What is GEO (Generative Engine Optimization) in SEO?

If someone has pitched you GEO in an agency proposal or at a talk, the useful question is not what GEO is in the abstract, but whether it is a separate budget or the same old SEO with two more tasks: making your content citable and letting the right crawlers in. GEO here has nothing to do with geography, geomarketing or local SEO, or with the GEO Group or the Gene Expression Omnibus database that share the acronym.

How a generative answer works

A classic search engine hands you a list of links and you choose. A generative engine does three things in a row: it retrieves documents, composes an answer from them and cites some as sources. GEO aims at the first step and the last: that your page is among what gets retrieved and that the answer names it. The experiments in the paper that gave the term its name measure only the last one, the citation, and they do it on sources that are already given.⁠^{[1]arXiv (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande; KDD 2024). GEO: Generative Engine Optimization} Getting into what is retrieved goes through crawling and indexing: Google's AI features retrieve pages from its index with its usual ranking systems, which is why Google treats that part as SEO.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search}

Google describes one piece of that process. AI Overviews and Google AI Mode may use a technique called query fan-out: they run several related searches across subtopics and data sources to build the answer, which is why they can show a wider and more varied set of links than a classic search.⁠^{[3]Google Search Central. AI features and your website} Its guide to optimizing for these features describes it as a set of simultaneous, related queries the model generates to ask for more information and bring back more results.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search} For a store, the question stops being only "what position am I in?" and becomes "which sources am I among?" as well. What you should not do is answer it with one page per possible query: Google warns that creating separate content for each search variation, fan-out queries included, mainly to manipulate rankings or generative AI answers in Search, violates its spam policy on scaled content abuse.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search}

The paper that named it

The term comes from a research paper by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, published on arXiv on November 16, 2023 and accepted at the KDD 2024 conference. In it they present Generative Engine Optimization as a new framework for content creators to improve their visibility in generative engine responses.⁠^{[1]arXiv (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande; KDD 2024). GEO: Generative Engine Optimization}

To measure it they built GEO-bench, a benchmark of 10,000 queries across 25 domains, 80% of them informational and the rest split evenly between transactional and navigational. They tried nine ways of rewriting a source: adding quotations, adding statistics, citing sources, writing with more authority, improving fluency, using technical terms, using unique words, making it easier to understand and the classic keyword stuffing. And they measured two things: what share of the words in the answer sit in sentences that cite the source, weighted so that sentences appearing earlier count for more, and a subjective impression of how much weight the source carries in it.⁠^{[1]arXiv (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande; KDD 2024). GEO: Generative Engine Optimization}

The results, on that position-adjusted word metric: with the text untouched, the source scored 19.3; with quotations it reached 27.2, with statistics 25.2 and citing sources 24.6. Keyword stuffing dropped to 17.7, below doing nothing. That is where the figure that circulates most comes from, that GEO can boost visibility "by up to 40%". And a less quoted finding that matters more for a small store: lower-ranked pages gained more. Citing sources improved the visibility of the fifth result by 115.1% and worsened the first one's by 30.3%.⁠^{[1]arXiv (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande; KDD 2024). GEO: Generative Engine Optimization}

Read those numbers for what they are. The engine was simulated: GPT-3.5 Turbo composing an answer from the full text of Google's top five results. The separate check, on 200 queries in Perplexity, supplied the sources' text as uploaded files, because Perplexity does not let you specify source URLs. So neither experiment measured whether a page makes it into what is retrieved: both started from given sources. These are 2023 and 2024 engines, most of the queries were informational, and the authors themselves warn that the methods may need to adapt as engines evolve, just as SEO has kept changing.⁠^{[1]arXiv (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande; KDD 2024). GEO: Generative Engine Optimization} A 40% in one experiment is not a forecast for your catalog.

GEO in SEO, according to Google

Google has its own guide on how to optimize for its generative AI features, which it added to its documentation on May 15, 2026.⁠^{[4]Google Search Central. Latest documentation updates} The guide names AEO (Answer Engine Optimization) and GEO, and answers plainly: from the standpoint of Google Search, optimizing for generative search is optimizing for the search experience, and so it is still SEO.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search} The specifics:

• You do not need to create machine-readable files, AI text files, markup or Markdown to appear in Google Search, generative features included.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search} In June 2026 it added that llms.txt files are not needed for Search and affect your visibility neither for better nor for worse.⁠^{[4]Google Search Central. Latest documentation updates}
• Structured data is still good SEO: the guide advises you to keep using it as part of your SEO strategy, because it helps you qualify for rich results in Search. What it makes clear is that there is no special schema.org markup for generative search.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search} Google uses the structured data it finds to understand the content of the page, and among the cases it publishes, Rotten Tomatoes measured a 25% higher click-through rate on pages with structured data than on pages without it.⁠^{[5]Google Search Central. Introduction to structured data markup in Google Search}
• You do not need to chop content into small pieces for AI to understand it better.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search}
• Hunting for inauthentic "mentions" around the web does not help as much as it seems. What it does ask for is content that is not interchangeable: expert opinion or first-hand experience that goes beyond what anyone knows.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search}
• To appear as a supporting link in AI Overviews or AI Mode, the page has to be indexed and eligible to be shown in Search with a snippet.⁠^{[3]Google Search Central. AI features and your website} On top of meeting Search's technical requirements, the site has to be included in the generative AI features of Search from Search Console.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search}

Bing reports citations; Google reports impressions. Since February 10, 2026, Bing Webmaster Tools has had a public preview report, AI Performance, that counts how many times your content is cited as a source in Copilot, in Bing's AI-generated summaries and in some partner integrations, with the cited pages and the grounding queries the AI used to retrieve them.⁠^{[6]Microsoft, Webmaster Tools blog. Introducing AI Performance in Bing Webmaster Tools Public Preview} The Search Console generative AI report, by contrast, gives impressions in AI Overviews and AI Mode.⁠^{[7]Search Console Help. Generative AI performance report (Search)} Bing's recommendations for publishers are those of a good reference page: clear headings, tables and FAQ sections, claims backed by examples, data and sources, fresh content and telling it about every change with IndexNow.⁠^{[6]Microsoft, Webmaster Tools blog. Introducing AI Performance in Bing Webmaster Tools Public Preview}

Who reads your site to answer

Each company uses different agents to search, to train its models and to serve one person's specific request, and in robots.txt each one is allowed or blocked separately. OpenAI gives the example itself: you can allow OAI-SearchBot so you appear in ChatGPT search results and block GPTBot so what it crawls is not used for training.⁠^{[8]OpenAI. Overview of OpenAI Crawlers}

Google has no separate row. For AI Overviews and AI Mode, the page has to be in the Search index, and what is shown is limited with the usual controls: nosnippet, data-nosnippet, max-snippet or noindex. Google-Extended, the name that tends to come up in these conversations, serves according to Google to limit training and grounding "in some of Google's other systems."⁠^{[3]Google Search Central. AI features and your website}

In a store

What we measured ourselves. In June 2026, Clione enriched 39 product pages for BenroUSA, Benro's photography gear store in the United States: the JSON-LD graph, the title, the meta description and, on nine of them, the FAQs. In Search Console, from June 13 to 22 and against their starting point before enrichment, those pages gained 13% in impressions while the whole site gained 8%, and the nine with FAQs gained 25%, most of them improving their position. It is a correlation measured on one store over ten days, not a forecast for your catalog, and it points the same way as the cases Google publishes.

How to check it

1. Open your robots.txt and see which agents you let through. For answers with search, OAI-SearchBot, PerplexityBot and Claude-SearchBot count; GPTBot and ClaudeBot are the training ones.⁠^{[8]OpenAI. Overview of OpenAI Crawlers}⁠^{[9]Perplexity. Perplexity Crawlers}⁠^{[10]Anthropic. Does Anthropic crawl data from the web, and how can site owners block the crawler?} ChatGPT-User, Perplexity-User and Claude-User are the agents for visits a person requests, and Anthropic warns that blocking Claude-User may reduce your visibility in user-directed searches.⁠^{[10]Anthropic. Does Anthropic crawl data from the web, and how can site owners block the crawler?} In OpenAI's systems, a change to robots.txt can take about 24 hours to show.⁠^{[8]OpenAI. Overview of OpenAI Crawlers}
2. Confirm in Search Console that your product pages are indexed, that nothing stops them from showing a snippet and that the site is included in the generative AI features of Search, a condition for appearing in them.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search}⁠^{[7]Search Console Help. Generative AI performance report (Search)} A nosnippet or a very short max-snippet also cuts down what AI Overviews and AI Mode can use.⁠^{[3]Google Search Central. AI features and your website}
3. Open the Search Console generative AI performance report. It shows impressions in AI Overviews and AI Mode, without clicks or queries. It has been available for all sites since August 31, 2026, and it only appears if your site accumulates enough impressions in those features.⁠^{[7]Search Console Help. Generative AI performance report (Search)}
4. If your site is verified in Bing Webmaster Tools, look at AI Performance: the citations by URL and the grounding queries that brought them in.⁠^{[6]Microsoft, Webmaster Tools blog. Introducing AI Performance in Bing Webmaster Tools Public Preview}
5. Repeat a purchase query. Pick a question your customer would ask, without your brand, run it several times on different days in each assistant, always in a new chat, and write down which sources it cites each time. A single screenshot is a sample of one, not a position.

Related concept: AEO

The closest term is AEO (Answer Engine Optimization). Google names them together and defines them the same way: both describe the work of improving visibility in AI search experiences.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search} What GEO has of its own is an academic origin, with a benchmark and visibility metrics designed for composed answers, not for lists of links.⁠^{[1]arXiv (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande; KDD 2024). GEO: Generative Engine Optimization} And Bing already uses the word in its tools: it presents AI Performance as a first step toward GEO tools in Bing Webmaster Tools.⁠^{[6]Microsoft, Webmaster Tools blog. Introducing AI Performance in Bing Webmaster Tools Public Preview}

When you get an AEO or GEO proposal, ask what it will do and how it will measure it. Google points those third-party services to its guide on evaluating outside SEO advice, asks you to be wary of tools that promise rankings or claim to use internal Google metrics, and says it without hedging: no third-party tool has access to its internal ranking or AI systems.⁠^{[2]Google Search Central. Optimizing your website for generative AI features on Google Search}

To go further, the comparison SEO vs GEO and AEO: what changes for an online store puts the three side by side, the guide to what AI bots see in your store goes down to the detail of each agent, and the AI search reading path orders the rest of the terms.