The GEO glossary. AI search, in plain words.

The AI Assistant channel is a channel in Google Analytics 4’s default channel group for visits from AI assistants such as ChatGPT, Gemini, DeepSeek, Copilot and Grok. Google documents that it excludes AI Overviews and AI Mode, whose clicks stay in Organic Search.

When the referrer matches Google’s list of assistants, GA4 sets the medium to ai-assistant. Assistants missing from that list still land in Referral; a custom channel group with a Source regex, placed above Referral, catches them and can be applied to past data.

Read the guide AI referral trafficAI Overviews (Google)AI Mode (Google)

AI Mode is a conversational mode of Google Search that writes an answer with links to supporting pages and accepts follow-up questions, designed for queries that need exploration, reasoning or complex comparisons. Google introduced it in March 2025 as a Search Labs experiment, then rolled it out widely.

AI Mode and AI Overviews may both use query fan-out, but Google says they may use different models and techniques, so their answers and links vary. Google says no special files, schema or technical requirements are needed beyond being indexed and snippet-eligible.

Read the guide AI Overviews (Google)Query fan-outGrounding

AI Overviews are AI-generated summaries that Google Search shows above the classic results for some queries, with links to the pages that support the answer. Google displays them only when it judges they add value, so many searches have none.

To be eligible as a supporting link, a page must be indexed and eligible to show with a snippet. AI Overview clicks are counted in Search Console’s “Web” search type and in GA4’s Organic Search channel, not in the AI Assistant channel.

Read the guide AI Mode (Google)Zero-click searchQuery fan-outGrounding

AI referral traffic is the visits a website receives when people click links in AI assistants’ answers, such as from chatgpt.com, perplexity.ai or gemini.google.com. In analytics it is always a floor, because some AI visits arrive without a referrer and are counted as Direct.

Links opened from desktop or mobile apps, copied URLs and strict referrer policies strip the source. Analytics also never shows the prompt behind a visit, and it misses every answer that mentioned you without a click.

Read the guide AI Assistant channel (GA4)Zero-click searchAI visibility

AI search is any search experience where an AI model writes an answer to the question instead of, or above, a list of links, as in ChatGPT search, Perplexity, Gemini, Google AI Overviews and AI Mode. The answer usually links to a few supporting sources.

Many AI search products run web searches before answering, often several at once (query fan-out), then read some results and write a synthesis. That is why classic SEO still feeds AI search, while what gets measured changes from rankings to mentions and citations.

Read the guide Answer engineQuery fan-outAI Overviews (Google)AI Mode (Google)

AI share of voice is the percentage of brand mentions in AI assistant answers that belong to your brand, measured across a fixed set of prompts and assistants. The formula is your brand’s mentions divided by the mentions of all tracked brands.

The prompt set defines the metric, so keep it stable and mostly unbranded. Calculate it per assistant before blending. There is no universal good score: compare against the competitors you lose deals to and against your own trend.

Read the guide AI visibilityBrand mention (in AI answers)Branded vs non-branded promptPosition (in an AI answer)

AI visibility is how often and how prominently a brand appears in AI assistants’ answers to the questions its buyers ask. It is usually measured over a fixed set of prompts with mentions, position, citations and share of voice.

Answers vary between runs, accounts and locations, so AI visibility is read as a trend over repeated runs, per assistant, rather than as a single snapshot.

Read the guide AI share of voiceBrand mention (in AI answers)Position (in an AI answer)Citation (cited source)

An answer engine is any system that responds to a question with a direct answer rather than a list of links. Google’s featured snippets were an early example; AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini and Claude are the current ones.

Each answer engine builds its answer differently: featured snippets quote a ranking page, Google’s AI features draw on its index, ChatGPT combines training data with web search when it decides to search, and Perplexity searches live for most questions.

Read the guide Answer engine optimization (AEO)AI searchZero-click search

Answer engine optimization (AEO) is the practice of making your content and brand easy for answer engines, such as Google AI Overviews, ChatGPT and Perplexity, to find, trust and use when they write a direct answer to a question. In practice it covers the same work as GEO.

AEO first described optimizing for featured snippets and voice assistants. Google’s 2026 guide to generative AI features names AEO and GEO and says that optimizing for them is “still SEO”, with no special files, chunking or markup required for Google Search.

Read the guide Generative engine optimization (GEO)Answer engineLLM SEO / LLMO

A brand mention is any appearance of a brand’s name in the text of an AI assistant’s answer, whether or not the answer links to the brand’s website. It differs from a citation, which is a source link shown with the answer.

A recommendation is a mention with intent behind it, such as “for a small sales team, X is a good fit”. The share of answers that mention you is often called mention rate; compared with competitors, it becomes share of voice.

Read the guide Citation (cited source)Position (in an AI answer)AI share of voiceSentiment (in AI answers)

A branded prompt names a specific brand (“Is HubSpot good for startups?”), while a non-branded prompt asks about a category or problem without naming one (“best CRM for startups”). Branded prompts show how an assistant describes you; non-branded prompts show whether you get discovered.

A branded prompt almost always mentions the brand it names, so counting it in share of voice inflates the score. Keep the core set mostly non-branded and report branded results separately, as a check on accuracy and framing.

Read the guide Prompt (tracked prompt)AI share of voiceHallucination (about a brand)

A citation is a link an AI assistant shows as a source for its answer, pointing to a page it used to write the response. A brand can be cited without being recommended, and recommended on the strength of pages it does not own.

For “best X” questions, assistants often cite review sites, comparison articles and forums. Looking at which domains are cited in the answers where you are absent shows where to earn coverage. Citation share is the fraction of cited URLs that point to your domain.

Read the guide Brand mention (in AI answers)GroundingQuery fan-outAI visibility

ClaudeBot is Anthropic’s crawler that collects web content that could contribute to training its Claude models. Blocking it signals that your future content should be excluded from training; Claude’s search uses a separate crawler, Claude-SearchBot.

Anthropic says blocking Claude-SearchBot may reduce your visibility and accuracy in Claude’s search results, and that its bots honor robots.txt. It advises against blocking its bots by IP, because they then cannot read your robots.txt.

Read the guide GPTBotrobots.txt (for AI crawlers)User-triggered fetcher

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness, the criteria Google’s Search Quality Rater Guidelines use to assess content quality, with trust as the most important. It is not a single ranking factor but describes what Google’s systems aim to reward.

Google added the first E, for first-hand experience, in December 2022. Its 2026 guidance on AI features asks for the same thing: non-commodity content that is helpful, reliable and people-first, such as real pricing, real limits and honest comparisons.

Read the guide EntityCitation (cited source)Answer engine optimization (AEO)

An entity is a uniquely identifiable thing, such as a company, person, product or place, that search engines and AI models recognize as distinct from other things with similar names. Clear, consistent facts about your brand across the web help systems treat it as one well-defined entity.

Keep the same category, pricing, positioning and product names on your site, your profiles and the directories that list you. Organization markup with sameAs links to your official profiles ties them together in a machine-readable way.

Read the guide Knowledge panelStructured data (schema markup)Hallucination (about a brand)

Generative engine optimization (GEO) is the practice of improving how often, and how favorably, AI assistants such as ChatGPT, Gemini and Perplexity mention, recommend and cite a brand in their answers. SEO aims for a high position on a results page; GEO aims for a place inside the answer itself.

The term comes from a 2023 academic paper on generative engines. In practice, GEO combines SEO, content strategy and reputation work, and is measured at the level of the answer: whether you are mentioned, at what position, next to which competitors and with which cited sources.

Read the guide Answer engine optimization (AEO)LLM SEO / LLMOAI visibilityAI share of voice

Google-Extended is a robots.txt control token that lets sites manage whether Google uses their crawled content for Gemini model training and grounding. Google says it does not affect inclusion in Google Search, is not a ranking signal, and does not control AI Overviews or AI Mode.

It has no user-agent string of its own: Google crawls with its usual crawlers and reads the token as a permission. AI Overviews and AI Mode follow Googlebot rules, snippet controls and the Search generative AI setting in Search Console.

Read the guide robots.txt (for AI crawlers)GroundingAI Overviews (Google)AI Mode (Google)

GPTBot is OpenAI’s web crawler for collecting content that may be used to train its foundation models. Blocking it in robots.txt opts your site out of that training but, per OpenAI, does not remove it from ChatGPT search.

OpenAI documents each of its robots.txt settings as independent. To appear in ChatGPT search, keep OAI-SearchBot allowed, whatever you decide for GPTBot.

Read the guide OAI-SearchBotrobots.txt (for AI crawlers)User-triggered fetcher

Grounding is connecting an AI model’s answer to external sources, such as web search results or a search index, so the answer is based on retrieved content rather than only on what the model learned in training. Grounded answers usually show the sources they used as citations.

Google describes its AI features as grounded in the Search index: it retrieves relevant pages with its core ranking systems, generates a response from them and links to them. A page that is not indexed cannot be a supporting link.

Read the guide Retrieval-augmented generation (RAG)Query fan-outCitation (cited source)Google-Extended

A hallucination is a statement an AI model presents as fact that is false or unsupported, such as a wrong price, a discontinued feature or an invented integration for a brand. It can come from outdated training data, from inaccurate sources the assistant retrieved, or from the model filling a gap.

You find them by asking branded prompts and reading the answers. The fix is usually upstream: state facts plainly on your own pages, keep them consistent across your profiles, and correct the third-party pages the assistant cites.

Read the guide Branded vs non-branded promptEntitySentiment (in AI answers)

A knowledge panel is the information box Google Search shows for entities in its Knowledge Graph, such as companies, people and places, summarizing key facts drawn from sources across the web. Its presence signals that Google recognizes the brand as a distinct entity.

Google generates knowledge panels automatically; you cannot create one, but an organization or person can claim theirs through Google’s verification process and suggest corrections. It is separate from a Google Business Profile.

Read the guide EntityStructured data (schema markup)

LLMO (large language model optimization), also called LLM SEO, is the practice of influencing how large language models describe and recommend a brand, both in answers built from web search and in answers drawn from training data. It is largely another name for GEO and AEO.

GEO, AEO, LLMO and “AI SEO” overlap so much that the label matters less than the metrics behind it. Pick the term your team already uses and define it by what you measure: mentions, positions, citations and share of voice.

Read the guide Generative engine optimization (GEO)Answer engine optimization (AEO)AI search

llms.txt is a proposed Markdown file placed at the root of a website (/llms.txt) that gives language models and AI agents a short, curated summary of the site with links to its most useful pages. It is a community proposal from llmstxt.org, not an official standard, and Google says its Search and AI features don’t use it.

Jeremy Howard of Answer.AI published the proposal in September 2024. OpenAI, Anthropic and Perplexity do not document it as a signal for which sources get cited, and it controls neither crawling nor training. It is most useful for developer docs and AI agents.

Read the guide robots.txt (for AI crawlers)Structured data (schema markup)

OAI-SearchBot is OpenAI’s search crawler, used to surface websites in ChatGPT’s search features. Sites that block it are not shown in ChatGPT search answers, although they can still appear as navigational links.

OpenAI notes that robots.txt changes can take about 24 hours to reach its search systems. ChatGPT also adds utm_source=chatgpt.com to links in its answers, which helps attribute the resulting visits in analytics.

Read the guide GPTBotrobots.txt (for AI crawlers)User-triggered fetcherAI referral traffic

PerplexityBot is Perplexity’s search crawler, used to surface and link websites in Perplexity’s answers; Perplexity says it is not used to train foundation models. Blocking it can keep your pages out of Perplexity results.

Perplexity recommends allowing PerplexityBot so your site can appear in its results, and publishes IP ranges plus CDN and firewall instructions. Its user-triggered fetcher, Perplexity-User, generally ignores robots.txt rules.

Read the guide robots.txt (for AI crawlers)User-triggered fetcherOAI-SearchBot

Position in an AI answer is the rank at which a brand is mentioned among the brands named in that answer, for example first or fifth in a shortlist. It is the closest equivalent to a search ranking for AI assistants.

Position changes between runs of the same prompt, so it is best read as an average or a trend. Being mentioned first in a three-brand shortlist is worth more than appearing last in a list of ten.

Read the guide Brand mention (in AI answers)AI share of voiceAI visibility

A tracked prompt is a question a buyer would realistically ask an AI assistant, such as “best CRM for a 20-person startup”, that you run repeatedly in one or more assistants to measure whether and how your brand appears. The set of tracked prompts defines what your AI visibility metrics mean.

A good starting set is 10 to 30 prompts covering category discovery, use cases, comparisons and problems, written in buyers’ words and mostly unbranded. Keep the wording fixed and date any change, or the trend stops being comparable.

Read the guide Branded vs non-branded promptPrompt volumeAI share of voice

Prompt volume is an estimate of how often people ask AI assistants a given question or topic. No AI assistant publishes this data, so every prompt volume figure is a third-party model, typically extrapolated from panels or keyword search volume, not a measured count.

Unlike Google search volume, there is no official source to check these numbers against. Use them as a rough signal of interest, and choose tracked prompts mainly on how close they are to revenue.

Read the guide Prompt (tracked prompt)Branded vs non-branded prompt

Query fan-out is when an AI assistant turns one question into several narrower web searches, runs them, and builds its answer from what comes back. Google defines it as “a set of concurrent, related queries generated by the model to request more information”.

Google uses the term for AI Mode and AI Overviews, and similar behavior can be observed in ChatGPT when it searches. The pages that rank for those sub-queries are the pool the assistant cites from. Fan-out is not deterministic: the same prompt can produce different searches on different runs.

Read the guide GroundingRetrieval-augmented generation (RAG)AI Mode (Google)Citation (cited source)

Retrieval-augmented generation (RAG) is a technique where a system first retrieves relevant documents, for example through a web search, and then has a language model generate an answer from them. It is how most AI search products produce answers with up-to-date information and citations.

For GEO, RAG means there are two gates: your page has to be retrieved (crawlable, indexed, ranking for the sub-queries) and then judged useful enough to be used and cited in the answer.

Read the guide GroundingQuery fan-outCitation (cited source)

robots.txt is the plain-text file at the root of a site that tells crawlers which URLs they may crawl, and it is how site owners allow or block each AI vendor’s crawlers by user-agent token. Training crawlers, search crawlers and user-triggered fetchers each read their own rule.

Blocking “AI bots” as one category can remove you from AI answers. Blocking GPTBot opts out of OpenAI training but keeps you in ChatGPT search; blocking OAI-SearchBot removes you. A crawler follows the group that names it most specifically and ignores the User-agent: * group.

Read the guide GPTBotOAI-SearchBotClaudeBotPerplexityBotGoogle-ExtendedUser-triggered fetcher

Sentiment in AI answers is the tone in which an assistant describes a brand: positive, neutral or negative, including the strengths, weaknesses and caveats it attaches to it. Two answers can both mention you while one recommends you and the other warns against you.

Sentiment scores are usually produced by a model classifying the answer text, so they are an approximation. Reading the answers themselves shows how you are framed and which sources the framing comes from.

Read the guide Brand mention (in AI answers)Hallucination (about a brand)AI share of voice

Structured data is code, usually JSON-LD using the schema.org vocabulary, that describes a page’s content (an organization, a product, an article) in a standard machine-readable format. Google says it is not required for its generative AI features and that there is no special schema.org markup for them.

Google still recommends structured data for rich result eligibility and page understanding. In a 2026 Ahrefs controlled test, pages that added JSON-LD saw no meaningful lift in AI citations. Markup should match the visible page, and facts should appear in plain text first.

Read the guide EntityKnowledge panelllms.txt

A user-triggered fetcher is an AI agent that visits a page because a person asked for it in a chat, for example to summarize a pasted URL, rather than crawling the web automatically. Examples are ChatGPT-User, Claude-User and Perplexity-User.

Because a human requested the page, some vendors say robots.txt may not apply: OpenAI says rules “may not apply” to ChatGPT-User, and Perplexity says Perplexity-User “generally ignores” them. Anthropic says its bots, including Claude-User, honor robots.txt.

Read the guide robots.txt (for AI crawlers)GPTBotClaudeBotPerplexityBot

A zero-click search is a search where the user gets what they need on the results page or in the AI answer and does not click through to any website. AI Overviews and assistant answers make zero-click outcomes more common, especially for simple informational questions.

A buyer can read a shortlist of three tools in an assistant, pick one and never click. Your analytics record nothing either way, which is why AI visibility (being mentioned) needs its own metrics alongside AI referral traffic.

Read the guide AI Overviews (Google)AI referral trafficAI visibility