GEO glossary: the terms of AI search visibility

This GEO glossary defines 34 terms that come up in AI search visibility work, from the acronyms GEO, AEO and LLMO to query fan-out, grounding and llms.txt. GEO, short for generative engine optimization, is the practice of getting a brand cited when people ask Perplexity, ChatGPT, Gemini or Google's AI features about its market. Each entry starts with a one-sentence definition you can quote without the rest.

Three of these terms, citation rate, average position and share of voice, are the numbers that Hypervisible, the GEO software we publish, reports for up to five AI engines, which is why this glossary defines them the way they are measured.

GEO, AEO, LLMO, SEO

The four acronyms overlap heavily. SEO targets a ranked list of links; GEO, AEO and LLMO target the answer an AI writes, and most teams use those three almost interchangeably.

GEO

GEO (generative engine optimization) is the practice of making a brand or a page more likely to be cited in answers written by AI engines such as ChatGPT, Gemini, Perplexity and Google AI Mode. The name comes from a paper by researchers at Princeton University and IIT Delhi, first posted in November 2023 and accepted at KDD 2024; in their experiments, some content changes raised visibility in generative engine answers by up to 40%. Our guide to generative engine optimization covers the method end to end.

AEO

AEO (answer engine optimization) is the practice of shaping content so an engine can lift it as the direct answer to a question. It puts format first: the question in plain words, a short answer right below, then the supporting detail. Start with our explainer on what AEO is, then read AEO vs GEO vs SEO for where the lines blur. Americans search “aeo vs geo” about 1,600 times a month (Google Ads estimate, October 2026).

LLMO

LLMO (large language model optimization) is the practice of getting large language models to know a brand, describe it accurately and name it when someone asks about its category. It frames the GEO goal around the models rather than the search products built on them, and Google Ads estimated about 480 US searches a month for it in October 2026. How to measure it is covered in LLM optimization.

SEO

SEO (search engine optimization) is the work of helping search engines understand your content and helping people find your site through search. That's close to how Google's own SEO Starter Guide puts it, and it remains the ground floor of GEO: Google's AI features only cite indexed pages that are allowed to show a snippet. Pages that can't be crawled don't get quoted.

AI engines and their search features

For a US brand in October 2026, the engines worth checking first are ChatGPT, Gemini, Perplexity and the two AI features inside Google Search. Each one answers in prose, and each can name brands.

Answer engine

An answer engine is a search tool that replies with a written answer built from sources it retrieves, instead of a list of links for you to sort through. The researchers who coined GEO call the same thing a generative engine. For a brand, the change is in what the buyer sees: a handful of names inside a sentence. Usage numbers are collected in AI search statistics for 2026, and AI search engines compared shows how each one cites brands.

AI Overviews

AI Overviews are the AI-written summaries Google places above regular results on some searches, with links to the pages behind them. Google started showing them to everyone in the US on May 14, 2024. A summary appears only when Google's systems judge that it helps, and plenty of queries never get one. Our AI Overviews guide traces where their links come from.

AI Mode

Google AI Mode is a conversational mode of Google Search that writes an answer with Gemini models, links to the pages it drew on, and keeps context for follow-up questions. The US-wide rollout began on May 20, 2025, and by May 19, 2026, Google was reporting more than one billion monthly users. People reach it at google.com/ai or through the AI Mode button after a search. The full picture is in our guide to Google AI Mode.

ChatGPT search

ChatGPT search is the feature that lets ChatGPT look things up on the web during a conversation and answer with links to the sources it used. OpenAI released it on October 31, 2024; from February 5, 2025, it was open to everyone, including people who weren't signed in. ChatGPT decides when a question needs fresh information, then rewrites it into one or more targeted queries for partner search providers. OpenAI's help center says it ranks results on several factors and promises no fixed placement. More in ChatGPT search: how it works.

Perplexity

Perplexity is an AI-powered search engine that searches the internet in real time and gives every answer numbered citations linking to the original sources. Because each claim carries a number, it's an easy engine to audit when you want to see which websites shape answers in your category. ChatGPT vs Gemini vs Perplexity compares how the three pick the brands they name.

Gemini

Gemini is both Google's family of AI models and its consumer assistant, the Gemini app, which was called Bard until February 8, 2024. Gemini models also run AI Mode, yet the app has its own interface and can recommend different brands than Google Search for the same question. Track them as two engines.

Zero-click search

A zero-click search ends with no click to any website, because the results page or the AI's text already holds the answer. Pew Research Center tracked 900 US adults in March 2025: they clicked a regular result on 8% of visits when an AI summary appeared, against 15% when none did. Being named starts to matter more than being linked. Our article on zero-click search looks at what organic traffic is left.

AI visibility is measured by putting the same questions to the same engines on a schedule and counting what comes back. Citation rate, average position and share of voice do most of the work.

Brand mention

A brand mention is any appearance of a brand's name in the text of an AI answer, linked or not. Mentions often rest on pages the brand doesn't own, like a review platform or a Reddit thread. Our guide to brand mentions in AI answers explains how to find and track them.

Source citation

A source citation is a link an AI engine attaches to its answer, pointing to a page that backs up what it says. Citations and mentions don't always line up: your buying guide can be cited as a source while a competitor gets the recommendation in the text. The domains that keep showing up as sources in your category make a ready-made to-do list, and how ChatGPT chooses its sources explains why that list shifts.

Citation rate

Citation rate is the share of AI answers that name your brand, out of a fixed set of questions. If Perplexity names a Denver HVAC contractor in 6 of 24 answers, that contractor's citation rate on Perplexity is 25%. The number only means something per engine and over time, on the same questions. For benchmarks, see what a good AI visibility score looks like.

Average position

Average position is the mean rank at which your brand appears in the AI answers that name it, where 1 means named first. Being first of five recommended firms and being fifth are different outcomes. It has nothing to do with average position in Google Search Console, which averages the topmost position of your links in classic results. An AI rank tracker follows it week to week.

AI share of voice is your brand's share of all the brand mentions in a set of AI answers, against the competitors you track. If you and four rivals collect 60 mentions and 18 are yours, your share of voice is 30%. The term comes from advertising. The formula and common mistakes are in AI share of voice.

Share of model is the share of a language model's brand mentions in a category that go to your brand, measured against the competitors you track. In effect it's share of voice counted inside one model's answers, and people often use the two terms for the same thing; Google Ads put “share of model” at about 70 US searches a month in October 2026. Finding out who gets named instead of you is the core of competitive intelligence in AI search.

Sentiment

Sentiment is the tone an AI answer takes toward a brand, positive, neutral or negative, judged from the wording and the context. A brand can be named often and described poorly, for instance still tied to a billing complaint it fixed two years ago. Read it next to citation rate; our guide to measuring brand awareness shows where it fits with your other brand metrics.

Measuring these metrics with Hypervisible

Hypervisible is the GEO software we publish, and most of the metrics above are what it reports: citation rate, average position, your rank against competitors in the same answers, the domains each engine cites as sources, and the sentiment of what the answers say about you. It reads ChatGPT, Gemini and Google AI Overviews on every plan, plus Perplexity and Google AI Mode on the five-engine plans. On paid plans, the same questions return to those engines each Monday and the gap with the previous week appears in points, so a citation rate can be followed as a trend. Readings can be run for buyers in ten countries, American, British, Canadian, French and German ones among them, in the language each of them searches in. Each reading draws on data from Google and the AI engines.

Martin Noale set up NEXUS GEO, the French GEO agency that makes Hypervisible, in Marly-le-Roi in January 2026. On nexus-geo.fr the agency keeps a public grid of 47 criteria that scores a brand's presence in AI answers, next to roughly a hundred guides, so the definitions on this page come from the people who built the tool.

Its limits fit in a sentence: readings are weekly, and Claude and Copilot fall outside what it measures.

The free AI visibility checker is the short version: 10 questions, five sector questions where your name doesn't appear and five where it does, each asked of three engines (ChatGPT, Gemini and Google AI Overviews), so 30 answers come back within 5 to 10 minutes. It takes no card, and results are kept in your account. Other tools are compared in AI visibility tools.

Mechanisms: fan-out, grounding, hallucination

Every answer with citations comes out of the same sequence: the engine splits the question, fetches pages, then writes from them. Hallucinations creep in when the writing drifts from what was fetched.

LLM

An LLM (large language model) is an AI model trained on huge amounts of text to predict the next token in a sequence, which is what lets it write text and answer questions. OpenAI's GPT models and Google's Gemini models are LLMs; ChatGPT and the Gemini app are products built on them. On its own, an LLM knows only its training data, which is why answer engines add a web search step.

Prompt

A prompt is the instruction or question a person types into an AI to get a response. In AI visibility tracking, the questions you monitor are also called prompts, and most tools price by how many you track. Wording changes results: “best CRM” and “best CRM for a 20-person law firm in Chicago” can bring up different brands.

RAG

RAG (retrieval-augmented generation) is a technique in which a language model retrieves documents before writing, so its answer rests on those documents rather than on its memory alone. The term comes from a May 2020 paper by researchers at Facebook AI Research, University College London and New York University. Google describes its AI features in the same terms, with a blunt consequence: a page the retrieval step never finds can't be cited.

Grounding

Grounding is the ability of an AI model to connect its output to verifiable sources of information. Google uses the word for how its AI features build answers from pages in its index and treats it as another name for RAG; Google Cloud also offers “Grounding with Google Search” to developers. The more grounded an answer, the more of its claims you can click through and check.

Query fan-out

Query fan-out is the technique by which an AI engine splits a question into several related sub-queries, runs them at the same time and assembles one answer from the results. Google's documentation calls it “a set of concurrent, related queries generated by the model to request more information,” and ChatGPT search does something close when it rewrites a request into several queries. A question about hot tub dealers in Phoenix that offer financing might branch into searches on dealers, financing terms and local reviews (our own example). That's why a narrow page that answers one branch well can be cited on a broad question.

Hallucination

A hallucination is a plausible but false statement produced by a language model and stated with the same confidence as a true one. OpenAI attributes part of the problem to training and evaluation methods that reward guessing over admitting uncertainty. For a brand, it can mean a discontinued plan quoted as current, or a service you never offered. AI reputation management covers how to catch and correct them.

Trust signals: E-E-A-T, structured data, entities

Google says its AI features rest on the same ranking and quality systems as regular Search. A page earns trust the way it always has, and the business behind it has to be easy to identify.

E-E-A-T

E-E-A-T stands for experience, expertise, authoritativeness and trustworthiness, the qualities Google's search quality raters are asked to look for in content. Google added the first E, for experience, on December 15, 2022, and says trust matters most of the four. E-E-A-T itself isn't a ranking factor, and raters have no control over how pages rank; their feedback tells Google whether its systems are working.

Structured data

Structured data is a standardized format, added to a page's code, that describes the page's content to search engines, such as a product and its price or a business and its opening hours. Google recommends JSON-LD and notes that most structured data in Search uses the schema.org vocabulary. For AI features, Google says no special markup is needed, and any markup you publish should match the visible text.

Schema.org

Schema.org is the shared structured data vocabulary founded by Google, Microsoft, Yahoo and Yandex, which names types such as Organization, Product and LocalBusiness and the properties that go with them. The project's site says that as of 2024, more than 45 million web domains marked up their pages with it. For most brands, the first job is an accurate Organization or LocalBusiness block with name, logo, address and official profiles.

Entity

An entity is a thing a search engine can identify without ambiguity, such as a person, place, organization or product, together with its attributes and its links to other entities. Your brand works as an entity for an AI engine once it can tell you apart from a namesake and tie you to a category and a city. Consistent facts across your site, your Google Business Profile, directories and press coverage make that easier.

Knowledge Graph

The Knowledge Graph is Google's database of billions of facts about people, places and things, and the relationships between them. Google introduced it on May 16, 2012, under the line “things, not strings,” with more than 500 million objects and 3.5 billion facts at launch. Knowledge panels in Google Search draw on it and are created automatically when there is enough information about an entity on the open web.

Crawler access: robots.txt, llms.txt and AI bots

AI engines generally cite only what their crawlers are allowed to read. These files and user agents decide which engine reads what, and what it may do with it.

robots.txt

robots.txt is a plain text file at the root of a website that tells crawlers which URLs they may or may not access. The convention goes back to 1994 and became RFC 9309 in September 2022, which states that the rules are not a form of access authorization, only rules crawlers are asked to honor. One overly broad Disallow line can take a site out of an AI engine's answers, which is why an AI visibility audit checks the file first.

llms.txt

llms.txt is a proposed Markdown file, usually placed at a site's root, that gives language models and AI agents a short overview of the site and links to its most useful pages. Jeremy Howard published the proposal on September 3, 2024, and it is still a proposal. Google says its Search doesn't use llms.txt files, though other services may. Our view: it costs little to publish and won't rescue unclear pages.

GPTBot

GPTBot is the OpenAI crawler that collects content which may be used to train its foundation models. Disallowing it in robots.txt tells OpenAI to keep your pages out of training. OpenAI says each of its settings is independent, so blocking GPTBot doesn't take you out of ChatGPT search.

OAI-SearchBot

OAI-SearchBot is the OpenAI crawler used for ChatGPT's search features. Block it and your site stops appearing in ChatGPT search answers, except as a navigational link; OpenAI says changes take about 24 hours. Your host or CDN also has to let OpenAI's published IP addresses through.

Google-Extended

Google-Extended is a robots.txt product token, with no crawler of its own, that keeps content Google crawls from being used to train future Gemini models or to ground answers in Gemini Apps and on Vertex AI. Google states that it has no effect on a site's inclusion or ranking in Google Search. Leaving AI Overviews and AI Mode is a different matter, handled by a Search Console setting, and Google warns that opted-out sites receive none of the traffic or impressions those features send.

Roles and effects from OpenAI's and Google's crawler documentation, read October 22, 2026.
Name	Company	What it does	If you block it
GPTBot	OpenAI	Collects content that may train models	Your pages stay out of training; ChatGPT search is unaffected
OAI-SearchBot	OpenAI	Crawls for ChatGPT search	Your site drops out of ChatGPT search answers
ChatGPT-User	OpenAI	Visits a page when a user asks	OpenAI says robots.txt rules may not apply to these visits
Google-Extended	Google	Control token, no crawler of its own	Content stays out of Gemini training and grounding; Google Search is unaffected
Googlebot	Google	Crawls for Google Search, AI features included	Blocked pages can't be sources for AI Overviews or AI Mode

For a brand that wants to be cited, the usual setup lets OAI-SearchBot and Googlebot in and settles the training question separately, through GPTBot and Google-Extended. Then check the outcome: a free scan shows whether ChatGPT, Gemini and Google AI Overviews name you today.

Your brand

The free analysis asks 10 questions to three AI engines (ChatGPT, Gemini and Google AI Overviews), reads the 30 answers in 5 to 10 minutes and keeps your dashboard in your account.

Frequently asked questions

What is GEO?

GEO stands for generative engine optimization: the practice of getting a brand or a page cited in answers written by AI engines such as ChatGPT, Perplexity, Gemini and Google AI Mode. The term comes from a 2023 research paper by a team from Princeton University and IIT Delhi. In practice it combines classic SEO, content written so it can be quoted, and work on the third-party sites AI engines draw on, measured by how often the AI names you.

What is the difference between GEO and AEO?

AEO, answer engine optimization, focuses on format: writing content an engine can lift as the direct answer to a question. GEO, generative engine optimization, is broader and covers being cited by AI engines that write their own answers, including through what other websites say about you. The methods and metrics overlap so much that many teams use the two terms interchangeably. Either way, the useful test is whether AI answers cite you.

What is a citation rate?

A citation rate is the percentage of AI answers that name your brand, out of a fixed list of questions asked on a schedule. If ChatGPT names you in 14 of 40 answers, your citation rate on ChatGPT is 35%. Track it per engine, since ChatGPT, Gemini and Perplexity often name different brands, and over several weeks, since answers vary from one session to the next.

How do you track GEO metrics?

Hypervisible, the software we publish, is what we recommend: it puts a fixed list of questions to ChatGPT, Gemini and Google AI Overviews, with Perplexity and Google AI Mode on its five-engine plans, then reports your citation rate, average position and rank against competitors per engine, with the change in points every Monday. Its free check reads 30 answers from three of those engines (ChatGPT, Gemini and Google AI Overviews) in 5 to 10 minutes, without a card. By hand, a few questions are manageable if the wording, the engines and the schedule stay the same.

What is query fan-out?

Query fan-out is how AI engines such as Google AI Mode handle a complex question: they split it into several related sub-queries, search for each at the same time, and combine the results into one answer. Google uses it in its AI features, and ChatGPT search rewrites requests into multiple queries in a similar way. As a result, a page that answers one narrow sub-question well can be cited on a much broader query.

What is share of model?

Share of model is the portion of a language model's brand mentions in your category that go to your brand, compared with the competitors you track on the same questions. It applies the advertising idea of share of voice to AI engines. A brand at 40% share of model on questions about payroll software is named far more often than rivals splitting the other 60%, whatever their Google rankings.

Do I need an llms.txt file?

No. llms.txt is a September 2024 proposal, and Google says its Search doesn't use these files, though other AI services may read them. Publishing one costs little, so there's no harm in adding it. Get the basics right first: a robots.txt that lets AI search crawlers such as OAI-SearchBot in, and clear pages with facts a model can check.