Glossary

The language of GEO

The words you'll meet in our reports and conversations, defined plainly — how AI engines choose what to say, and how we measure and move it for education brands.

The glass facade of a modern university building

How the pieces relate

The words below trace one path: a question becomes an answer that names its sources — which Besaid samples to measure your visibility.

Prompta question is askedRetrievalthe engine fetches sourcesAI answerit composes a replyCitationsand names its sourcesθmeasured over many samples

how an AI answer formswhat Besaid measures

(θ)

Measurementθ (Theta): Besaid's notation for visibility as a probability — the estimated chance that an AI answer to a given question mentions your brand. Writing it as θ is a reminder that it is an estimate drawn from samples, always reported with the sample size that supports it.; See alsoAI Visibility Sample Size Prompt Sampling

(A)

MethodAEO: Answer Engine Optimization — shaping content so that answer engines (AI assistants that reply directly instead of returning links) select and cite it. Often used interchangeably with GEO; AEO emphasizes the answer surface, GEO the generative model behind it.; See alsoGEO Answer Engine Citations
RetrievalAI Overviews: Google's AI-generated answer block that appears above the traditional search results, summarizing sources into a direct response. Being cited inside an AI Overview is a distinct visibility goal from ranking in the blue links below it.; See alsoAnswer Engine Citations Zero-Click
MeasurementAI Visibility: How present, and how favorably, a brand appears in the answers generative AI engines give. Unlike a search ranking it is probabilistic — the same question can yield different answers — so it is measured over many samples, not a single lookup.; See alsoθ (Theta)Share of Voice Sample Size
MethodAI Visibility Diagnostic: Besaid's free starting point: one real sampling run across the major AI engines on your brand, returned the same day as a report you can put in front of leadership — the AI's actual answers, how often you're recommended, and where to act first.; See alsoManaged GEO AI Visibility
RetrievalAnswer Engine: An AI system that responds to a question with a synthesized answer rather than a list of links — ChatGPT, Gemini, Perplexity, AI Overviews, DeepSeek. It is the unit that GEO optimizes for.; See alsoGEO AEO LLM

(B)

MeasurementBenchmark: A repeated, standardized measurement of visibility across a fixed set of questions and competitors, used to track movement over time and to place a brand against its category. A benchmark is only meaningful if the question set and sample size stay stable between runs.; See alsoShare of Voice Sample Size Prompt Sampling

(C)

RetrievalChunking: How a retrieval system splits a page into smaller passages before indexing them, so it can pull the most relevant fragment into an answer. Clear structure — headings, self-contained paragraphs — makes your content chunk cleanly, which makes the right part of it easier to retrieve and cite.; See alsoRetrieval RAG Grounding
ContentCitations: The sources an answer engine links to, or names, as the basis for its answer. Citations are the most direct lever in GEO: appearing as a cited source both drives referral traffic and raises the chance the engine repeats your framing.; See alsoGrounding Owned Media Zero-Click
ContentCrawler Access: Whether an AI company's crawler (GPTBot, ClaudeBot, Google-Extended, PerplexityBot and others) is permitted to fetch your pages — typically governed by robots.txt. If the crawler that feeds an engine's training or retrieval is blocked, the engine cannot ground its answers in your content.; See alsorobots.txt Grounding Freshness

(D)

RetrievalDeep Research: An engine mode that answers a question by running many searches over several minutes, reading dozens of sources, and composing a long, cited report — instead of replying in one shot. It widens the fan-out dramatically, so brands well outside the top results can still be pulled in if their content is clear and retrievable.; See alsoFan-out Queries Retrieval Citations

(E)

RetrievalEmbedding: A numerical representation of a piece of text that places it in a high-dimensional space where similar meanings sit close together. Retrieval systems embed both the question and your content, then match by nearness — which is why phrasing your pages the way real buyers phrase questions helps you get retrieved.; See alsoRetrieval RAG Knowledge Graph
ContentEntity: A distinct thing an engine recognizes and reasons about — a brand, product, person or place. AI engines resolve the words in a question to entities; being a well-defined, consistently described entity makes you easier to retrieve and harder to confuse with others.; See alsoKnowledge Graph Schema Markup Hallucination
MethodEvidence Chain: Besaid's principle that every metric traces back to the raw material behind it — the AI's verbatim answer, the exact prompt, the engine and model, the timestamp. It is what separates an auditable measurement from a black-box score.; See alsoPrompt Sampling Sample Size Mention

(F)

RetrievalFan-out Queries: The set of sub-questions an engine silently generates from a single user prompt, to retrieve broader context before it answers. Fan-out matters because your content may be pulled in through a reformulated query you never see.; See alsoRetrieval Deep Research Grounding
RetrievalFreshness: How recent the content is that an engine relies on when it answers. Some engines fetch live pages; others describe you from a months-old snapshot. Freshness explains why a change can show up quickly in one engine and lag in another — and why keeping key facts current matters.; See alsoGrounding Crawler Access Retrieval

(G)

MethodGEO: Generative Engine Optimization — the discipline of measuring and improving how generative AI engines represent a brand in their answers. The AI-era counterpart to SEO, but optimizing for synthesized answers and citations rather than ranked links.; See alsoAEO AI Visibility Answer Engine
RetrievalGrounding: Tying an AI's answer to specific external sources retrieved at query time, rather than to unverifiable memorized knowledge. Grounded answers cite where their claims come from — and grounded engines are where fresh, well-structured content can win visibility fastest.; See alsoRAG Retrieval Citations

(H)

ModelHallucination: When an AI states something false or unsupported with confidence — a wrong fact about your brand, an invented product, a competitor credited with your work. In GEO a hallucination is both a risk to monitor and, once caught with evidence, a correctable problem.; See alsoGrounding Entity Evidence Chain

(K)

ContentKnowledge Graph: A structured network of entities and the relationships between them that an engine draws on to answer factual questions. Consistent, machine-readable descriptions of your brand improve how it is represented in these graphs.; See alsoEntity Schema Markup Grounding

(L)

ModelLLM: Large Language Model — the neural network, trained on vast text, that gives a generative engine its ability to understand a question and compose an answer. The model behind names like GPT, Gemini and DeepSeek.; See alsoAnswer Engine Token Temperature
Contentllms.txt: A proposed plain-text file at a site's root that tells large language models what the site is, what matters on it, and how to describe it — llms.txt is to AI engines what robots.txt is to search crawlers. Besaid publishes its own as a demonstration.; See alsorobots.txt Crawler Access Grounding

(M)

MethodManaged GEO: GEO delivered as a service rather than a tool: Besaid's team runs the whole program — measuring how AI talks about you, doing the optimization work, and proving the change with before-and-after evidence. You read reports; you never operate a dashboard.; See alsoGEO AI Visibility Diagnostic
MeasurementMention: A single instance of your brand appearing in an AI's answer, whether or not it is linked as a citation. Counting mentions across many samples is the raw material behind visibility and share-of-voice metrics.; See alsoCitations Position Share of Voice

(O)

ContentOwned Media: Content on domains you control — your site, your help center, your blog — as opposed to third-party pages that talk about you. Owned-territory share, the fraction of an engine's citations that point to your own domains, is the most direct battleground of GEO: it is the content you can change tomorrow.; See alsoCitations Share of Voice Schema Markup

(P)

MeasurementPersona: The user identity a question is asked from when sampling an engine — for example a prospective undergraduate at home versus a parent researching abroad. Because engines tailor answers to who seems to be asking, persona is part of a vantage and changes the story a measurement tells.; See alsoVantage Prompt Sampling Sentiment
MeasurementPosition: Where your brand falls in the order of an AI's answer — named first, buried third, or listed last. Because readers weight what comes first, position is tracked alongside whether you are mentioned at all.; See alsoMention Share of Voice Sentiment
MeasurementPrompt Sampling: Asking an engine the same or related questions many times, across models and conditions, to estimate a probabilistic answer rather than trust a single reply. It is the method that turns a chat interface into a measurement instrument — and its rigor is only as good as its disclosed sample size.; See alsoSample Size θ (Theta)Temperature

(R)

RetrievalRAG: Retrieval-Augmented Generation — an architecture in which an engine first retrieves relevant documents, then generates its answer from them. RAG is why publishing clear, retrievable content can influence answers without any change to the underlying model.; See alsoRetrieval Grounding Chunking
RetrievalRetrieval: The step in which an engine fetches relevant documents at query time before it writes an answer — by live search, a vector index, or both. Retrieval is the moment your content either makes it into the answer or does not, which makes being retrievable the practical goal of most GEO work.; See alsoRAG Embedding Chunking
Contentrobots.txt: The plain-text file at a site's root that tells automated crawlers which paths they may fetch. In the AI era it governs whether engine crawlers can read you at all — block the wrong bot and you remove yourself from that engine's grounding, often without realizing it.; See alsoCrawler Access llms.txt Grounding

(S)

MeasurementSample Size: The number of samples — prompts, engines, runs — behind a reported metric, and the denominator that makes it trustworthy. A visibility figure without its sample size is a number without a scale; Besaid attaches one to every percentage.; See alsoθ (Theta)Prompt Sampling Benchmark
ContentSchema Markup: Structured data (often Schema.org / JSON-LD) added to a page that spells out, in machine-readable form, what the page is about — an organization, a course, a review. It helps engines resolve your content to the right entity and cite it correctly, and is one of the most reliable owned-media levers.; See alsoEntity Knowledge Graph Owned Media
MeasurementSentiment: The tone an engine takes toward your brand when it mentions you — positive, neutral, cautious, negative. Because an answer can name you and still frame you unfavorably, sentiment is measured separately from whether you appear.; See alsoMention Position Hallucination
MeasurementShare of Voice: The proportion of relevant AI answers in which your brand appears, measured against the competitors in the same set — your slice of the total conversation. It reframes visibility as a competitive standing, not an absolute count.; See alsoMention Benchmark AI Visibility
ModelSystem Prompt: The hidden instructions an engine's operator gives the model before your question reaches it — setting its persona, guardrails and default style. You cannot see it, but it shapes every answer, which is one reason the same model can behave differently inside different products.; See alsoLLM Prompt Sampling Temperature

(T)

ModelTemperature: A setting that controls how random an engine's output is: low temperature makes answers more repeatable, high temperature more varied. It is one reason the same question yields different replies, and one reason visibility must be estimated over samples rather than read from a single run.; See alsoPrompt Sampling Sample Size LLM
ModelToken: The unit of text a language model reads and writes — roughly a word-piece. Models process and price by the token, and their context is capped in tokens, which is why concise, well-structured content is easier for an engine to take in and reuse.; See alsoLLM Chunking Embedding

(V)

MeasurementVantage: The perspective a question is asked from — the language, location and persona simulated or genuinely used when sampling an engine. Answers shift with vantage, so a home-market view and a real overseas node can tell very different stories about the same brand.; See alsoPersona Prompt Sampling Sentiment

(Z)

MeasurementZero-Click: When an AI answers a question completely enough that the user never clicks through to a source. Zero-click is why being mentioned and described accurately inside the answer now matters as much as — sometimes more than — earning the click, since the click may never come.; See alsoCitations AI Visibility Answer Engine

See what AI is telling your future students.

Get a free diagnostic