Short answer
Generative engine optimization (GEO) is the work of making your pages the sources AI assistants such as ChatGPT, Perplexity and Google’s AI Overviews retrieve and cite when they write an answer. Assistants split a question into several searches, pull candidate passages from search indexes, and write from the passages they can quote and trust. GEO covers each step: crawl access, content that answers the sub-questions, clear entities, and mentions on the sites the web already trusts.
Key takeaways
- GEO and AEO (answer engine optimization) chase one result, a brand cited in AI answers. GEO starts from how the engines retrieve and assemble their sources.
- Google, OpenAI and Perplexity all retrieve from search before they answer, so a page has to be crawled and indexed to be cited.
- One question becomes many searches. In Ahrefs’ 2026 data, 38% of the pages cited in AI Overviews ranked in the top 10 for the query itself.
- In the original GEO research, adding citations, quotations and statistics raised a source’s visibility by 30 to 40%, and keyword stuffing did worse than no change.
- Topical authority, entity coverage and mentions on other sites decide which retrieved sources an assistant trusts enough to quote.
What generative engine optimization is
Generative engine optimization (GEO) is the practice of getting a brand’s pages retrieved, quoted and cited by the AI systems that write answers instead of listing links. The name comes from a 2023 research paper by Pranjal Aggarwal and colleagues, accepted at KDD 2024, which calls these systems generative engines: they search for information, then use a language model to write one answer from several sources [1].
GEO and AEO name one goal from two sides. AEO starts from the brand: its pages, its entity and the answers it gives, as I explain in what is answer engine optimization (AEO). GEO starts from the engine: how it breaks a question down, where it retrieves from and which passages it keeps. I run them as one program on an SEO foundation, and I compare all three in AEO vs GEO vs SEO.
How AI assistants choose their sources
The engines differ in detail, and their own documentation describes the same four stages:
- They break the question down. Google says AI Overviews and AI Mode may use a query fan-out technique, issuing multiple related searches across subtopics and data sources to develop a response [2]. OpenAI says ChatGPT search typically rewrites a prompt into one or more targeted queries for the search providers it works with [3].
- They retrieve candidates. Those searches run against search indexes: Google’s own, the providers ChatGPT works with, and the pages each engine’s crawler has fetched. OpenAI’s OAI-SearchBot and Perplexity’s PerplexityBot exist to surface websites in their search results [4] [5].
- They select passages. From everything retrieved, the model keeps a short list of passages that answer a sub-question directly and that it can attribute to a source.
- They write and cite. The answer links to the sources it used: Google shows them as supporting links [2], and ChatGPT answers that use search may include citations [3].
Two consequences follow. A page has to be crawled and indexed before it can be a source, so SEO is the entry ticket. Google’s requirements for AI Overviews and AI Mode are a page that is indexed and eligible to show a snippet [2], on a site left included in Search Console’s Search generative AI control, the default since the control reached every site on 31 August 2026 [7].
And a ranking for the head query covers less of the answer than it used to. In Ahrefs’ study of 863,000 search results and 4 million cited URLs, 38% of the pages cited in AI Overviews ranked in the top 10 for the query itself, down from about 76% in July 2025 [6]. The rest come from the searches behind the answer, the fan-out.
What the GEO research found
The GEO paper rewrote sources in nine different ways and measured how much of each generated answer a source earned, across GEO-bench, a benchmark of queries from many domains [1]. Three methods led: citing sources, adding quotations and adding statistics, with relative gains of 30 to 40% on the paper’s main visibility measure. On Perplexity, a live engine, the gains reached 37% [1].
Keyword stuffing, the old SEO habit of repeating the query’s words, did worse than leaving the page alone: on Perplexity it scored 10% below the baseline [1].
The finding that matters most for smaller brands is where the gains landed. The methods helped lower-ranked sources most: adding citations raised the visibility of sources ranked fifth in search by 115%, while the top-ranked source lost 30% on average [1]. Retrieval puts a page in the candidate set, and the writing decides how much of it the answer uses.
The tests ran on research systems and on Perplexity in 2023 and 2024, so I read the numbers as direction and test each change on a client’s own prompts.
What GEO work covers
In my programs, GEO runs in five layers, and each one feeds a stage of the pipeline above:
- Topical authority: a planned set of pages that covers a whole subject, the main questions and the fan-out questions behind them, organised as hubs and spokes with clear internal links, so search after search behind a prompt returns your site.
- Content strategy built for extraction: every page answers its question in its first lines, then adds what the research rewards, figures, quotations and cited sources [1].
- Entities: the brand, its products, its people and its category named the same way everywhere, with structured data that matches the page, so a model connects the brand to the topic.
- Mentions: the comparison pages, reviews, publications and videos the engines already retrieve, with the brand in them. In Ahrefs’ study, YouTube URLs made up 18.2% of the AI Overview citations that sat outside Google’s top 100 for the query [6].
- Access: OAI-SearchBot, PerplexityBot and Googlebot allowed in robots.txt and at the CDN and firewall, as OpenAI and Perplexity both recommend, with the pages indexed in Google and Bing [4] [5].
I find the gaps with tools I build: the sub-questions behind each priority prompt, the entities a model associates with the category, and the sources it cites in place of the brand. The GEO consulting page shows how that runs as a program.
How Google, ChatGPT and Perplexity differ
| Engine | Where answers come from | Crawler to allow | What its documentation adds |
|---|---|---|---|
| Google AI Overviews and AI Mode | Google’s index, through query fan-out [2] | Googlebot | The Search generative AI control in Search Console, on by default [7]; Google-Extended governs training and grounding in some of Google’s other systems [2] |
| ChatGPT search | Its search providers, Microsoft among them, and OpenAI’s crawler [3] [4] | OAI-SearchBot; GPTBot is the training crawler [4] | Robots.txt changes take about 24 hours to reach search [4] |
| Perplexity | Pages PerplexityBot crawls for its search results [5] | PerplexityBot, which does not crawl for model training [5] | Robots.txt changes can take up to 24 hours [5] |
The common thread is search. Each engine answers from pages it can crawl and retrieve, so one technical foundation serves all three, and the differences sit in which crawler to allow and where each engine looks. I cover ChatGPT step by step in how to get your brand cited by ChatGPT.
How to measure GEO
Measure GEO by engine, on the prompts your buyers use. Every month, run a fixed set of prompts in ChatGPT, Perplexity, Gemini and Google’s AI features, and record whether your brand is cited, which page is cited and which competitors appear in its place. Three numbers come out of that: share of voice, the proportion of answers that cite you; citations by page; and fan-out coverage, how many of the sub-questions behind your priority prompts your pages answer. Add AI referral sessions from analytics, and the program has a business number next to the visibility ones.
On one program for a leading news portal, share of voice in AI answers rose from 9% to 41% against its top five competitors, and more than 120 queries moved into cited positions through entity coverage.
A worked example: one question, three engines
The brand and numbers here are illustrative. A project management software company asks ChatGPT, Perplexity and AI Mode the same question: “What is the best project management tool for a construction firm?” Each engine names four or five products, and the company appears in none of the answers.
The cited sources show the same sub-questions in all three: price per user, offline access on site, integrations with accounting software, and reviews from construction teams. The company ranks well for “project management software”, and it has no page on offline access or on construction workflows.
That gives the program its first jobs: a page for each missing sub-question with the answer in its first lines, a comparison page with real prices, and outreach to the two review sites every engine cited. The same question is checked again next month, in all three engines.
Frequently asked questions
Is GEO different from SEO?
GEO builds on SEO. Assistants retrieve from search indexes, so the same crawling, indexing and authority work applies. GEO adds the fan-out questions behind a prompt, passages written to be quoted, and tracking inside AI answers next to rankings.
Is GEO the same as AEO?
They share the goal, a brand cited in AI answers, and differ in where they start: AEO works on the brand’s own pages and entity, GEO works on the queries and sources an engine assembles an answer from. I run them as one program because the fixes land on the same mechanism.
Does Google need special markup for AI Overviews?
No. Google says no special markup or optimizations are needed for AI Overviews or AI Mode: a page that is indexed and eligible to show a snippet can appear as a supporting link, on a site left included in Search Console’s Search generative AI control, the default.
How long does GEO take?
Access and page fixes can show within weeks, because ChatGPT and Perplexity pick up robots.txt changes within about a day and search the live web for many answers. Topical authority and mentions on other sites take months, because they depend on new pages and on other publishers.
Which engines should GEO cover first?
The ones your buyers use. For most brands that means Google’s AI features, ChatGPT and Perplexity first, then Gemini and Copilot. Track them all on one prompt set, so you see where each engine differs.
Sources
- Pranjal Aggarwal and others, “GEO: Generative Engine Optimization”, arXiv, 16 November 2023, revised 28 June 2024 (accepted to KDD 2024). arxiv.org
- Google Search Central, “AI features and your website”, last updated 10 December 2025. developers.google.com
- OpenAI Help Center, “Searching the web with ChatGPT”, updated October 2026. help.openai.com
- OpenAI, “Overview of OpenAI Crawlers”, read 28 October 2026. developers.openai.com
- Perplexity, “Perplexity Crawlers”, read 28 October 2026. docs.perplexity.ai
- Ahrefs, “Update: 38% of AI Overview Citations Pull From The Top 10”, 2 March 2026. ahrefs.com
- Search Console Help, “Search generative AI control”, read 28 October 2026. support.google.com