buyer’s questions and note which ones your company is missing from. murmur’s free diagnostic gives you that list in advance. Enter the site URL and see, without signing up and in less than two minutes, what AI models answer about your market. murmur’s sector studies offer a snapshot of one round of purchase questions in other markets.
Read also
- How to write text to appear in ChatGPT: answer up front and numbers with sources
- Why AI recommends some companies and not others: the four reasons
- 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
References
- Google, “Optimizing your website for generative AI features on Google Search”, Google Search Central, updated on 10 July 2026. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide (accessed on 1 October 2026). Applies to Google search, with AI Overview and AI Mode.
- Google, “Políticas de spam para a Pesquisa Google na Web” (“Spam policies for Google Web Search”), Google Search Central, updated on 16 September 2026. https://developers.google.com/search/docs/essentials/spam-policies?hl=pt-br (accessed on 1 October 2026).
- Microsoft, “Bing Webmaster Guidelines”, Bing Webmaster Tools, undated page. https://www.bing.com/webmasters/help/webmaster-guidelines-30fba23a (accessed on 1 October 2026).
- OpenAI, “Searching the web with ChatGPT”, OpenAI Help Center, no exact date on the page. Lists Microsoft among ChatGPT’s search providers. https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt (accessed on 3 October 2026).
- Olivier Martinez, “Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026)”, arXiv 2607.14035, 15 July 2026. https://arxiv.org/abs/2607.14035 (accessed on 1 October 2026).
- murmur, case studies for FotoRegistro, Mindo and Visio, published with the clients’ authorization: https://murmur.marketing/en/cases/fotoregistro/, https://murmur.marketing/en/cases/mindo/ and https://murmur.marketing/en/cases/visio/; free diagnostic: https://murmur.marketing/en/diagnostic/; sector studies: https://murmur.marketing/estudos/ (accessed on 1 October 2026).