Appendix D — Glossary

AEO / GEO — Answer engine optimization, generative engine optimization. Used interchangeably. The practice of getting a brand included and well-positioned in AI-generated answers.

Chunk — A segment of a document, typically a few hundred tokens, indexed independently. The actual unit of retrieval.

Citation rate — Share of tracked questions where your domain appears as a source. Different from mention rate.

Entity — A specific, identifiable thing — a company, person, product — that a knowledge graph can resolve and attach facts to.

Ghost citation — An appearance where your site is cited as a source but your brand is never named in the answer. Roughly 62 percent of appearances.

Grounding — Checking generated claims against retrieved sources. Produces support scores, and can suppress low-confidence output entirely.

Information gain — What a passage offers that the passage it’s compared against does not. The one content principle that survives the evidence.

Mention rate — Share of tracked questions where your brand is named in the answer. The recommendation metric.

Query fan-out — Breaking one question into several sub-questions run in parallel. Confirmed by Google; the number of sub-questions is not public.

RAG — Retrieval-augmented generation. Retrieve relevant passages, put them in the model’s context, generate an answer grounded in them.

Retrieval threshold — The score below which a system answers from its own knowledge without searching. Google’s documented default is 0.3.

Share of voice — How often and how prominently a brand appears in AI answers relative to competitors. Widely sold, rarely sampled adequately.

UCP / ACP / MCP / A2A — Universal Commerce Protocol (Google), Agentic Commerce Protocol (OpenAI and Stripe), Model Context Protocol (Anthropic-originated), Agent2Agent (Linux Foundation).