Working GEO in 2026 demands a hybrid vocabulary: classic SEO, academic NLP, European regulation, and AI-platform product knowledge. This glossary gathers the 100 terms that appear most often in briefings, contracts, and team meetings. For extended definitions and per-term examples, see GEOMOND's complete glossary.
Models and architectures (1-20)
LLM: large language model (GPT-4, Claude, Gemini). Transformer: base architecture of modern LLMs. Token: minimum processed unit (~0.75 words in English). Context window: amount of tokens the model processes per query. Parameters: model weights; GPT-4 has hundreds of billions. Fine-tuning: model adjustment with own data. RLHF: reinforcement learning with human feedback. RAG: Retrieval Augmented Generation. Embedding: vector representation of text. Vector DB: database for embeddings (Pinecone, Weaviate). Multimodal: model that processes text + image + voice + video. AI agent: autonomous system executing multi-step tasks. MoE: Mixture of Experts, modular architecture. SLM: Small Language Model (Llama 3 8B, Phi-3). Open-weight: model whose weights are publicly distributed. Closed-source: model accessible only via API (GPT-4). Inference: model execution in response to input. Latency: time from request to first response token. Throughput: tokens per second. Hallucination: response invented by the model.
GEO metrics (21-40)
Presence Index: probability of citation in AI responses (0-100). Share of Voice: share of mentions in the sector. Answer share: percentage of prompts where you appear. Citation rate: rate of citations with link. Mention sentiment: tone of mention (positive, neutral, negative). Position in answer: order within the AI response. Brand recall AI: spontaneous recall in AI query. Co-mention: joint appearance with competitor. Source dominance: weight of your domain as a source. AIO inclusion rate: % of Google searches with AI Overview that include you. AI crawl rate: AI bot crawl frequency (GPTBot, PerplexityBot). Disambiguation score: clarity of brand identity for the model. Topic authority: topical authority measured in the AI corpus. Knowledge graph presence: existence of entity in public graphs. Wiki coverage: coverage on Wikipedia/Wikidata. NER recall: rate at which your brand is recognized as an entity. Prompt coverage: number of sector prompts where you appear. Conversational depth: how much your brand is explored in multi-turn. Answer share trend: temporal evolution of answer share. SAIDR: Search-AI Decoupling Rate, emerging metric.
Techniques and operations (41-65)
Prompt: input sent to the model. System prompt: persistent instruction. Prompt engineering: design of effective prompts. Few-shot: examples in the prompt. Zero-shot: no prior examples. Chain of thought: step-by-step reasoning. Schema.org: structured-data vocabulary. JSON-LD: serialization format for Schema.org. FAQPage schema: AI-optimized FAQ markup. Article schema: article markup. Organization schema: brand-entity markup. Author schema: authorship markup. HowTo schema: procedure markup. llms.txt: emerging file to guide LLMs. robots.txt: crawler-control file. Canonical tag: preferred URL indicator. Hreflang: language version indicator. Pillar content: cluster trunk content. Topic cluster: thematic content grouping. Internal linking: semantic internal linking. External authority: weighted external mentions. Backlink: incoming link. Brand backlink: link with brand-text. Citation: mention without link. Mention monitoring: monitoring of mentions.
Platforms and models (66-85)
ChatGPT, GPT-4, GPT-4o, GPT-5: OpenAI family. Claude, Claude 3.5 Sonnet, Claude Haiku: Anthropic family. Gemini, Gemini 2.0, Gemini Pro: Google DeepMind family. Perplexity: cited-answer engine. Copilot: Microsoft assistant (over OpenAI). Grok: xAI model. DeepSeek: Chinese open-source family. Llama: Meta open-source family. Mistral: French open-source family. SearchGPT: OpenAI search product. AI Overviews: AI responses in Google. Featured snippets: classic featured snippets. SGE: Search Generative Experience (precursor to AIO). Bing Chat: Bing assistant. You.com: answer engine. Operator: OpenAI agent. Comet: Perplexity agentic browser. Pulse: ChatGPT proactive agent. Agent Mode: Gemini agent mode. Computer Use: Anthropic API for desktop control.
Regulation and infrastructure (86-100)
AI Act: European AI Regulation. GDPR: European data-protection regulation. C2PA: content provenance standard. Watermarking: synthetic-content watermark. Data poisoning: deliberate corpus contamination. Adversarial prompt: prompt designed to break alignment. Jailbreak: bypass of model restrictions. Alignment: model alignment with human values. Red teaming: adversarial testing of models. Model card: official model technical sheet. Inference cost: per-query model cost. API rate limit: requests-per-time-unit limit. Webhook: automatic HTTP notification. Function calling: function invocation from the model. MCP: Model Context Protocol, emerging Anthropic standard.
This glossary updates quarterly. Request the free audit at GEOMOND and receive a personalized version with the most relevant terms for your sector.
Frequently asked questions
Why do I need to know 100 GEO terms?
Because the field evolves fast and the technical vocabulary (RAG, embeddings, prompt injection, AI Overviews, llms.txt, agentic search) is the foundation for understanding reports, briefings and agency proposals. An updated glossary avoids costly misunderstandings.
Which 5 terms are most important to start with?
Presencia Index, RAG, Schema.org, AI Overviews and AI agent. With those five you understand 80% of technical GEO conversations in 2026 with any agency or in-house team.
Is this glossary updated as AI evolves?
Yes, we revise it quarterly. Terms like "prompt engineering" lose weight while "agentic workflow" or "context window optimization" gain it. We recommend reviewing the glossary every time you start a new project phase.