GEO & AI Search Glossary
Plain-language definitions of the terms behind generative engine optimization, AI brand monitoring, and how large language models decide which brands to mention.
Agentic Search
Agentic search is research delegated to an AI agent. Instead of answering one question, the agent plans a task (compare vendors, find the best price, shortlist tools), runs many searches, reads the pages itself, and returns a decision. The human sees the conclusion, not the browsing.
AI Assistant
An AI assistant is the product wrapped around a language model: the chat interface, the web search integration, the memory, the app. ChatGPT, Claude, Gemini, Perplexity, and Copilot are assistants. When people say 'AI told me', they mean an assistant's answer.
AI Brand Monitoring
AI brand monitoring is the ongoing practice of checking how AI assistants talk about your brand. It tracks whether you get mentioned, how accurately you are described, and how you stack up against competitors in AI answers.
AI Citation
An AI citation is a source that an assistant points to when it answers a question. When the model cites your domain, you are not just mentioned. You are the evidence, which is the strongest position to hold in an AI answer.
AI Crawler
An AI crawler is a bot that fetches your web pages for an AI system. Some collect content for model training (GPTBot, ClaudeBot, CCBot), others retrieve pages live so an assistant can answer with current information (OAI-SearchBot, PerplexityBot). Whether you allow them in robots.txt shapes whether AI systems can learn about and cite your site.
AI Mode
AI Mode is Google's conversational search tab. Instead of ten blue links, it generates a full answer with Gemini, supports follow-up questions, and pulls sources through a technique called query fan-out, where one question is expanded into many sub-queries searched in parallel.
Google AI Overviews
Google AI Overviews are AI-generated summaries that appear above the traditional results for many searches. They answer the question directly and cite a handful of sources, which changes where attention and clicks go.
AI Referral Traffic
AI referral traffic is the visits your site gets from AI assistants: someone asks ChatGPT or Perplexity a question, the answer cites your page, and they click through. It shows up in analytics with referrers like chatgpt.com, perplexity.ai, and gemini.google.com.
AI Search
AI search is when a person asks an AI assistant a question and gets a synthesized answer, rather than a list of links to click. Tools like ChatGPT, Perplexity, Gemini, and Google's AI Overviews read across sources and return one response, often naming specific brands and citing some pages.
AI Share of Voice
AI share of voice is the share of relevant AI answers where your brand gets mentioned, compared with competitors. It is the AI-era version of the visibility metric marketers have always tracked.
AI Visibility
AI visibility is how often, how prominently, and how favorably a brand shows up in AI-generated answers. It covers whether an assistant mentions you at all, where in the answer you appear, whether your site gets cited as a source, and what the answer actually says about you.
Answer Engine
An answer engine is a system that gives you a direct answer rather than a page of links to pick from. ChatGPT, Perplexity, Gemini, and Google's AI Overviews all behave like answer engines.
Answer Engine Optimization (AEO)
Answer Engine Optimization (AEO) is the practice of structuring content so an answer engine can pull a clear, direct answer straight out of it. It overlaps heavily with GEO, with more emphasis on the question-and-answer format.
Answer Position
Answer position is where in an AI answer your brand shows up. Being the first name the assistant gives carries most of the value; a mention at the end of a list of seven carries little. It is the closest AI-search equivalent to a ranking position.
Brand Mention
A brand mention is any point where an AI assistant names your brand in its answer. In GEO, the position and the framing matter. Being mentioned first as a recommended option is very different from being a footnote.
Chunking
Chunking is the process of splitting a page or document into smaller passages so a retrieval system can index each one and return only the most relevant part to an AI answer. Well-structured content chunks cleanly along clear headings and self-contained paragraphs, which makes the right passage easier to retrieve and quote.
Citation Share
Citation share is the share of AI answers, across a set of relevant buyer questions, that cite your domain as a source. If assistants answer 100 questions about your category and your site is cited in 12 of them, your citation share is 12 percent. It is the evidence-level counterpart to mention rate.
Content Freshness
Content freshness is how current a page is: when it was last updated and whether its facts still hold. Retrieval-backed AI search favors recent sources for anything time-sensitive, so a maintained page with a visible update date and a changelog beats an identical page that looks abandoned.
Context Window
A context window is the maximum amount of text a language model can hold in mind at once, measured in tokens. It includes the prompt, any retrieved sources, and the answer being written. When retrieved pages exceed the window, some content is dropped, so what fits, and where it sits, affects what the model uses.
Conversational Search
Conversational search is research done as a dialogue: you ask an assistant a question, get an answer, and refine with follow-ups, with the assistant remembering the thread. It replaces a series of separate keyword searches with one evolving conversation.
Digital PR
Digital PR is the practice of earning coverage on publications, newsletters, and industry sites: data stories, expert quotes, product coverage. For GEO it is load-bearing, because AI assistants assemble recommendations mostly from third-party sources, and digital PR is how your brand gets onto them.
E-E-A-T
E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trustworthiness. It is the framework Google uses to judge content quality, and the same signals help AI systems decide which sources to trust and cite. Strong E-E-A-T makes a page more likely to be retrieved and quoted in an AI answer.
Entity (Entity SEO)
An entity is a distinct, recognizable thing, such as your brand, a product, or a person, that a search or AI system can identify and reason about. The clearer your entity, the more reliably you get described and recommended.
Fine-Tuning
Fine-tuning takes a trained model and continues training it on a focused dataset to shape its behavior: a support tone, a domain vocabulary, a task format. It is how companies customize models, and one of the ways assistant behavior changes between releases without a new base model.
Generative AI
Generative AI is software that produces new content rather than retrieving existing content: it writes text, generates images, drafts code. Large language models are the text-generating branch, and they power the assistants that now answer buyer questions directly.
Generative Engine
A generative engine is a search system that answers a question by writing the answer, not by listing links. It combines a large language model with retrieval: the engine fetches relevant sources, then generates a synthesized response that may cite a few of them.
Generative Engine Optimization (GEO)
Generative Engine Optimization (GEO) is the practice of shaping how AI assistants describe and recommend your brand. The goal is simple. When a buyer asks an AI a question in your category, your brand shows up, sounds right, and gets recommended.
GPTBot
GPTBot is OpenAI's crawler for gathering content that may be used to train its models. It is not the same as OAI-SearchBot, which fetches pages so ChatGPT can answer with live web content, or ChatGPT-User, which fetches a page when a user asks about it directly. Each can be allowed or blocked separately in robots.txt.
Grounding
Grounding is the practice of tying an AI answer to actual sources the model retrieves, rather than relying only on its training memory. Grounded answers tend to cite, which is exactly where a brand can earn its place.
AI Hallucination
An AI hallucination is when a model produces a confident answer that is simply wrong. For a brand, that can look like an outdated price, a feature you do not offer, or a claim that quietly damages trust.
Inference
Inference is the moment a trained model runs: it takes the prompt plus any retrieved context and generates the answer, token by token. Training happens rarely and bakes in knowledge; inference happens on every question and assembles the actual answer a buyer reads.
Information Gain
Information gain is what your page adds that the rest of the web has not already said: a new number, a firsthand test, an original dataset, a real example. Pages that only restate existing consensus add zero gain, and both search ranking and AI citation increasingly discount them.
Knowledge Cutoff
A knowledge cutoff is the point in time where a model's training data ends. A model with a January 2026 cutoff knows nothing that happened after that date on its own. Live AI search works around this with retrieval, fetching current web content at answer time.
Knowledge Graph
A knowledge graph is a structured network of entities, like brands, people, and products, and the relationships between them. Search engines and AI systems use it to understand what your brand is, not just the words on your page.
Large Language Model (LLM)
A large language model (LLM) is an AI system trained on huge amounts of text to predict and generate language. It powers assistants like ChatGPT, Claude, and Gemini, and it is the thing deciding how your brand gets described.
LLM Optimization (LLMO)
LLM Optimization (LLMO) is the practice of shaping how large language models represent your brand, so that when a model answers a question in your category, it describes you accurately and includes you among its recommendations. It is a near-synonym for generative engine optimization, framed around the model itself.
llms.txt
llms.txt is a simple file at the root of your domain that points AI models to your most important, machine-readable content. Think of it as a friendly map for assistants, the way robots.txt is for crawlers.
Mention Rate
Mention rate is the share of relevant buyer questions where an AI assistant names your brand in its answer. Ask the questions your buyers actually ask, across ChatGPT, Claude, and Gemini, and count how often you appear. That percentage, tracked over time, is your mention rate.
Prompt
A prompt is the question or instruction a person gives an AI assistant. In GEO, the prompts your buyers actually use are the queries you need to win, the same way keywords were in classic search.
Prompt Engineering
Prompt engineering is the practice of crafting the input you give an AI model to get a more accurate, relevant, or useful output. In brand monitoring, it is how you design the questions used to test what models say about your category, so results are consistent, realistic, and comparable over time.
Query Fan-Out
Query fan-out is how AI search systems research a question. Instead of running your query once, the system generates many related sub-queries, searches them all in parallel, and builds its answer from the combined results. Google describes AI Mode working this way.
Retrieval-Augmented Generation (RAG)
Retrieval-Augmented Generation (RAG) is a technique where an AI retrieves relevant documents at answer time and uses them to ground its response. It is why up-to-date, well-structured content can shift what an assistant says about you.
robots.txt
robots.txt is a plain text file at yoursite.com/robots.txt that tells crawlers which parts of the site they may fetch. It has been a search engine convention for decades, and it now doubles as the control panel for AI: rules for GPTBot, ClaudeBot, PerplexityBot, and Google-Extended decide whether AI systems can learn about your brand and cite your pages.
Semantic Search
Semantic search matches the meaning and intent behind a query, not just the exact keywords. It is why AI assistants can answer a question phrased in a dozen different ways and still understand what the buyer wants.
Sentiment Analysis
Sentiment analysis is the work of judging the tone attached to a mention, whether positive, neutral, or negative. In AI answers, how you are framed can matter more than whether you appear.
Structured Data (Schema Markup)
Structured data, often added with schema.org markup, labels parts of your page so machines know what each one means. It helps AI assistants and search engines read your brand facts without guessing.
System Prompt
A system prompt is the standing instruction an AI assistant receives before any user input: who it is, how to format answers, when to search the web, how to cite sources. Users never see it, but it shapes every answer, including how brands get recommended and attributed.
Token
A token is the chunk of text a language model actually processes: a word, part of a word, or punctuation. As a rule of thumb, 1,000 tokens is about 750 English words. Model context windows, API pricing, and answer length limits are all counted in tokens.
Topical Authority
Topical authority is being the site that has genuinely covered a subject: the definitions, the how-tos, the comparisons, the data, the edge cases. Search engines use it to decide who ranks; AI systems reflect it in who gets retrieved and cited when questions in that subject come up.
Training Data
Training data is the corpus of text a language model learns from, largely web content gathered by crawlers, plus licensed and curated datasets. A model's default beliefs about your brand are a compression of what that corpus says, which is why what the web wrote about you two years ago can still shape AI answers today.
Trust Gap
A trust gap is the distance between being mentioned and being cited. If an AI names you but backs its answer with competitor sources, the model trusts them as evidence and treats you as an afterthought.
Vector Embedding
A vector embedding is a list of numbers that represents the meaning of a piece of text, so AI systems can compare how similar two texts are. Retrieval systems embed both the user's question and candidate pages, then pull the pages whose embeddings sit closest in meaning, which is how relevant sources reach an AI answer.
Zero-Click Search
A zero-click search is a query the user gets answered without visiting any website, because the answer appears directly on the search or assistant surface. AI answers and AI Overviews accelerate this: the model resolves the question in place, so brands can be described or recommended without ever receiving a visit.