AI Visibility Glossary

Adaptive RAG

RAG Signal's proprietary 5-step methodology (MAP → BUILD → WEIGHT → REINFORCE → MEASURE) for engineering brand visibility in AI retrieval systems. Unlike standard RAG, Adaptive RAG continuously adjusts to model updates and competitive changes.

AI Visibility

Whether a brand appears and is cited in AI model responses (ChatGPT, Claude, Perplexity, Gemini, etc.) when users ask relevant prompts. Measured by Citation Rate.

Brand Memory

The structured knowledge representation of a brand that AI models retrieve when processing prompts. Comprises entity definitions, factual assertions, relationship maps, and citation anchors.

Citation

When an AI model mentions a brand by name in its response. A citation can be positive, neutral, or negative. Citation + positive framing = the goal state.

Citation Anchor

A high-authority, high-visibility source that AI models reference when retrieving brand information. Similar to backlinks in traditional SEO but measured by retrieval impact rather than link equity.

Citation Delta

The change in Citation Rate between two measurement points, typically reported at 30/60/90 days. The primary success metric for RAG Signal sprints.

Citation Rate

The percentage of relevant buyer prompts where an AI model cites a specific brand. Formula: (Prompts Citing Brand ÷ Total Prompts Tracked) × 100.

Cross-Model Persistence

Whether a brand is cited consistently for the same prompt across different AI models. High persistence = strong Brand Memory. Low persistence = model-dependent retrieval.

Entity

A defined person, organization, place, product, or concept in structured data. AI models use entities to build knowledge graphs that power retrieval.

Entity Intelligence

RAG Signal's proprietary technology for mapping, constructing, and weighting brand entities for AI retrieval optimization.

llms.txt

A proposed standard (by Anthropic) for websites to provide structured, AI-readable information about their content. Used by Claude to improve retrieval accuracy.

MAP-BUILD-WEIGHT-REINFORCE-MEASURE

The 5-step Adaptive RAG methodology. Each phase feeds the next: MAP discovers prompts, BUILD constructs Brand Memory, WEIGHT scores signals, REINFORCE deploys across models, MEASURE tracks Citation Delta.

Model Update

When an AI model is upgraded or retrained. Model updates can reset citation patterns — brands without continuous reinforcement can drop to zero overnight.

Prompt Discovery

The process of identifying real buyer questions where a brand should be cited. Uses GSC data, competitor citation analysis, LLM query patterns, and category research.

RAG (Retrieval-Augmented Generation)

The architecture AI models use to retrieve information from knowledge bases before generating responses. The foundation of how AI decides which brands to cite.

RAG Scoring Algorithm

RAG Signal's proprietary 7-dimension signal evaluation system: source authority (25%), factual consistency (20%), entity linkage (15%), cross-model persistence (12%), temporal relevance (12%), citation frequency (10%), competitive differentiation (6%).

Signal Decay

The gradual degradation of Brand Memory signals over time, especially after model updates. Monitored and corrected by the retainer service.

Zero-Click Retrieval

When an AI model answers a user's question directly — citing sources by name — without providing clickable links. This is why Citation Rate matters more than click-through rate for AI visibility.

GEO (Generative Engine Optimization)

The practice of optimizing brand content so AI models (ChatGPT, Claude, Perplexity, Gemini) retrieve and cite it in generated answers. Unlike SEO — which targets search engine result pages — GEO targets the retrieval layer of generative engines.

Tokenization

The process of splitting text into tokens — the smallest units AI models process (roughly word fragments). How your content is tokenized affects chunking, context limits, and retrieval quality in RAG systems.

Semantic Search

Search based on meaning rather than exact keywords. Semantic search matches queries to content through vector similarity, which is why AI models can retrieve your brand for prompts that never contain your brand name.

Vector Database

A database that stores content as high-dimensional vectors (embeddings) for similarity search. RAG systems query vector databases to find the most relevant passages to cite when generating answers.

Embedding

A numerical vector representation of text that captures its meaning. Embeddings let AI models compare the semantic similarity between a user query and your content — the foundation of modern retrieval.

Hallucination

When an AI model generates plausible but factually incorrect information. Grounding answers in retrievable, well-structured content reduces hallucination — and brands with strong source signals get cited instead.

Knowledge Graph

A structured network of entities and their relationships (people, organizations, products, concepts). AI models build and query knowledge graphs to determine which brands are authoritative and how they connect.

Chunking

The practice of splitting long content into smaller retrievable passages. Chunk quality determines whether an AI model can find a complete, self-contained answer in your content — a core audit step in the RAG-Ready Data process.

Context Window

The maximum amount of text (in tokens) an AI model can consider when generating a response. Limited context windows make retrieval quality critical — only the most relevant chunks get cited.

Temperature

A model parameter controlling output randomness — lower values produce more deterministic, factual responses. Higher temperature settings amplify retrieval differences between brands, making source quality more decisive.

Fine-Tuning

Training a base AI model further on specific data to specialize its behavior. While most public models aren't fine-tuned per brand, retrieval optimization (Brand Memory) achieves similar visibility effects without model access.

What is Adaptive RAG?

Adaptive RAG is RAG Signal's proprietary 5-step methodology (MAP → BUILD → WEIGHT → REINFORCE → MEASURE) for engineering brand visibility in AI retrieval systems. Unlike standard RAG, Adaptive RAG continuously adjusts to model updates and competitive changes.

What is Signal Weighting?

Signal Weighting is the third phase of Adaptive RAG, where RAG Signal scores brand signals across 7 dimensions: source authority (25%), factual consistency (20%), entity linkage (15%), cross-model persistence (12%), temporal relevance (12%), citation frequency (10%), and competitive differentiation (6%).

What is Citation Rate?

Citation Rate is the percentage of relevant buyer prompts where an AI model cites a specific brand. Formula: (Prompts Citing Brand ÷ Total Prompts Tracked) × 100.

What is Brand Memory?

Brand Memory is the structured knowledge representation of a brand that AI models retrieve when processing prompts. It comprises entity definitions, factual assertions, relationship maps, and citation anchors.