🚀 OrcaScope private beta — first 10 design partners get 3 months free Join now!

Guides2026-07-24~4 min read阅读中文版 →

The GEO glossary: 30 terms, bilingual

From GEO and AEO to mention rate, llms.txt and prompt injection — one-line bilingual definitions of 30 core terms in the field.

Working in GEO means colliding with SEO jargon, AI terminology and product slang all at once. Here are 30 core terms, one line each, bilingual — for aligning your own team, and you are welcome to cite it.

Fundamentals

  • GEO (Generative Engine Optimization / 生成式引擎优化) — the practice of getting a brand accurately mentioned, recommended and cited inside generative AI engines' answers. See What is GEO.
  • AEO (Answer Engine Optimization / 答案引擎优化) — GEO's near-synonym with earlier roots (the featured-snippet and voice-assistant era); in practice the two overlap almost entirely.
  • SEO (Search Engine Optimization / 搜索引擎优化) — the thirty-year-old discipline of ranking on results pages; GEO's foundation, not its rival.
  • AI visibility (AI 可见度) — the degree to which a brand is mentioned, recommended and cited in AI answers; can be normalized into a score and tracked over time.
  • Generative engine (生成式引擎) — an AI system that retrieves from the web, synthesizes multiple sources and generates a natural-language answer; the academic term for ChatGPT-class systems.
  • Answer engine (答案引擎) — a retrieval system that returns an answer rather than a list of links; the predecessor and synonym of the generative engine.
  • AI-native search (AI 原生搜索) — search products whose core interaction is an AI-generated answer (e.g. Metaso, Perplexity).
  • Zero-click (零点击) — the user gets what they need inside the answer and clicks nothing; the default shape of brand exposure in the GEO era.

Engines and crawling

  • LLM (large language model / 大语言模型) — the model underneath a generative engine. A model is not a product entrance: the same model can power many products.
  • RAG (retrieval-augmented generation / 检索增强生成) — retrieve first, then answer from the retrieved material; it makes "being retrievable" the precondition of "being mentioned".
  • Web-connected retrieval (联网检索) — an engine's ability to fetch live web pages; behaviour varies enormously between engines.
  • AI crawler (AI 爬虫) — the bots AI vendors use to fetch web pages (GPTBot, ClaudeBot and friends); whether robots.txt lets them in directly affects whether you can appear in answers.
  • llms.txt — a convention for a site-description file at the web root, aimed at AI engines, so one fetch explains what the site is (ours is at /llms.txt).
  • Knowledge cutoff (知识截止) — the time boundary of a model's training corpus; brands and facts newer than it can only enter answers via web retrieval.
  • Hallucination (幻觉) — a model generating false information; brands should monitor not just "were we mentioned" but "what was said wrongly".
  • MCP (Model Context Protocol / 模型上下文协议) — an open protocol that lets AI agents call external tools and data in a standard way (our MCP endpoint is documented on the developers page).

Measurement and metrics

  • Monitoring prompt (监测问题) — a fixed set of questions sent to engines repeatedly under a fixed methodology; the basic unit of continuous measurement.
  • Mention (提及) — the brand name (including aliases) appears in the answer body. Our detection is tone-neutral — a mention is a mention.
  • Citation (引用) — a reference link attached to an answer; counted per domain, it reveals "who the AI considers authoritative".
  • Mention rate (提及率) — the share of monitoring prompts in which the brand is mentioned; normalized, it becomes the engine sub-score.
  • Recommendation rank (推荐位次) — the brand's position in an answer's recommendation list; first and fifth are both "mentions", with very different value.
  • Sample size (样本量) — the number of samples behind a score; flagging "not enough data" honestly beats faking precision (our rules: how we score).
  • Score version (评分版本) — the version number of the scoring methodology; historical batches keep their version tag so trends compare like with like.

Content and strategy

  • Pillar content (支柱内容) — a deep, systematic answer to one core question; a high-frequency citation source type for AI answers.
  • Structured data (结构化数据) — Schema.org markup (Organization, FAQPage, Article…) that makes a page machine-readable to engines.
  • Authoritative source (权威来源) — the third parties engines prefer to cite: media, Q&A communities, industry lists; China and the West run two separate ecosystems (see our observations).
  • Brand alias (品牌别名) — every name a brand goes by: local and English names, product names, nicknames. Monitoring without an alias table systematically understates visibility.
  • White-hat GEO (白帽 GEO) — improving visibility only through public, sustainable means: content, structure, authority building, continuous measurement.
  • Prompt injection (提示注入) — hiding instructions in web pages to manipulate AI answers — a black-hat tactic. We don't do it and don't recommend it: engine vendors will claw it back.
  • Bilingual content assets (双语内容资产) — content estates built separately for the Chinese and Western sides; translation is not localization, because the two citation ecosystems differ.

Related reading

Theory read — now measure reality: what do AI answers say about your brand today?

Run my free checkup

How is the score computed? Full methodology →

The GEO glossary: 30 terms, bilingual · OrcaScope