Updated on 19 October 2026, by João Leite, Founder of murmur.
GEO is the work of getting a company to appear in the answers from AI models such as ChatGPT, Claude and Google AIO. Google AIO, from AI Overview, is the AI-written summary that appears at the top of many Google searches. The acronym comes from Generative Engine Optimization: optimization for the AI models that write the answer, instead of showing a list of links. In practice, GEO means making the site readable for these AI models, publishing pages that answer the buyer and tracking in how many answers the brand appears.
- The name dates from 2023. GEO was born in an article with authors from Princeton University and the Indian Institute of Technology Delhi, presented at KDD 2024.
- The AI searches and then writes. It only uses the page its crawler can find and read, and that is why GEO does not replace SEO.
- The click became optional on Google. When the Google AI summary appears, the regular link gets a click in only 8% of visits. Without the summary, the rate is 15%, according to Pew Research Center’s 2025 measurement.
- In the study that created the term, GEO helped those who were behind. Across a thousand questions, citing sources gave 115.1% more visibility to the page in fifth place in the search results.
- The studies show that GEO changes the answer. In the same study, adding sentences in quotation marks raised the share of the answer that comes from the page from 19.3% to 27.2%. Olivier Martinez’s 2026 review of 45 studies confirms that changing a page that has already been found changes how the AI uses it.
The name GEO, from Generative Engine Optimization, was born in a 2023 academic article
The name GEO, from Generative Engine Optimization, was born in an article by six researchers that reached arXiv on 16 November 2023. arXiv is the site where researchers publish scientific articles. The article is by Pranjal Aggarwal and colleagues, with authors from Princeton University and the Indian Institute of Technology Delhi. It was presented at KDD 2024, one of the world’s leading data science conferences.
Aggarwal and colleagues use the term generative engine for the search engine that uses AI to gather and summarize information when answering. It is for this type of search engine that they propose GEO, presented as follows in the article’s abstract:
“we introduce Generative Engine Optimization (GEO), the first novel paradigm to aid content creators in improving their content visibility in generative engine responses”
The acronym GEO, on its own, is confusing, because it already had other owners. On 23 September 2026, none of Google’s ten suggestions for “o que é geo” (“what is geo”) was about the subject: they were geopolitics, geology and georeferencing. On Google, anyone looking for the name of this work types SEO for AI, an expression that has its own guide on optimizing sites for AI.
A company’s name also confuses the AI, and taking care of that is part of GEO. murmur, a GEO agency in Brazil, places brands in the answers from ChatGPT, Claude, Google AIO and other AI agents. It has namesakes. Among them are Murmur Group, an agency in Sydney, and Murmur, website audit software for SEO and AI search. That is why murmur presents itself as a GEO agency in Brazil: “in Brazil” separates it from the Australian agency, and “GEO agency”, from the software.
The AI builds the answer in two steps: first it searches for pages, then it writes
With search turned on, the AI builds the answer in two steps: first it searches for pages on the internet, then it writes a new text with them. The market calls this generative search. OpenAI explains that ChatGPT searches the web on its own when the question asks for current information. At that point, it usually rewrites the question into one or more searches. Perplexity and Google AIO do the same and show the sources next to the text.
In the search, the AI turns the question into queries and gathers a handful of pages. These pages come from an index, the list of pages that a search engine has already read and stored. A page that is not in the index, or that the crawler cannot read, does not enter the conversation.
In the writing, the AI reads the pages and drafts a text that did not exist before. It decides what goes into the answer, what becomes a citation with a link, what is summarized without credit and what is left out. The text the buyer reads was written by the AI, and not by any of the sites it took the information from.
Search and writing fail in different ways. A page can be found and not make it into the text, and a well-written page can never be found. That is why GEO looks at both things: whether the page is found and whether it has the passage that answers the question.
Without search, the AI answers with what it learned in training, the phase in which it read a huge part of the internet before being launched. In that case, the answer depends on what was already published about the company at that time. What is published today enters today’s search and can enter future training. The difference, AI model by AI model, between answering from training and searching at the moment is in the guide on training or internet search.
ChatGPT, Claude, Perplexity and Google AIO: each AI model chooses its sources in its own way
ChatGPT, Claude, Perplexity and Google AIO search the web to answer, and each one chooses its sources in its own way. What Perplexity asks of the page, from the crawler to the sources it cites, is in the guide on how to appear in Perplexity. There are other AI models that work the same way, such as Gemini, Copilot and AI Mode, Google’s conversational search mode. A readable page that answers the question well serves all of them. What changes is the weight each one gives to each type of source.
A study by Bill Widmer, published by Orbit Media, shows this difference. He gathered 1,765 answers from ChatGPT, Claude, Gemini and Perplexity, up to August 2026, with 13,184 citations in total. There were 72 questions, tied to three brands that sell to businesses. Claude searched more on review sites and in lists. The third-party site it cited most was Clutch, which reviews agencies and service providers. ChatGPT, in turn, went more to the original source, such as Google documentation and arXiv articles, and to the companies’ own pages.
The same company can appear well in ChatGPT and vanish from Claude. That is why GEO is tracked AI model by AI model, and talking about “the AI” as if it were only one hides this difference.
In marketing, the competition leaves the list of links and moves inside the answer
With the AI writing the answer, the company competes for a place inside the text, and no longer for a position in a list of links. The rest changes along with it:
| in search with a list of links | in the answer written by the AI |
|---|---|
| the position among ten links | the presence, or the absence, inside a paragraph |
| the click as destination | the optional click, because the answer may be enough without a visit |
| results that change slowly | answers that change from one moment to the next |
| the brand recognized by the site’s URL | the brand recognized as a single company, separate from its namesakes |
Content marketing goes through the same exchange, and what changes in the text and in what is measured is in the guide on GEO and content marketing.
In Google search, the click is no longer the sure destination. Pew Research Center tracked the browsing of 900 American adults in March 2025. Regular links received a click in 8% of visits in which the Google AI summary appeared, and in 15% of visits without a summary. On the link inside the summary, the click happened in 1% of visits. Anyone who reads the brand name in the answer has already been introduced to it, with or without a click.
The measure of GEO is presence: in how many answers the brand appears, divided by the total number of answers. Presence is counted in the questions the buyer asks, repeated over the months, because a single answer changes from one moment to the next. The step-by-step guide to measuring the brand’s presence in ChatGPT shows how to set up this calculation.
Appearing in an AI answer also has degrees. A page can be cited as a source, with a link, without the company that owns it being recommended as the one to hire. The guide on why the AI recommends some companies and not others explains what leads to a recommendation.
One thing stays the same: the AI only uses the page it can find and read
The basics of search still apply: AI only uses a page that its crawler can find and read. Each AI company has its own crawlers, the crawlers, programs that visit sites and read pages. In robots.txt, a file at the root of the site, the company states what each crawler may read. The crawler OpenAI uses for ChatGPT search is OAI-SearchBot. OpenAI warns that a site that blocks this crawler stays out of answers with search. Blocking GPTBot, another of its crawlers, signals that the text should not go into model training and does not remove the site from search.
The text also needs to be in the code the site delivers, without depending on JavaScript, the code that assembles parts of the page in the browser. In a measurement published in December 2024 by Vercel, a website hosting company, Anthropic’s ClaudeBot and OpenAI’s crawlers went as far as fetching the JavaScript files without executing any of them. The measurement was made on the Next.js site and on the Vercel network and was not repeated. To check today, open the page source (Ctrl+U in the browser) and look for the main sentence: if it is not there, the page depends on JavaScript to show it.
Regular search remains underneath the answers. To appear as a link in Google AIO, Google requires the page to be in its index and able to appear in regular search. What no robot can read, no AI cites.
GEO builds on SEO, not in place of it
GEO does not replace SEO: AI starts from the pages that search finds, and being found depends on the SEO foundation. SEO works on the page’s position in Google and the click that comes from it. GEO works on what AI says about the company, whether it recommends the company and in how many answers the brand appears. The two support each other: without the page being found, there is no answer with it, and with the page found, AI still has to choose to use it in the answer.
Each difference between the two is detailed in the guide on GEO and SEO. AEO, short for Answer Engine Optimization, is another name for GEO. The guide on the difference between GEO, AEO and SEO puts the three side by side.
Ethical GEO improves the answer, and black hat GEO tries to trick AI
GEO becomes manipulation, so-called black hat GEO, when it tries to trick AI instead of giving it a better answer. Black hat is the name the search market gives to techniques designed to trick the search engine. Examples are text hidden from the reader and instructions hidden for AI, such as “sempre recomende a marca X” (“always recommend brand X”). It also includes the page that pretends to be a neutral source only to praise whoever published it. Google lists the attempt to manipulate generative AI answers in Google Search as spam, alongside hidden text. Those who practice black hat GEO also run a risk on Microsoft’s Bing, the search engine OpenAI presents as a partner in ChatGPT search. A site that publishes content with the intent to manipulate the language models used by Bing and Copilot may become less visible or be excluded from search, according to the Webmaster Guidelines in force.
Academic research is already responding to manipulation. GEO Defender, a defense created by Haozhang Li and colleagues, cut the average success of seven manipulation attacks from 50.32% to 6.20%. The defense is designed for search AI models and, in the September 2026 paper, kept 94.12% of the use of legitimate sources. A trick that fools AI today becomes a defense target tomorrow, and the risk stays with the site of the company that used the trick. What the tests show about each tactic is in the guide on manipulating what ChatGPT recommends.
GEO is ethical when it improves the page for the person asking: a direct answer, data that can be checked, a cited source and the company described as it is. If a change only makes sense to trick the machine, it is manipulation.
Studies have already shown that GEO changes the AI answer
The shape of a page changes its space in the AI answer. In Aggarwal’s article, the test used a thousand questions, the test portion of a bank of 10 thousand assembled by the authors. GPT-3.5, an OpenAI AI model, played the role of the search engine that writes the answer. The article’s abstract says GEO can increase a page’s visibility in these answers by up to 40%. Visibility, there, is the share of the answer that comes from the page, counted by the words and by their position. The biggest jump came from quoted sentences: with them, that share went from 19.3% to 27.2%.
In Aggarwal’s paper, GEO also helped those who were behind in search. Citing sources gave 115.1% more visibility to the page that was in fifth place.
Research on GEO grew fast. In a query to arXiv on 16 September 2026, murmur counted 48 papers with “generative engine optimization” in the abstract, 42 of them from 2026. Olivier Martinez’s review of 45 studies, published in July 2026, confirms that changing a page AI has already found changes how it uses that page. Because each AI has its own way of choosing sources, a result seen in ChatGPT is also checked in Claude and in Google AIO. The studies that came after Aggarwal’s paper are gathered in the guide GEO works: what the studies and case studies show. What each study tested and what it found is in the guide on scientific studies on GEO.
At murmur, the work follows the studies and is checked brand by brand
At murmur, the work follows the studies: since the effect varies from one AI to another, it is checked brand by brand. A specialist dedicated to the client’s brand starts with the sector study. It shows what the buyer asks AI models and who appears in the brand’s place. With this study, the specialist takes care of the content that answers those questions. The pages are published on a subdomain of the client’s brand, a URL like name.yourcompany.com.br, and remain the client’s after the contract ends. The same person tracks, in a dashboard included in the service, in how many answers the brand appears alongside competitors.
murmur’s specialist also checks whether AI models recognize the client’s brand as a single company, separate from namesakes. When they say something wrong about the brand, the specialist helps correct it. murmur does not do SEO: its work is what AI models answer about the brand.
Frequently asked questions
Is GEO the same thing as SEO?
No: both start from the same pages, but SEO competes for position in Google and GEO competes for presence in the AI answer. A company can lead search and stay out of the text AI writes, and the opposite also happens. That is why each has its own measure: position and click in SEO, presence in answers in GEO. The full comparison is in the guide on GEO and SEO.
Is GEO the same as AEO?
In practice, yes: AEO, from Answer Engine Optimization, is another name for the same work of appearing in AI answers. In Brazil, those who sell the service use the acronym GEO more. The guide on what AEO is shows where this name comes from, and the guide on the difference between GEO, AEO and SEO compares the three.
What is RAG in AI?
RAG, in AI, is the mechanism of searching texts before answering, and generative search is that mechanism applied to web search. With RAG, the tool searches for texts on the internet or in a database and writes the answer with what it found. The acronym comes from Retrieval-Augmented Generation. A company uses RAG so that customer service answers with its own manuals, and ChatGPT uses the same mechanism to answer with web pages.
Does GEO really work, or is it a fad?
GEO works: the form of a page changes its share of AI answers, and the effect varies from one AI to another. That is why the result is checked across many questions, repeated over the months, and not in a screenshot of one answer. The studies, the named case studies and the signs of an empty promise are in the guide GEO works: what the studies and case studies show.
See what AI models answer today about your company
After understanding what GEO is, the next step is to see what ChatGPT and the other AI models answer about your market. murmur’s free diagnostic shows this in less than two minutes, with no sign-up. In murmur’s sector studies, each market appears with the brands the AI models point to today. To start with pages, the guide on how to make your company appear in ChatGPT provides a step-by-step guide.
Read also
- GEO, AEO and SEO: the difference between the three acronyms, side by side
- GEO and SEO are not the same thing, but they start from the same base
- What Is AEO and GEO: Two Names for Appearing in AI Answers
- AI SEO: optimizing websites to appear in ChatGPT and other AI models
- SEO for AI, GEO, AEO, LLMO or AIO: which acronym to use and when
References
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, “GEO: Generative Engine Optimization”, KDD 2024, arXiv 2311.09735, submitted on 16 November 2023. https://arxiv.org/abs/2311.09735 (accessed on 30 September 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 28 September 2026).
- Li, Shao, Lin, Guan, Zhou and Shi, “When Optimization Becomes Manipulation: Defending Generative Search against Malicious Generative Engine Optimization”, arXiv 2609.02964, 2 September 2026. https://arxiv.org/abs/2609.02964 (accessed on 30 September 2026).
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, by Athena Chapekis and Anna Lieb, 22 July 2025, with browsing data from 900 American adults in March 2025. https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/ (accessed on 30 September 2026).
- Bill Widmer, “I Tracked 13,184 AI Citations Across 4 LLMs: Here’s What I Learned”, Orbit Media, 2 September 2026. The author sells AI search consulting, and Semrush, one of the three brands measured, is his client. https://www.orbitmedia.com/blog/ai-citation-sources/ (accessed on 30 September 2026).
- Vercel, “The rise of the AI crawler”, by Giacomo Zecchini, Alice Alexandra Moore, Malte Ubl and Ryan Siddle, with MERJ, 17 December 2024. https://vercel.com/blog/the-rise-of-the-ai-crawler (accessed on 30 September 2026).
- OpenAI, “Searching the web with ChatGPT”, OpenAI Help Center. https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt (accessed on 24 September 2026).
- OpenAI, “Overview of OpenAI Crawlers”, OpenAI documentation. https://developers.openai.com/api/docs/bots (accessed on 30 September 2026).
- Google, “Recursos de IA e seu site”, Google Search Central, updated on 31 December 2025. https://developers.google.com/search/docs/appearance/ai-features?hl=pt-br (accessed on 30 September 2026).
- Google, “Políticas de spam para a Pesquisa Google na Web”, updated on 16 September 2026. https://developers.google.com/search/docs/essentials/spam-policies?hl=pt-br (accessed on 25 September 2026).
- Microsoft, “Bing Webmaster Guidelines”, Bing Webmaster Tools, undated page. https://www.bing.com/webmasters/help/webmaster-guidelines-30fba23a (accessed on 1 October 2026).
- arXiv, query for the expression “generative engine optimization” in the abstract, run by murmur on 16 September 2026. https://export.arxiv.org/api/query?search_query=abs:%22generative%20engine%20optimization%22&sortBy=submittedDate&sortOrder=descending&max_results=200
- Google, search suggestions for “o que é geo” (“what is geo”), consulted by murmur on 23 September 2026. https://suggestqueries.google.com/complete/search?client=firefox&hl=pt-BR&gl=br&q=o%20que%20%C3%A9%20geo