AI Visibility Glossary: 30 Terms Defined
The vocabulary around AI search is inconsistent, and several terms are used to mean different things by different vendors. These are plain definitions, each written to stand alone.
Core metrics
Share of Answer (SoA) — A brand visibility metric that measures the proportion of AI-generated responses mentioning a brand for questions relevant to its category. It is the answer-era analogue of share of voice.
Unaided visibility — Brand visibility measured with prompts that never name the brand. It measures genuine discovery, and it is the more meaningful of the two visibility measures.
Aided visibility — Brand visibility measured with prompts that do name the brand. It measures how a brand is characterised once already known, not whether it would be discovered.
Citation rate — How often a specific page or domain is cited as a source in AI responses, as distinct from how often the brand is mentioned.
Competitive answer set — The group of brands that appear alongside yours in AI responses. It frequently differs from a company's assumed competitive set.
Description accuracy — Whether AI systems describe a brand correctly. Inaccurate mentions can be more damaging than absence.
Disciplines
AEO (Answer Engine Optimization) — The practice of structuring content so answer engines can extract and reuse it. Focuses on passage-level clarity, direct answers, and structured data.
GEO (Generative Engine Optimization) — The practice of improving brand presence in AI-generated responses. Broader than AEO, encompassing entity signals, authority, and third-party presence.
SEO (Search Engine Optimization) — The practice of improving visibility in traditional search results, measured by ranking position. Still the foundation, since several AI systems ground answers in indexed web content.
Entity SEO — Work aimed at making a brand recognisable as a distinct entity rather than a keyword string. Centres on naming consistency, verified profiles, and factual coherence.
How the systems work
LLM (Large Language Model) — A model trained on large text corpora that generates responses by predicting likely continuations. The engine underneath modern AI assistants.
Grounding — Anchoring a generated response in retrieved source documents rather than relying only on training data. Grounded responses typically carry citations.
RAG (Retrieval-Augmented Generation) — An architecture that retrieves relevant documents at query time and supplies them to the model as context. This is why fresh, well-structured, publicly reachable content can influence answers without the model being retrained.
Passage-level retrieval — Retrieving individual paragraphs or sections rather than whole pages. It is why a single well-written paragraph can earn a citation for an otherwise unremarkable page.
Training data cutoff — The date beyond which a model has no built-in knowledge. Content published after the cutoff can still influence answers through retrieval.
Hallucination — A confident, fluent, factually wrong output. For brands, the common form is an assistant inventing details about your product or pricing.
Non-determinism — The property that identical prompts can produce different responses. It is why single-run measurement is unreliable and repeated sampling is necessary.
Surfaces
AI Overview — Google's AI-generated summary appearing above traditional results, with links to the sources used.
Answer engine — Any system that returns a direct answer rather than a list of links.
AI assistant — A conversational interface built on an LLM, such as ChatGPT, Claude, Gemini, or Perplexity.
Zero-click — A search interaction resolved without visiting any website. AI answers increase the proportion of these substantially.
Technical
llms.txt — A plaintext file at a domain root that provides AI systems with a curated summary of a site and its key facts. Distinct from robots.txt: it curates information rather than controlling access.
robots.txt — A file specifying which crawlers may access which paths. Now commonly used to allow or block AI crawlers specifically.
AI crawler — A bot that fetches web content for AI systems, whether for training, live retrieval, or indexing. Different crawlers serve different purposes and can be permitted selectively.
Structured data — Machine-readable markup, usually JSON-LD, describing what a page contains. Helps systems parse content and attribute it correctly.
Schema.org — The shared vocabulary used by structured data, defining types such as Organization, Article, Product, and FAQPage.
sameAs — A schema property linking an entity to its other verified profiles. A sameAs URL that does not resolve is a broken identity claim and does active harm.
Entity disambiguation — Determining which real-world entity a name refers to. Where two organisations share a name, weak disambiguation causes systems to omit both rather than risk error.
Knowledge graph — A structured database of entities and their relationships, used to resolve references and supply factual grounding.
Content concepts
Quotable passage — A self-contained span of text that can be lifted into an answer without losing meaning. The practical unit of AI citation.
Self-contained content — Content whose sections make sense in isolation, without pronouns or references pointing outside the passage. The single most important structural property for citation.
Original evidence — Data, research, or observation that exists in only one place. The strongest differentiator for citation, because there is no substitute source.
E-E-A-T — Experience, Expertise, Authoritativeness, Trustworthiness. A framework originally written for human quality raters, now useful as a checklist of credibility signals that machines can detect.
Frequently asked questions
What is the difference between AEO and GEO? AEO is about content structure and extractability. GEO is broader, covering entity signals, authority, and off-domain presence as well. In practice the terms are often used interchangeably.
Is Share of Answer a standard metric? The concept is widely shared; the exact computation differs between vendors. Compare trends within one tool rather than absolute values across tools.
Do I need llms.txt? It is cheap, low-risk, and increasingly consulted. Worth having, but not a substitute for good content and clear entity signals.
Is GEO replacing SEO? No. Several AI systems ground answers in indexed web content, so SEO remains the foundation. GEO adds a layer; it does not replace one.
Published by the SIQA Editorial Team.