Share of answer is the operating metric for this practice.

How we count it
Request an audit

Insight · 10 min

SEO, AEO, and GEO, defined and compared

SEO, AEO, and GEO differ in what the engine hands the user. Search engine optimization (SEO) competes for a ranked link, answer engine optimization (AEO) competes to have a passage lifted into a snippet or AI Overview, and generative engine optimization (GEO) competes to be cited as a source in an answer a language model writes. The inputs overlap heavily. The selection mechanism, the unit of win, and the metric do not.

In short

  • SEO optimizes for a list of ranked URLs. The win is a position and a click.
  • AEO optimizes for a passage that an engine extracts and shows directly. The win is the quoted text, and whether it is accurate.
  • GEO optimizes for inclusion in a model-written answer that retrieves sources and cites some of them. The win is a citation, ideally with a link.
  • GEO is the only one of the three with an academic origin: a 2023 paper by researchers at Princeton University and IIT Delhi, with independent co-authors.
  • All three depend on the same precondition: a page that crawlers can fetch and the engine has indexed.

What is SEO?

SEO is the practice of making pages crawlable, indexable, and relevant enough to rank in a search engine’s results for the queries people use. Google describes its own pipeline in three stages: crawling, where automated programs download pages; indexing, where Google analyzes the content and stores it; and serving, where it returns the results it judges relevant to a query.

Ranking is the selection step. Google says relevance is determined by hundreds of factors, which can include the user’s location, language, and device, and that it does not accept payment to rank pages higher. The output is an ordered list. The user still chooses which result to open.

What is AEO?

AEO is the practice of shaping a page so that an engine can take one passage from it and present that passage as the answer. The classic surfaces are Google featured snippets and People Also Ask. Google’s AI Overviews and AI Mode extend the same idea with a generated summary and supporting links.

The term has no single founding paper. It grew out of SEO practice as answers began to appear on the results page itself. Google introduced featured snippets in January 2014, according to a 2018 post from its Search Liaison, Danny Sullivan. AI Overviews began rolling out to everyone in the U.S. on May 14, 2024.

Two facts from Google’s documentation define the discipline. First, Google’s systems decide whether a page makes a good featured snippet. There is no markup that designates one. Second, there are no additional requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible to show in Search with a snippet, and no special schema.org structured data is needed. AEO is therefore not a separate technical channel. It is the craft of writing the passage that wins an extraction the engine was going to perform anyway.

What is GEO?

GEO is the practice of increasing how visibly a source appears in answers composed by generative engines such as ChatGPT search, Perplexity, Gemini, Claude, and Microsoft Copilot. The engine retrieves documents, a language model writes an answer from them, and some of those documents are attached as citations. Which ones, and how to become one, is the subject of how to get cited by ChatGPT and Perplexity; much of the weight sits off your own site.

Where the term came from

The term was coined in “GEO: Generative Engine Optimization” by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande. It was first posted to arXiv on November 16, 2023 (2311.09735) and accepted to KDD 2024. The paper formalizes “generative engines” as systems that synthesize information from multiple sources, and frames GEO as a black-box optimization problem for the people who publish those sources.

What the paper actually found

The paper’s engine was a research setup, not a commercial product. For each query it fetched the top five Google results and had gpt-3.5-turbo write an answer citing them. The authors then rewrote one source with nine methods and measured how much more visible that source became.

  • What they measured. Two visibility metrics: Position-Adjusted Word Count, the share of the answer’s words attributed to the source, discounted by citation position; and Subjective Impression, a model-graded score covering relevance, influence, uniqueness, position, count, click likelihood, and diversity.
  • The benchmark. GEO-bench, 10,000 queries drawn from nine sources across multiple domains.
  • What worked. Cite Sources, Quotation Addition, and Statistics Addition performed best, which is where the headline “up to 40%” visibility boost comes from. Fluency and readability rewrites helped less.
  • What did not. Keyword Stuffing, the move the authors associate with classic SEO, offered little to no improvement and scored below the unoptimized baseline on Position-Adjusted Word Count.
  • Rank was not destiny. For the source ranked fifth in search, Cite Sources raised visibility by 115.1%. For the source ranked first, the same method lowered it by 30.3%.
  • On a live engine. On a 200-query subset run through Perplexity.ai, gains were smaller but present, with improvements up to 37%.
  • Domain mattered. Effectiveness varied by domain, and the authors argue for domain-specific methods.

Read the paper as evidence about mechanism, not as a ranking-factor list. It tested one model on five fetched sources in 2023. Commercial engines have since changed their retrieval, models, and citation interfaces. The durable finding is that a generative engine rewards verifiable substance in the passage, and that a lower-ranked source can still win the citation.

How does each engine choose what to show?

Each engine runs a different selection step on top of a shared index of crawled pages: ranking orders documents, extraction picks a passage, and retrieval-augmented generation writes new text and attributes parts of it.

Ranking

A search engine scores documents against the query and returns an ordered list of links with titles and snippets. Selection happens at the level of the URL. Two pages that say the same thing compete for one position.

Passage extraction

An answer engine takes a query, identifies a candidate page, and lifts a specific span of text. For a featured snippet, that span is quoted with a link to its page. For AI Overviews and AI Mode, Google says its systems may use a “query fan-out” technique, issuing multiple related searches across subtopics and data sources, and then display a wider and more diverse set of supporting links than a classic result. Selection happens at the level of the passage. A page that ranks well but buries its answer can lose the extraction. How AI Overviews choose a source covers what fan-out means for which pages get cited.

Retrieval-augmented generation

A generative engine retrieves documents, places them in the model’s context, and generates an answer conditioned on them. The pattern was named in “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks” by Patrick Lewis and colleagues (arXiv, May 2020; NeurIPS 2020), which paired a language model with a retriever over an index. Commercial engines add an attribution layer on top. Anthropic’s documentation for Claude’s web search tool, for example, states that citations are always enabled for web search, and each citation carries the URL, title, and up to 150 characters of cited text. Selection happens at the level of the claim. The model may paraphrase a source, merge it with others, or use it without linking it.

Who fetches the page

Each kind of engine has its own crawlers, and as of November 2026 the major vendors separate training crawlers from search crawlers:

VendorSearch or answer crawlerTraining crawler or tokenUser-triggered fetcher
OpenAIOAI-SearchBot, used to surface sites in ChatGPT searchGPTBotChatGPT-User
AnthropicClaude-SearchBotClaudeBotClaude-User
PerplexityPerplexityBot, which Perplexity says is not used for foundation-model trainingNone listedPerplexity-User
GoogleGooglebot (Search, including AI features)Google-Extended, a robots.txt token for Gemini training and groundingListed separately by Google

Google states that Google-Extended does not affect a site’s inclusion in Google Search and is not a ranking signal. Perplexity says Perplexity-User generally ignores robots.txt because a person requested the fetch, and OpenAI says robots.txt rules may not apply to ChatGPT-User for the same reason. The access decisions behind this table are covered in robots.txt for AI crawlers.

What does each one measure?

SEO measures position and clicks, AEO measures presence and accuracy of an extracted passage, and GEO measures mentions, citations, and links inside generated answers.

SEOAEOGEO
Engine outputRanked list of linksExtracted passage or AI summary with supporting linksModel-written answer with cited sources
Unit of selectionURLPassageClaim
Unit of winA position and a clickYour passage is shown, and it is rightYou are named as a source, ideally with a link
Selection mechanismRankingPassage extractionRetrieval-augmented generation
Typical surfacesGoogle and Bing resultsFeatured snippets, People Also Ask, AI Overviews, AI ModeChatGPT search, Perplexity, Gemini, Claude, Copilot
Core metricsRank, impressions, clicks, CTRSnippet or overview presence, passage accuracy, cited URLMention rate, citation rate, link rate, share of answer
First-party data, as of November 2026Google Search Console, Bing Webmaster ToolsSearch Console counts AI feature traffic within the Web search typeBing Webmaster Tools AI Performance, for Copilot and Bing AI summaries
Origin of the termIndustry practiceIndustry practiceAggarwal et al., arXiv 2311.09735 (2023)

The first-party data is uneven. Google’s documentation says sites appearing in AI features are included in overall search traffic in Search Console, reported under the Web search type. Microsoft launched AI Performance in Bing Webmaster Tools as a public preview on February 10, 2026, with total citations, average cited pages, a sample of grounding queries, and page-level citation counts across Copilot, AI-generated summaries in Bing, and select partner integrations. For the other generative engines, measurement means running a fixed panel of queries and logging the answers and sources yourself. The method is laid out in how to measure AI visibility, and tying the result to revenue is covered in attributing AI search to pipeline.

Where do SEO, AEO, and GEO overlap?

They overlap at the foundation: every surface reads pages that were crawled and indexed. Google says the underpinnings of what it has long advised carry across to its AI experiences, and its own requirement for AI features is simply an indexed page that is eligible for a snippet. A page blocked by robots.txt, set to noindex, or rendered in a way the crawler cannot read is absent from all three.

They also share a preference for checkable content. Snippets favor a passage that answers the question directly. The GEO paper found that adding citations, quotations, and statistics raised visibility. Google asks that structured data match content visible on the page. The common thread is a specific claim in plain HTML, which is the subject of what gets cited.

Where do they conflict?

They conflict in four places, and each is a trade-off a site has to decide rather than optimize away.

  1. Visibility versus the click. A ranked link sends a visitor. An extracted passage or a generated answer may satisfy the query on the page. Google says clicks from AI Overviews are higher quality, but it does not publish how many fewer there are, so a win on one surface can read as a loss on another.
  2. Preview controls. nosnippet, data-nosnippet, max-snippet, and noindex limit how content appears in Google’s AI experiences as well as in classic snippets. A control set to protect a passage from extraction also removes it from AEO and from Google’s AI features.
  3. Crawler policy. Blocking a training crawler and blocking a search crawler are separate decisions with separate tokens. A blanket block written to stay out of training data can remove a site from a generative engine’s retrieval. Allowing everything is also a decision, with its own consequences for how content is reused.
  4. Rank versus citation. In the GEO paper, the methods that lifted the fifth-ranked source lowered the first-ranked one. Ranking first in search did not guarantee the largest share of the answer (commercial engines widen the gap through query fan-out), and keyword density, a traditional on-page lever, did not help.

The case for running all three from one query list is made separately in SEO, AEO, and GEO are one system, which also answers whether GEO is just SEO. This guide is narrower: the three are different selection problems, and they should be reported as different numbers.

Sources

Bring us the domain.

We will tell you where you are absent, which page is in the way, and what has to be true before a citation is possible.

Request an audit