# AI search glossary

> Short, plain-English definitions for the AEO, GEO, and AI-search terms you will run into.

By Marius C. Paun · Reviewed by Joey Kudish · Published 2026-07-27 · Updated 2026-07-28

A quick reference for the terms that come up across these guides and across the AI search world. Definitions are kept short and plain. Where a term is used loosely in the industry, this says so, because pretending these are all settled would be its own kind of jargon.

## The big three (and their cousins)

**SEO (Search Engine Optimization)** is the long-standing practice of earning visibility in a ranked list of search results, the model where users click through to a site.

**AEO (Answer Engine Optimization)** is optimizing your content so AI answer engines understand, trust, and cite you inside their written answers. It is a practitioner term with no single canonical definition. See [what is AEO](/guides/what-is-aeo).

**GEO (Generative Engine Optimization)** is the same idea, focused on generative AI responses. It is the one term with an academic origin, coined in a [2023 research paper](https://arxiv.org/abs/2311.09735). In practice, AEO and GEO are used almost interchangeably; do not over-invest in the distinction.

**Answer engine** is a tool that returns a direct, synthesized answer with citations instead of a list of links. Perplexity describes itself as one; ChatGPT search and Google's AI surfaces behave like ones.

**AI Overviews / AI Mode (Google)** are Google's AI search surfaces. AI Overviews are the AI summaries shown above traditional results (launched in the US in May 2024). AI Mode is a more conversational version for complex, multi-part questions.

## How the engines work

**LLM (large language model)** is the kind of AI model, trained on huge amounts of text, that understands and generates language. It is what powers every answer engine.

**RAG (retrieval-augmented generation)** is the retrieve-then-answer pattern: the system fetches relevant pages first, then writes an answer grounded in them, usually with citations. Named in a [2020 paper](https://arxiv.org/abs/2005.11401). See [how AI engines work](/guides/how-ai-engines-work).

**Grounding** means tying an AI's answer to retrieved, verifiable sources rather than its memory alone, to keep it accurate and citable.

**Hallucination** is when an AI states something false but confident. Grounding and citations are the main ways engines reduce it (and let you check it).

**Token / tokenisation** : a token is the small chunk of text a model processes (roughly four characters, or about three-quarters of a word). Tokenisation is splitting text into those units.

**Context window** is the maximum amount of text, measured in tokens, a model can consider at once. It is shared across the question, the retrieved pages, and the answer, which is why concise, well-structured content helps.

**Embeddings / semantic search** : an embedding turns text into a list of numbers (a vector) that captures its meaning, so systems can match content by meaning rather than exact keywords.

**Query fan-out** is Google's term for breaking one question into multiple related sub-queries, running them at once, and combining the results. The high-level mechanism is Google's; the finer detail is partly industry interpretation.

## Understanding and trust

**Knowledge graph** is a structured map of real-world entities (people, places, organizations) and how they relate, the "things, not strings" idea Google [introduced in 2012](https://blog.google/products/search/introducing-knowledge-graph-things-not/).

**Entity** is a distinct, recognized "thing" an engine can identify and connect to other things, the unit of meaning behind your brand.

**Structured data / schema.org** is machine-readable markup that labels what your content means. Schema.org is the shared vocabulary; JSON-LD is the recommended format. See [JSON-LD structured data](/guides/json-ld).

**E-E-A-T** stands for Experience, Expertise, Authoritativeness, and Trust, the criteria Google's human quality raters use to judge content. Important: it is not a direct ranking factor and there is no "E-E-A-T score." See [authority and proof](/guides/authority-and-proof).

**sameAs** is the structured-data property that links your site to your authoritative profiles (LinkedIn, Crunchbase, Wikidata) so engines can confirm your identity.

## Crawling and discovery

**Crawler / user agent** : a crawler is automated software that fetches web pages; its user agent is the name it identifies itself with (like `GPTBot` or `OAI-SearchBot`), which you can allow or block. See [AI crawler access](/guides/robots-ai-bots).

**llms.txt** is a proposed Markdown file at your site root that gives AI models a curated map of your key pages. It is an emerging convention, not an adopted standard, and no major AI engine has confirmed it uses one. See [llms.txt](/guides/llms-txt).

## Visibility and measurement

**Citation vs mention** : a mention names your brand in an answer; a citation links to or attributes your page as a source. Tools count them differently.

**Share of voice** is how often you appear in AI answers relative to competitors, for a defined set of prompts. The concept is borrowed from traditional marketing; each tool computes it its own way.

**Zero-click search** is a search answered right on the results page, with no click through to a website. AI answers are accelerating it. See [measuring your AI visibility](/guides/measuring-ai-visibility).