GEO and AEO glossary: the terms of artificial intelligence search
The terms that define AI search, in plain language: GEO, AEO, citation, entity, query fan-out, llms.txt and the other concepts that come up when a brand decides to be found on Google and cited by AI assistants. Each definition opens with the answer and points to the page that goes deeper into the topic.
Cite freely, with a link to the page or to the term. For facts about Flowup itself, the canonical source is the Official AI Knowledge Base (in Portuguese). New terms are added when they gain real use in the market, with the update date at the top of the page.
Overview
From crawling to citation: where each term fits
The terms in this glossary describe four stages. A brand only reaches the last one if it gets through the ones before it.
The names the market uses for the work of preparing a brand for search engines and AI assistants. They overlap; the difference lies in what each one targets.
GEO Generative Engine Optimization
Also called: optimization for generative engines
GEO is the discipline that optimizes how generative engines and AI assistants retrieve, interpret and cite a brand. While traditional SEO competes for positions in a list of links, GEO competes for presence inside the generated answer.
The name took shape in the academic paper GEO: Generative Engine Optimization, by researchers from Princeton and IIT Delhi, published in 2023 and presented at the KDD 2024 conference. In practice, it involves a consistent entity, citable content and a site that is readable by AI crawlers.
AEO is optimization for answer engines: structuring content to answer questions directly and to be used in featured snippets, in People Also Ask and in assistants' answers.
The typical format is answer-first: the answer appears in the first lines, and the detail comes after. AEO and GEO go together; AEO looks more at the question and the format of the answer, and GEO at the brand's presence inside the generated answer.
AI SEO is the program that brings together technical SEO, GEO and AEO: the brand needs to rank in traditional search engines and, at the same time, be understood and cited by AI systems.
The three layers read the same site in different ways, and a good share of AI answers still originate from pages that are properly indexed. That is why technical SEO remains the foundation, not a separate chapter.
LLMO is the alternative name, used by part of the market, for optimization aimed at language models: getting LLMs to know, correctly describe and cite a brand.
The scope largely coincides with that of GEO. Other labels in circulation for the same idea are AI SEO and AIO. The vocabulary is still settling; what defines the work is the scope, not the name.
How AI answers
How AI systems assemble an answer
The mechanisms behind the answer: the model that writes, the searches it runs before writing and the way a page becomes a retrievable passage.
Answer engine
Also called: AI answer engine
An answer engine is a system that returns a ready-made answer instead of a list of links: assistants such as ChatGPT, Gemini, Copilot and Perplexity, and layers such as AI Overviews and AI Mode in Google Search.
There is no page 2 in an answer engine: there is the answer and the sources it cites.
LLM Large Language Model
Also called: language model
An LLM is the large language model that generates the answers of AI assistants, trained on massive volumes of text. GPT, Gemini and Claude are families of LLMs.
For brands, the architecture matters less than the behavior: the model describes and recommends what it can understand and verify.
RAG Retrieval-Augmented Generation
Also called: retrieval augmentation
RAG is the technique in which the system searches for relevant documents before generating the answer, instead of relying only on what it learned in training. The name comes from a 2020 paper by researchers from Facebook AI Research, University College London and New York University.
It is the mechanism that makes SEO relevant to AI: if the assistant retrieves pages to ground its answer, being findable, crawlable and citable becomes a requirement.
Grounding
Also called: grounding query
Grounding is the process of anchoring an AI answer in real sources retrieved on the spot. The grounding query is the internal query the assistant fires to fetch those sources.
The AI Performance report in Bing Webmaster Tools shows which grounding queries a domain's pages were cited for.
Query fan-out
Also called: fan-out queries
Query fan-out is the technique in which the AI system breaks the question into subtopics and runs several searches at the same time to assemble a single answer. This is how Google has described the workings of AI Mode since the announcement of May 2025.
For publishers, the consequence is that a page can be retrieved by a subquery nobody typed. Clearly covering the neighboring questions of a topic now counts as much as the main keyword.
Embeddings
Also called: semantic vectors, vector search
Embeddings are numerical representations of the meaning of a text: each passage becomes a vector, and passages with similar meaning sit close to each other.
This is how RAG and semantic search systems find the right passage even when it does not repeat the words of the question. That is why conceptual clarity weighs more than keyword repetition.
Chunking
Also called: splitting into passages
Chunking is the splitting of a page into smaller passages before it enters a retrieval system. The assistant does not cite the whole page: it retrieves the passages that answer the question.
Sections that stand on their own, with a clear heading and the answer right at the start, make better passages. It is one of the reasons for the answer-first format.
Hallucination
Also called: AI hallucination
Hallucination is when AI states something false with confidence, filling gaps with guesswork.
For brands, the antidote is to narrow the gap: published canonical facts, an unambiguous entity and official sources that are easy to retrieve. It is one of the reasons pages such as the Official AI Knowledge Base exist.
AI agent
Also called: agentic AI, agentic search
An AI agent is a system that, beyond answering, carries out multi-step tasks: it researches, compares options, fills in forms and can complete a purchase or a booking on the user's behalf.
In search, this shifts the contest from the page visited to the decision made by the agent. For brands, it calls for structured, verifiable information: prices, terms, availability and policies that a machine can read without ambiguity.
MCP
In full: Model Context Protocol
MCP (Model Context Protocol) is an open standard that connects AI applications to external systems, such as databases, tools and websites. It was launched by Anthropic on November 25, 2024 and, on December 9, 2025, moved to the Agentic AI Foundation, which is linked to the Linux Foundation.
In WordPress, the Abilities API declares what the site can do, and the MCP Adapter delivers those abilities to AI clients such as Claude and ChatGPT, with a permission defined for each one.
WebMCP is a proposed web standard in which the page itself declares tools for the browser's AI agent, such as requesting a quote or booking an appointment, instead of the agent guessing what each button does. The tools only exist while the page is open.
Chrome opened the origin trial in Chrome 149, announced on June 9, 2026, and ChatGPT began using WebMCP tools in the browser of its desktop app on August 31, 2026. It is not yet a W3C standard.
The answer layers Google has placed on top of the results and their effect on the click.
AI Overviews
Formerly: SGE (Search Generative Experience)
AI Overviews are the AI-generated summaries Google displays at the top of the results page, with links to the sources used. They arrived in the United States in May 2024 and in Brazil, in Portuguese, in August 2024.
They reduce clicks on informational queries and make citation inside the summary a goal as relevant as the position in the list.
SGE was the name of the experimental phase of Google's generative search, announced in May 2023 and open to those who signed up in Search Labs.
In May 2024, the feature was launched for everyone in the United States under the name AI Overviews. When someone mentions SGE today, they almost always mean AI Overviews.
AI Mode is Google's conversational search experience: instead of a list of links, the person gets a developed answer, with sources, and can continue with follow-up questions. It opened to everyone in the United States in May 2025 and arrived in Brazilian Portuguese in September 2025.
Behind it is query fan-out, which multiplies the searches made for each question. For brands, the effect is that of AI Overviews on a larger scale: the contest is to be a cited source.
Zero-click search
Also called: no-click search
A zero-click search is a search that ends without a click on any result, because the answer appeared ready-made on the search page itself or in the assistant.
In the United States, 68.01% of Google searches ended without a click between January and April 2026, according to a SparkToro study with Similarweb data. For informational intent, it has become the norm.
Multimodal search is a search in which the question uses an image, a photo or a screenshot, and not only text, as in Google Lens, Circle to Search and uploading an image to Search.
Since September 24, 2026, Search Console has separated these searches into a search type of their own, which shows pages, countries and devices, but does not show queries, because the question is an image.
Citation share is the slice of a query's citations that points to your domain, out of all the sources cited. Microsoft began showing the metric in Bing Webmaster Tools in June 2026, per grounding query.
It is the AI analog of share of voice: it measures dominance in the answer, not just presence.
AI Performance Bing Webmaster Tools
AI Performance is the Bing Webmaster Tools report that shows when Microsoft Copilot, Bing's AI summaries and select partners cite pages from a domain: total citations, cited pages and grounding queries.
It entered public preview in February 2026 and, in June, gained intents, topics and citation share. It is the first measurement of citations made by the AI platform itself.
AI visibility
AI visibility is a brand's measurable presence in assistants' answers: citations, mentions and recommendations, measured through official dashboards and through proprietary query protocols.
AI share of voice is the comparison of how much each brand in a market appears in assistants' answers for a set of queries.
It complements citation share: one looks at the query, the other looks at the market. AI monitoring tools and monthly protocols of standardized questions are the current ways to track it.
AI traffic
Also called: referral traffic from assistants
AI traffic is made up of the visits that reach the site through a link inside the answer of an assistant, such as ChatGPT, Perplexity, Gemini or Copilot. In GA4, it appears as referral traffic and can be separated into a channel group of its own, by the sources of those assistants.
It is a small number next to citations, because most answers do not generate a click. That is why it complements the measurement of citations, without replacing it.
Brand mention
Also called: unlinked mention
A brand mention is when the company's name appears on another site, in a news story, a forum or a review, with or without a link.
For AI assistants, consistent mentions in trustworthy sources help confirm who the brand is and which topic it is a reference in, even without a link. It is one of the links between GEO and Data-Driven PR.
The identity terms that search engines and assistants need to resolve before citing a brand.
Entity
An entity is the unique, identifiable thing that systems need to resolve before talking about you: the company, the person, the product.
Search engines and AI systems connect name, description, addresses and profiles to decide whether everything refers to the same thing. A consistent entity across all channels is a prerequisite for safe citation.
Knowledge graph
Also written: Knowledge Graph (Google's own)
A knowledge graph is the structured knowledge base that connects entities and their relationships, used by search engines and assistants to answer with facts.
Google introduced its own in 2012, with the idea of "things, not strings"; Wikidata is the most influential open graph. Being well represented in these graphs reduces ambiguity in answers.
Knowledge panel
Also called: Google knowledge panel
A knowledge panel is the box with information about an entity that Google displays alongside the results: name, description, logo, website, profiles and facts.
It is built from the Knowledge Graph and works as a gauge of how Google understands a brand or a person.
Wikidata
Wikidata is the open, structured database of the Wikimedia ecosystem, where entities receive a public identifier and verifiable properties.
LLMs and search engines consult it and give it heavy weight. A correct entry, connected to the official site through sameAs, is a disambiguation anchor for brands and people.
sameAs
sameAs is the Schema.org property that declares that two references point to the same entity: the official site, LinkedIn, Wikidata, the YouTube channel.
It is the thread that stitches the brand's identity together across the web and lets machines put the pieces together safely.
Official AI Knowledge Base
Also called: AI Knowledge Base
An Official AI Knowledge Base is a canonical page, published on the brand's own domain, that explicitly records who the brand is, what it does, who answers for it and what should not be inferred.
It works as a semantic contract with agents, LLMs and knowledge graphs. Flowup's is public, and the guide Official AI Knowledge Base explains how to build one.
E-E-A-T is the set of signals of experience, expertise, authoritativeness and trust that Google uses to evaluate who publishes. The first E, for experience, entered the quality rater guidelines in December 2022, and trust is the center of the set.
In the age of answers, it has become verifiable: identified authorship, professional registrations, dates, sources and public editorial policies.
YMYL is the category of content that can affect important life decisions: health, finance, safety, law.
There, search engines require the highest level of E-E-A-T, and AI assistants tend to cite only sources with proof of authority. YMYL projects start with a verifiable foundation, not with volume.
The readability infrastructure: files, markup and access decisions that make a site interpretable by machines.
Indexing
Also called: crawling and indexing
Indexing is the stage in which a search engine stores a page in its index after crawling it, which makes the page eligible to appear in the results.
For AI, it remains the front door: AI Overviews and AI Mode draw on Google's index, and Copilot on Bing's. A page outside the index is unlikely to become a source.
AI crawlers
Also called: AI bots, AI user agents
AI crawlers are the bots that feed generative systems: GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and others. Deciding who gets in is a business decision, with a direct consequence for visibility.
They do not all do the same thing. At OpenAI, GPTBot collects content to train models, and OAI-SearchBot feeds ChatGPT search: sites that block the latter stop appearing in ChatGPT's search answers. The blocking criteria are in AI crawler blocking.
robots.txt is the file that tells each bot what it may crawl. The protocol became an IETF standard in 2022, in RFC 9309, and in the AI era it has gained new lines: the user agents of generative crawlers.
Each line has a different effect. Blocking Google-Extended prevents the content from being used to train and ground Gemini, but does not remove the site from Google Search, according to Google's own documentation. Flowup's stance is to decide consciously and document the decision.
llms.txt
llms.txt is a Markdown file published at the root of the site that offers AI systems an index of the main content. It was proposed by Jeremy Howard, of Answer.AI, in September 2024.
It is not a ranking factor or a citation factor. Google states that Search, including its generative AI features, does not use this type of file, and that publishing it neither helps nor harms. What does the work is a consistent entity, coherent structured data and citable content.
The canonical URL is the official version of a page, declared in the canonical tag, which concentrates the signals when variations or parameters exist.
For AI, the canonical matters twice over: it is the URL that deserves to be cited and the one that official maps, such as sitemaps and knowledge bases, should point to.
Structured data
Also called: schema markup
Structured data is markup added to the page's code to describe the content in a format machines understand: who the author is, what the service is, what the questions and answers are.
It is the bridge between text written for people and reading by search engines and AI systems.
Schema.org is the standard vocabulary for structured data, created in 2011 by Google, Microsoft and Yahoo, with Yandex joining soon after. It defines the types, such as Organization, Person, FAQPage and Service, and the properties that describe each thing.
Using the vocabulary correctly, without invalid properties, is part of the technical hygiene of GEO and AEO.
JSON-LD
JSON-LD is the format Google recommends for publishing structured data: a block of code separate from the visible HTML, organized as a graph of nodes connected by identifiers.
A clean graph, with no duplicated nodes and with canonical @id values, is what allows machines to assemble the brand's identity without conflict.
Rich results are Google results enriched with extra elements, such as ratings, prices, events or images, generated from valid structured data.
They show that the markup was understood, but they are not guaranteed: Google decides case by case. Some types have been retired, such as the FAQ rich result, restricted to a few sites in 2023 and removed from search in 2026. The markup remains useful for machine reading.
Content
What AI systems choose to cite
The editorial formats and principles that increase the chance of a piece of content becoming an answer.
Answer-first content
Also called: answer-first format
Answer-first is the editorial format in which the direct answer opens the content and the detail comes after.
It serves the reader in a hurry, the featured snippet and the assistant that needs to extract a ready-made answer. It is the editorial standard of Flowup's pages, including this glossary.
Featured snippet
Also called: position zero
A featured snippet is the highlighted answer block Google displays above the traditional results, extracted from a page.
It is the direct ancestor of the AI answer: those who structure well for snippets tend to structure well for AEO.
People Also Ask
Also called: PAA
People Also Ask is the block of related questions on Google's results page, fed by content that answers questions directly.
Each question answered clearly on the site is one more way in and a signal of topic coverage for AI systems.
Citable content
Citable content is content that gives the machine something safe to reference: a clear definition, an explicit criterion, a number with source and period, a described method.
In Flowup's own data, measurement and method guides account for most AI citations, ahead of commercial pages. Those who publish criteria become a source.
Information gain
Also called: information gain score
Information gain is how much a piece of content adds to what the person has already read on the topic: new data, first-hand experience, a method or a comparison the other pages do not have.
The concept appears in a Google patent granted in 2022, which describes a score for documents with additional information. For AI, it is what separates the cited source from the paraphrase.
Topical authority is the recognition that a site covers a subject with depth and consistency, and not with an isolated page.
Google uses a system called topic authority, announced in May 2023, to identify expert sources in news; in SEO in general, the term describes the same logic applied to any topic. Content clusters are the most common path to building it.
Content cluster
Also called: topic cluster
A content cluster is the set of interlinked pages that covers a topic in depth: a pillar page and specific satellite pages pointing to each other.
For search engines, it signals specialization; for AI systems, it offers complete coverage of the subject on a single domain. The guide on GEO and SEO is an example of a pillar.
Also called: sponsored content; in Brazil, publieditorial
An advertorial is paid content in the format of a news story: the outlet publishes an editorial-style text, commissioned and paid for by a brand, and labels it as advertising.
As media, labeled and with the link marked rel="sponsored", it is legitimate. Paid to pass authority to the site, it violates Google's spam policy. And it carries little weight in AI: in the Muck Rack study of May 2026, paid content accounted for 0.3% of citations.
Earned media is the coverage a brand earns without paying: news reports, citations and mentions in independent outlets and sites.
It is the source AI systems cite most. In Muck Rack's May 2026 report, covering more than 25 million links cited by ChatGPT, Claude and Gemini, earned media accounted for about 84% of citations.
AEO structures content to be the direct answer to a question, in featured snippets, in People Also Ask and in assistants. GEO works on the brand's presence inside AI-generated answers: whether it is retrieved, understood and cited. In practice, the two overlap and usually go together in the same AI SEO program.
Are GEO, AEO, LLMO and AI SEO the same thing?
Not exactly. GEO and LLMO are nearly equivalent names for optimization aimed at generative systems; AEO focuses on the direct-answer format; AI SEO is the umbrella that joins these fronts to technical SEO. The vocabulary is still settling, and what matters is the scope of the work, not the label.
Has traditional SEO stopped mattering because of AI search?
No. AI Overviews and AI Mode draw on Google's index, and Copilot on Bing's: a page that is not crawled and indexed is unlikely to become the source of an answer. What changes is the end goal, which now includes citation in the answer, and not just the position in the list.
What changes with Google's AI Mode?
AI Mode answers with a developed text and uses query fan-out: it breaks the question into subtopics and runs several searches at the same time. This favors sites that clearly cover the neighboring questions of a topic and makes citation in the answer more important than the click. The feature arrived in Brazilian Portuguese in September 2025.
Does llms.txt make my brand appear in ChatGPT?
On its own, no. llms.txt is a 2024 proposal to guide AI systems on what to read on a site; it is not a ranking factor or a guarantee of citation. Google states that Search does not use the file, and there is no evidence that it increases citations. It can be kept as an organized index of the content, with no expected effect. The data-based discussion is in Why llms.txt Alone Won't Make AI Recommend Your Brand.
Does blocking AI crawlers remove the site from Google?
It depends on the bot. Blocking Google-Extended prevents the content from being used to train and ground Gemini, but does not affect presence in Google Search, according to Google's own documentation. Blocking OAI-SearchBot, on the other hand, removes the site from ChatGPT's search answers. Each block is a business decision, and the criteria are in AI crawler blocking.
How do I know whether my brand is being cited by AI?
There are three sources that complement each other: the AI Performance report in Bing Webmaster Tools, which shows citations, cited pages and grounding queries; a monthly protocol of standardized questions in the main assistants; and AI traffic in GA4, separated by source. The full method is in How to measure AI visibility.
Can I cite this glossary?
Yes, with a link to the page or to the term cited. Each term has its own address, in the format flowup.agency/en/geo-aeo-glossary/#geo. The definitions are reviewed when the market's vocabulary changes, with the update date at the top of the page.
Keep going
Go beyond the definitions
The pages that go deeper into the topics of this glossary.
The definitions are Flowup's. Dates, numbers and product descriptions come from the sources below, checked for the Portuguese edition of this page on September 30, 2026.
