AI search & GEO Glossary — 101 terms defined

agentic search Agentic search is AI that browses the web and takes multi-step actions on your behalf to complete a task, not just answering from memory. Rather than returning links, an agent plans steps, reads full pages, cross-checks facts, and can act (compare, book, buy) with little human input.
AI agent An AI agent is a software system built on a large language model that pursues a goal by reasoning, planning, and taking actions with tools, adjusting its next step based on feedback from the environment. Unlike a fixed workflow, an agent decides for itself how to reach the goal.
AI brand monitoring AI brand monitoring is the practice of tracking how, where, and how often a brand appears in AI-generated answers from tools like ChatGPT, Perplexity, Gemini, and Google AI Overviews. It measures mentions, citations, sentiment, and share of voice against competitors across sampled prompts.
AI Citations AI citations are the sources an AI answer names or links to when it responds to a query. In AI search (Google's AI Overviews and AI Mode, ChatGPT, Perplexity), being cited is the equivalent of ranking: if the model draws on your page, your brand becomes part of the answer the user reads.
AI crawler An AI crawler is an automated bot that fetches web pages to train large language models or to ground the answers AI assistants give. Examples include OpenAI's GPTBot, Anthropic's ClaudeBot, and Google's Google-Extended token, most of which site owners manage through robots.txt.
AI Overviews AI Overviews are Google's AI-generated answer summaries shown at the top of search results, above the classic blue links. They use Gemini models to pull together information from multiple web pages and cite the sources those answers draw from.
AI search AI search is a search experience where a large language model reads across web sources and returns a synthesized, conversational answer with supporting links, instead of only a ranked list of blue links. Examples include Google's AI Overviews and AI Mode, ChatGPT search, and Perplexity.
AI search optimization AI search optimization is the practice of shaping content and site signals so AI-powered search surfaces (Google AI Overviews and AI Mode, ChatGPT, Perplexity, Gemini) retrieve, summarize, and cite your pages. It is an umbrella term for doing SEO work aimed at generative answers, not just blue links.
AI SEO AI SEO is optimizing a website to stay visible as search shifts toward AI-generated answers, both by using AI tools in the workflow and by making content that AI surfaces like Google's AI Overviews, ChatGPT and Perplexity can cite. Google's position: optimizing for AI search is still SEO.
AI Traffic AI traffic is website visits that arrive from AI assistants and answer engines like ChatGPT, Perplexity, Gemini, and Copilot, when a user clicks a link or citation inside the AI's answer. In GA4 it appears under the AI Assistants channel (medium ai-assistant). Also called LLM referral traffic.
AI visibility AI visibility is how often and how prominently a brand shows up in, and gets cited by, the answers AI engines like ChatGPT, Google AI Overviews, Perplexity, and Gemini generate. It tracks presence inside AI-generated answers rather than ranking position in a classic list of blue links.
answer engine An answer engine is a system that responds to a query with a single, direct answer instead of a ranked list of links. Tools like Perplexity, ChatGPT search, and Google's AI Overviews read live web sources, synthesize them, and return a cited answer the user can act on without clicking through.
answer engine optimization Answer engine optimization (AEO) is the practice of structuring and formatting content so AI assistants and answer engines (ChatGPT, Perplexity, Google AI Overviews, voice assistants) can lift it directly as the answer to a question, rather than ranking it as one of ten blue links.
Answer Synthesis Answer synthesis is the step where an AI search or RAG system uses a language model to fuse retrieved passages from several sources into one generated answer, instead of returning a list of links. It is the generation half of retrieval-augmented generation.
Applebot-Extended Applebot-Extended is a robots.txt user-agent token Apple introduced in 2024 that lets publishers opt out of having their content used to train Apple's foundation models (Apple Intelligence). It does not crawl pages itself, and disallowing it does not remove your site from Apple search.
Brand Entity A brand entity is your company represented as a distinct "thing" in a search engine's Knowledge Graph, with its own identifier, attributes (name, logo, founders, products), and relationships to other entities, rather than as a keyword string. It is how Google and AI systems recognize your brand as a real-world object.
Brand SERP A brand SERP is the Google search results page returned when someone searches your exact brand name. It reflects how Google understands your brand as an entity and is often the first impression prospects, partners, and journalists get, which makes it a high-leverage reputation surface.
Bytespider Bytespider is the web crawler run by ByteDance, the company behind TikTok. It gathers web content to train ByteDance's large language models (including its ChatGPT rival Doubao) and to feed other ByteDance AI products.
CCBot CCBot is the web crawler run by Common Crawl, a nonprofit that publishes a free, petabyte-scale public archive of web pages. It identifies itself with the user-agent CCBot/2.0, and its open dataset has been used to train large language models such as GPT-3, LLaMA, and BLOOM.
ChatGPT ChatGPT is a generative AI chatbot from OpenAI, launched in November 2022, that answers questions in plain conversation using its GPT large language models. With its built-in web search, it can also pull live information from the internet and cite the pages it draws from.
ChatGPT search ChatGPT search is ChatGPT's web-connected mode that answers questions with current information pulled from the live web and links to the sources it cites. Instead of a list of blue links, it returns one synthesized answer with clickable inline citations to the original pages.
chunking Chunking is splitting a document into smaller passages so each piece can be embedded as a vector and retrieved on its own. In retrieval-augmented generation (RAG), chunks are the units an AI pulls to ground answers, so how you split text decides what gets found and cited.
citation-worthiness Citation-worthiness is how likely a specific passage of content is to be pulled into and cited by an AI answer engine such as ChatGPT, Google AI Overviews, or Perplexity. It hinges less on domain authority than on whether that passage gives a clear, sourced, self-contained answer the engine can lift.
Claude Claude is a family of large language models built by Anthropic and the AI assistant of the same name, available at claude.ai and in mobile, desktop, and coding apps. It answers questions, writes and debugs code, searches the web for current information, and runs autonomous agents.
ClaudeBot ClaudeBot is Anthropic's web crawler that collects public web content to help train its Claude AI models. It identifies itself with the user-agent token ClaudeBot, obeys robots.txt, and is one of three Anthropic crawlers alongside Claude-User and Claude-SearchBot.
Content Grounding Content grounding is the practice of writing clear, factual, well-sourced content that AI systems can safely use as the evidence behind their answers. In generative AI, grounding tethers a model's output to verifiable sources, so groundable content is content an LLM can retrieve, cite, and trust.
context window A context window is the maximum amount of text, measured in tokens, that a large language model can consider at once. It covers both the input (the prompt plus any retrieved documents) and the output the model generates, and it acts as the model's working memory for a single request.
conversational search Conversational search is search conducted as a multi-turn dialogue: you ask a question in natural language, then refine it with follow-ups, and the system carries context from earlier turns instead of treating each query as fresh. Google AI Mode, ChatGPT search, and Perplexity all work this way.
cosine similarity Cosine similarity measures how aligned two vectors are by the cosine of the angle between them, scored from -1 to 1 (1 = same direction, 0 = unrelated). In AI search it compares a query's embedding against passage embeddings to rank which content best matches the query's meaning.
DBpedia DBpedia is a community project that extracts structured facts about entities (people, places, companies, creative works) from Wikipedia and publishes them as a machine-readable knowledge graph of RDF triples, queryable with SPARQL and connected to the wider linked-data web.
deep research Deep research is an AI agent mode (in ChatGPT, Gemini, and Perplexity) that plans a multi-step investigation, browses and reads many web sources on its own, then returns a long, cited report. It trades speed for depth, usually taking minutes rather than seconds.
DeepSeek DeepSeek is a Chinese AI lab (founded 2023 in Hangzhou, spun out of the High-Flyer hedge fund) and the chatbot app of the same name. It builds open-weight large language models, notably the V3 general model and the R1 reasoning model, with R1 released under a permissive MIT license.
Disambiguation Page A disambiguation page lists the separate things that share one name and links each to its own entry, so a reader or search engine picks the intended one (Wikipedia's "Mercury" is the classic example). In SEO, disambiguation also means getting engines and AI to attach your content to the right entity.
embeddings Embeddings are numeric vector representations of text (or images, audio) that place items with similar meaning close together in a high-dimensional space. Search and AI systems compare these vectors to match content to a query by meaning instead of exact keywords.
Entity Disambiguation Entity disambiguation is the process of deciding which specific real-world entity a name refers to when several share the same label, for example Apple the fruit versus Apple Inc. Search engines and AI models resolve each mention to one unique node in a knowledge base such as the Google Knowledge Graph or Wikidata.
Entity Extraction Entity extraction is the natural language processing task of automatically identifying and classifying named things (people, places, organizations, dates, products) in unstructured text, then linking them to known records. It is also called named entity recognition (NER) or entity detection.
Entity Home An entity home is the single web page you designate as the authoritative source of facts about an entity (your brand, a person, a product). Search engines and AI systems treat it as the baseline they reconcile every other mention against, so it is usually the homepage or a dedicated About page.
Entity Linking Entity linking is the natural language processing task of mapping a mention in text (a person, place, or brand) to its unique entry in a knowledge base like Wikidata or the Google Knowledge Graph. It pairs named entity recognition with disambiguation to decide which real-world entity a name refers to.
Entity Salience Score Entity salience score is a 0.0 to 1.0 value that NLP systems like the Google Cloud Natural Language API assign to each entity in a document, measuring how central that entity is to the whole text. Scores near 1.0 mark the main subject; near 0 marks a passing mention.
Entity SEO Entity SEO is the practice of optimizing a site so search engines and AI systems recognize your brand, people, and topics as defined entities (things with attributes and relationships) rather than loose keyword strings. It uses structured data, sameAs links, and consistent naming to place you inside knowledge graphs.
Entity-based SEO Entity-based SEO optimizes content around the people, places, things, and concepts that search engines store as distinct entities in a knowledge graph, and the relationships between them, rather than matching raw keyword strings. It helps search and AI engines recognize your brand as a defined thing.
fine-tuning Fine-tuning is the process of taking a pretrained AI model and training it further on a smaller, task-specific dataset of example inputs and outputs, so it adapts to a particular domain, format, or behavior without being rebuilt from scratch.
foundation model A foundation model is a large AI model trained on broad data at scale (usually via self-supervision) that adapts to many downstream tasks through fine-tuning, prompting, or retrieval. GPT, Gemini, Claude, and Llama are foundation models; the LLMs behind AI search are a subset.
Freebase Freebase was a large, community-built knowledge base of structured facts about entities (people, places, things), launched by Metaweb in 2007 and acquired by Google in 2010. Its data helped seed Google's Knowledge Graph. Google shut Freebase down in 2016 and moved the data toward Wikidata.
Generative Engine Optimization Generative Engine Optimization (GEO) is the practice of structuring content and entity signals so AI answer engines (ChatGPT, Perplexity, Google AI Overviews) cite or mention you when they generate answers. The term was coined in a 2023 academic paper that measured content edits lifting visibility up to 40%.
Generative Snippet A generative snippet is a search result description that Google's AI writes from scratch instead of extracting verbatim text from your page or meta description. It summarizes the page to fit the query, and Google labels these AI-composed snippets with a Gemini icon.
Google AI Mode Google AI Mode is a conversational AI search experience in its own Search tab. It answers complex, multi-part questions with a synthesized response and supports follow-ups, using a query fan-out technique to run many related searches at once instead of returning a plain list of links.
Google Gemini Google Gemini is Google's family of natively multimodal large language models, built by Google DeepMind, that process text, images, audio, video, and documents. The name also covers Google's AI chatbot (renamed from Bard in February 2024), which runs on those models.
Google Knowledge Graph The Google Knowledge Graph is Google's database of real-world entities (people, places, things) and the connections between them. Launched in 2012, it powers knowledge panels and lets Search understand "things, not strings," reasoning about concepts instead of matching keyword text.
Google-Extended Google-Extended is a robots.txt control token that lets publishers decide whether Google may use their crawled content to train and ground Gemini models and Vertex AI. It is not a separate crawler, and it does not affect Google Search indexing or rankings.
GPTBot GPTBot is OpenAI's web crawler that collects publicly accessible content to train its GPT foundation models. Site owners allow or block it in robots.txt via the user-agent token "GPTBot," which is separate from OpenAI's search crawler (OAI-SearchBot) and its live user-fetch agent (ChatGPT-User).
Grok Grok is xAI's conversational AI assistant, a family of large language models that can search the open web and X (formerly Twitter) in real time and cite the live sources it uses. It runs at grok.com, in iOS and Android apps, and inside the X platform.
grounding Grounding is the practice of tying a large language model's output to verifiable external sources (retrieved documents, live web results, or a database) instead of the model's memory alone. Google defines it as connecting model output to verifiable sources of information, which reduces hallucinations.
hallucination An AI hallucination is when a large language model produces confident, plausible-sounding text that is factually wrong or fabricated, such as an invented statistic, a fake citation, or false details about a company. It happens because training rewards guessing over admitting uncertainty.
hybrid search Hybrid search is a retrieval method that runs keyword search (BM25) and vector (semantic) search in parallel over the same index, then merges the two ranked lists into one result set, usually with Reciprocal Rank Fusion (RRF). It captures both exact term matches and conceptual meaning.
information retrieval Information retrieval (IR) is the science of finding material, usually documents, that satisfies a user's information need from within large collections. It is the engine behind every search engine and the retrieval step that feeds AI answer systems like AI Overviews, ChatGPT, and Perplexity.
knowledge base A knowledge base is a structured store of facts and documents an AI retrieves from to answer questions, instead of relying only on what it memorized in training. Google's Knowledge Graph and the document indexes behind retrieval-augmented generation (RAG) are both examples.
Knowledge Vault Knowledge Vault was a 2014 Google research project that automatically extracted facts from the web (free text, HTML tables, page markup) and fused them with existing knowledge bases into a probabilistic store, giving each subject-predicate-object fact a calibrated confidence score.
large language model A large language model (LLM) is a neural network trained on huge amounts of text to predict the next token in a sequence, which lets it understand and generate human language. Transformer-based LLMs power tools like ChatGPT, Gemini, and Perplexity.
LLM citation An LLM citation is a source link that an AI assistant attributes part of its answer to, surfaced inline or in a sources panel when the model draws on live web results (ChatGPT Search, Perplexity, Google AI Overviews). Earning one places your page inside the AI answer itself, not just the blue-link list.
LLM SEO LLM SEO (also called LLMO, large language model optimization) is the practice of structuring your content, entities, and off-site mentions so large language models like ChatGPT, Gemini, and Claude reference and cite your brand in their answers. It is a near-synonym of generative engine optimization (GEO).
llms-full.txt llms-full.txt is a plain-text companion to llms.txt that holds the full body text of a site's key pages inline in one Markdown file, so an AI model can ingest the whole knowledge base in a single fetch instead of crawling many separate URLs.
llms.txt llms.txt is a proposed root-level Markdown file (served at /llms.txt) that lists a site's key pages in a clean, LLM-friendly format to help AI models find and read your content. It is voluntary, and no major AI search engine has confirmed using it as a ranking signal as of 2026.
Machine-readable Content Machine-readable content is web content structured and marked up so crawlers and AI models can parse its meaning without guessing, using standardized formats like semantic HTML and JSON-LD structured data that explicitly label entities, relationships, and page structure.
Meta AI Meta AI is Meta's conversational AI assistant, built on its open-weight Llama models and available inside WhatsApp, Instagram, Facebook, Messenger, the web (meta.ai), and Ray-Ban Meta glasses. It answers questions, generates images, and searches the live web without leaving Meta's apps.
Microsoft Copilot Microsoft Copilot is Microsoft's AI assistant, powered by large language models and built into Bing, Edge, Windows and Microsoft 365. It answers questions in natural language, grounds those answers in Bing's web index, and cites the pages it draws from. It was formerly called Bing Chat.
Model Context Protocol The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in November 2024, that lets AI applications like Claude and ChatGPT connect to external data sources, tools, and workflows through one shared interface instead of custom, one-off integrations.
multimodal AI Multimodal AI is a model that processes and reasons across more than one kind of input at once, such as text, images, audio, and video, instead of text alone. AI search systems like Google's AI Mode use it to interpret a photo and a spoken question together.
named entity recognition Named entity recognition (NER) is a natural language processing task that detects named things in text (people, places, organizations, dates, products) and classifies each into a type. Search engines and AI models use it to turn raw text into structured entities they can link, disambiguate, and reason about.
natural language generation Natural language generation (NLG) is the branch of AI and natural language processing that turns structured or non-linguistic data into readable human text. It is the inverse of natural language understanding, and the mechanism large language models and answer engines use to write prose.
natural language processing Natural language processing (NLP) is the field of AI that gives computers the ability to read, interpret, and generate human language. It powers how search engines and AI assistants grasp the meaning behind a query and produce answers, weighing each word in full context.
nearest neighbor search Nearest neighbor search (NNS) finds the data points closest to a query point in a vector space, ranked by a distance or similarity measure. In AI search it runs over embeddings to retrieve the most semantically similar passages, using approximate (ANN) methods at scale to trade a little accuracy for speed.
OAI-SearchBot OAI-SearchBot is OpenAI's web crawler that indexes pages so they can appear in ChatGPT's search answers and citations. It identifies itself with an OAI-SearchBot user-agent, obeys robots.txt, and does not collect content to train models (GPTBot does that).
Perplexity Perplexity is an AI answer engine that searches the live web and replies with a synthesized, conversational answer plus numbered inline citations to the sources it used, instead of returning a ranked list of links like a traditional search engine.
PerplexityBot PerplexityBot is the crawler Perplexity AI uses to discover and index web pages so it can surface and link to them in its answers. It identifies as PerplexityBot/1.0 and obeys robots.txt, unlike Perplexity-User, the separate agent that fetches pages live when a person asks a question.
prompt engineering Prompt engineering is the practice of writing and refining the instructions you give a large language model so it consistently produces the output you want. It covers how you phrase a request, the context and examples you include, and the role or system instructions that steer the model's response.
prompt injection Prompt injection is an attack that hides malicious instructions inside the text an AI model reads (a user message, a web page, a file) to override its original instructions and make it do something unintended. OWASP ranks it the top security risk for LLM apps (LLM01).
Prompt-level Visibility Prompt-level visibility is the practice of tracking, prompt by prompt, whether and how a brand appears in AI assistant answers (ChatGPT, Gemini, Perplexity, Google AI Mode) for specific user questions, instead of measuring rank at the keyword or domain level.
Query Expansion Query expansion reformulates a search query by adding synonyms, related terms, and subtopics so a system retrieves relevant results even when documents use different words than the searcher. Modern AI search applies it aggressively through query fan-out, turning one question into many parallel subqueries.
query fan-out Query fan-out is the technique Google's AI Mode and AI Overviews use to answer one prompt: instead of running a single search, the system breaks the question into many related sub-queries, runs them at once across subtopics and data sources, then synthesizes one answer.
Query Understanding Query understanding is how a search engine or AI assistant parses a search query to work out its intent, entities, and context instead of matching keywords literally. It uses natural-language models for spelling, synonyms, entity recognition, and intent classification to figure out what the user actually wants.
RDF RDF (Resource Description Framework) is a W3C standard data model that expresses information as triples: a subject, a predicate, and an object, each named by a global identifier (IRI). It represents entities and their relationships as a graph, and it underpins knowledge graphs, linked data, and JSON-LD structured data.
reranking Reranking is a second-stage step in AI search and retrieval systems that reorders an initial set of retrieved candidates by true relevance to the query, usually with a cross-encoder model that scores each query-document pair before the top results are passed to the LLM.
retrieval-augmented generation Retrieval-augmented generation (RAG) is an AI technique where a language model fetches relevant documents from an external source at query time and grounds its answer in that retrieved text, instead of relying only on what it memorized during training.
sameAs sameAs is a schema.org property that holds the URL of an authoritative page describing the same entity (its Wikipedia page, Wikidata entry, official site, or social profile), so search engines and AI systems can confirm identity and merge scattered references into one entity node.
Search Generative Experience Search Generative Experience (SGE) was Google's original beta name for the AI-generated answer summaries shown above regular search results. Launched inside Search Labs in May 2023, the feature graduated to general release in May 2024 under a new name: AI Overviews.
SearchGPT SearchGPT was OpenAI's 2024 prototype AI search engine that answered questions from live web results with inline citations and links to sources. OpenAI folded it into ChatGPT, launching ChatGPT search in October 2024, so today the name effectively refers to ChatGPT's built-in web search.
semantic chunking Semantic chunking is a text-splitting method for retrieval systems that places chunk boundaries where meaning shifts, using embedding similarity between adjacent sentences, instead of cutting at a fixed character or token count. Each resulting chunk holds sentences that are topically related.
Semantic Relevance Semantic relevance measures how closely a page's meaning matches the meaning and intent behind a query, judged by the distance between their embeddings (numeric vectors) in a shared space. Closer vectors signal a stronger match, and the comparison is on meaning, not exact keyword overlap.
Semantic Triple A semantic triple is a three-part statement in subject-predicate-object form (for example, SEOBRO -> provides -> SEO services) that encodes one fact so machines can read it. It is the atomic unit of RDF and the building block of the knowledge graphs behind semantic and AI search.
share of answer Share of answer is the percentage of AI-generated answers, across a tracked set of prompts and models, that name your brand in the answer text, measured against competitors. It is the AI-search version of share of voice: how often you win the answer versus rivals.
source attribution Source attribution is how an AI answer engine credits the web pages behind its generated response, shown as inline citations or supporting links so a reader can trace each claim back to the page it came from.
Things Not Strings "Things, not strings" is Google's phrase for its 2012 shift, launched with the Knowledge Graph, from matching the literal keyword strings in a query to understanding the real-world entities behind it (people, places, brands, concepts) and how they relate to one another.
tokenization Tokenization is the process of splitting text into tokens, the small units (whole words, subwords, or single characters) that a language model actually reads and generates. Most modern LLMs, including GPT, use subword methods like byte-pair encoding (BPE), where roughly four characters of English map to one token.
training data Training data is the large corpus of text (and sometimes code, images, or other media) a machine learning model learns from during pretraining. The model adjusts its internal weights to predict patterns in it, and that absorbed knowledge is what it draws on when generating answers.
transformer A transformer is a neural network architecture that processes a whole sequence at once using self-attention, a mechanism that weighs how strongly each token relates to every other token. Introduced by Google researchers in 2017, it is the architecture behind modern large language models.
Unlinked Brand Mention An unlinked brand mention is a reference to your brand by name in someone else's content with no hyperlink attached. Search engines call these implied links or citations, and both classic search and AI answer engines can read them as entity and trust signals without a clickable link.
vector database A vector database stores data as embeddings (high-dimensional numeric vectors) and retrieves records by semantic similarity rather than exact keyword match. It powers the retrieval step in RAG pipelines that ground AI assistants like ChatGPT and Perplexity in outside documents.
vector search Vector search retrieves results by semantic similarity instead of exact keywords. Content and queries are turned into embeddings (high-dimensional numeric vectors), and the engine returns the stored vectors closest to your query vector by distance, its nearest neighbors.
voice search Voice search is when someone speaks a query aloud and a search engine or assistant (Google Assistant, Siri, Alexa) answers it, often by reading back a single result instead of showing a full page of links.
Wikidata Wikidata is a free, collaboratively edited knowledge base of structured, machine-readable facts about entities (people, brands, places, things), each identified by a unique Q-number like Q42. Released under a CC0 public-domain license, it is one of the open sources that feed Google's Knowledge Graph.