Glossary
The terms, with their sources
Sixteen terms you will meet in this field. Each definition stands on its own, so it can be quoted without the ones around it. Each term that names a real thing links to that thing.
Every external link here was opened and checked, one by one, before being published. Two terms have no external entity of their own and say so instead of pointing at something merely similar.
GEO (Generative Engine Optimization)
The practice of preparing a site so that generative answer engines can extract, attribute and quote it. It works on the mechanical properties an engine reads — structure, entities, crawler access, rendering — not on the engine's decision to cite.
No external entity yetSix reproducible experiments →
SEO (Search Engine Optimization)
The practice of making a site findable and rankable in classic search engines. It is the foundation GEO sits on: a page an engine cannot crawl or render is a page it cannot quote either.
llms.txt
A plain-text file at the root of a site that lists its pages for large language models, in the spirit of robots.txt or sitemap.xml. It is a proposed convention, not a standard: no major engine has committed to reading it.
No external entity yetThe one this site publishes →
Schema.org
A shared vocabulary for describing things on a page — a person, a service, an article — so a machine reads them as entities and not as decorated text. Maintained jointly by Google, Microsoft, Yahoo and Yandex.
JSON-LD
The JSON format used to embed structured data in a page as a separate block, instead of scattering attributes through the markup. It is the format Google recommends and the one this site uses.
robots.txt
A file at the root of a site telling automated clients which paths they may fetch. A line too many here is one of the cheapest ways to become invisible: an AI crawler that is blocked has nothing of yours to quote.
Crawler
A program that fetches pages automatically to index or feed them to another system. GPTBot, ClaudeBot and PerplexityBot are crawlers; Google-Extended is not — it is a robots.txt token that governs training and grounding, and it does not affect Google Search.
LLM (Large Language Model)
A model trained on large amounts of text that predicts the next token. What it produces is not a lookup of stored facts, which is why the same question asked twice can come back with different sources.
RAG (Retrieval-Augmented Generation)
Fetching documents at answer time and giving them to the model as context, so the answer can cite sources instead of relying only on training. It is the mechanism behind most answers that show citations.
Knowledge graph
A network of entities and the relations between them, rather than a bag of documents. Declaring your entities with stable identifiers is what lets a machine tell your brand apart from another with the same name.
Entity linking
Tying a name on a page to the real-world thing it refers to, using an identifier a machine can resolve. In Schema.org this is what the sameAs property is for — and it is why every term here that has a real entity links to it.
Sitemap
An XML file listing the URLs of a site with their last modification date, so a crawler does not have to discover them by following links. It declares what exists; it does not promise anything about ranking.
Canonical URL
The address a page declares as its own official one when the same content can be reached several ways. Without it, the same text competing with itself splits whatever signal it had.
Jaccard index
A measure of how much two sets overlap, from 0 to 1. It is the measure used to compare which sources an answer engine cites across two identical runs — and the overlap is far below 1, which is why one run is not a measurement.
Confidence interval
The range a measured value is expected to fall in, given how many times it was measured. A percentage published without one — no run count, no spread — is not a measurement; it is an opinion with decimals.
DOI (Digital Object Identifier)
A permanent identifier for a document or dataset that keeps resolving even if the file moves. It is what turns "we measured this" into something a third party can fetch and check years later.