Glossary
Clear definitions and explanations of key terms in AI search, GEO, AEO, and digital marketing.
What is the Universal Commerce Protocol (UCP)? A complete explanation
The Universal Commerce Protocol (UCP) is the open standard that lets AI agents shop across stores with a single integration. Here is what UCP actually is, who is behind it, and how it differs from OpenAI's Agentic Commerce Protocol (ACP).
Read moreWhat is an MCP server? How AI connects to tools and data
The Model Context Protocol (MCP) is the open standard that lets AI models connect to external tools and data sources. An MCP server is the part that actually exposes those tools, data and templates to the model. Here is the difference between the protocol and the server, how tools, resources and prompts work, and why CitationLab publishes its own MCP servers.
Read moreWhat is an AI agent? Agentic AI explained
An AI agent is a system that plans, uses tools and completes multi-step tasks toward a goal — unlike a chatbot that just answers one question at a time. Here is what agentic AI actually is, how autonomy varies, and why it changes how brands get discovered and recommended.
Read moreWhat is NLWeb? Making your website agent-ready
NLWeb is an open project from Microsoft that makes websites conversational and queryable by AI agents — using schema.org and other existing web formats together with the Model Context Protocol. Every NLWeb instance can act as an MCP server. Here is what NLWeb is, how it relates to agent-ready websites, and what CitationLab does with it.
Read moreWhat is an agent-ready website? agents.md and llms.txt explained
An agent-ready website is built so AI agents can read, understand and act on its content — not just humans. It comes down to semantic HTML, schema.org, llms.txt, agents.md and clean APIs. Here is what each part does, why it decides whether AI cites you, and how Citation Sites are built agent-ready from the start.
Read moreWhat are agent payments? A2A and AP2 explained
Agent payments are when AI agents complete purchases and payments on the user's behalf — and they require new protocols for trust and identity. Agent2Agent (A2A) lets agents talk to each other; Agent Payments Protocol (AP2) proves a purchase is really what the user asked for. Here is how they work and how they relate to MCP, UCP and ACP.
Read moreWhat is grounding in AI answers? How AI ties claims to sources
Grounding is the technique where an AI model ties its claims to concrete, retrieved sources rather than generating answers purely from parametric memory. It is the mechanism that makes AI answers verifiable — and that produces citations. Here is what grounding is, why it reduces hallucination, and what it takes for your content to become the thing the model anchors its answer in.
Read moreWhat are embeddings and vector databases? How AI understands meaning
Embeddings are how AI models represent meaning: each piece of text is turned into a list of numbers (a vector) so that texts with similar meaning end up close together. Vector databases store those vectors and find the nearest ones in milliseconds — the foundation of all modern AI retrieval. Here is what they are, how they work, and why embedding proximity decides whether your content gets retrieved.
Read moreWhat is reranking? How AI picks which sources actually get used
Reranking is the second stage in a RAG pipeline: after the initial search has pulled in a batch of candidate sources, a dedicated model gives them a more precise relevance score and re-orders them before the answer is generated. This is where the genuinely relevant chunks are separated from the merely apparently relevant ones. Here is how reranking works, and why being retrieved is not enough — you have to survive reranking.
Read moreWhat is a foundation model? How large AI models learn and remember
A foundation model is a large, pretrained AI model — like GPT, Gemini, Claude or Llama — trained on vast amounts of text and built to be adapted to many different tasks. It 'remembers' knowledge in its weights (parametric memory), but what it did not learn during training has to be retrieved. Here is what foundation models are, the difference between pretraining and fine-tuning, and why brand facts have to be either in the training data or retrievable.
Read moreWhat is fine-tuning? And why AI visibility doesn't come from it
Fine-tuning is training a ready-made language model further on your own examples so it learns a specific style, format or task. It is often confused with RAG and prompting — and even more often mistaken for the path to AI visibility. Here is the difference, and why what you get cited for almost always comes from what the model retrieves, not what it was trained on.
Read moreWhat is Deep Research? The AI that does the research for you
Deep Research is the agentic research mode in ChatGPT, Gemini and Perplexity: instead of one quick answer, the AI writes a research plan, searches widely, reads dozens of sources and produces a long, cited report. Here is how it works, and why it makes being citation-worthy more important than ever.
Read moreWhat is an AI browser? The agents that browse for you
An AI browser is an agentic browser with an AI assistant built in — one that does not just display pages, but navigates, reads and acts on them for you. OpenAI Atlas, Perplexity Comet and CitationLab's own Cite Browser lead the way. Here is what they are, and what it means when an agent, not a human, visits your website.
Read moreWhat is function calling (tool calling)? How AI acts on structured data
Function calling — or tool calling — is how a language model invokes external tools and APIs: it returns structured JSON specifying which function to run and with what arguments. It is the very mechanism behind AI agents, and it is closely tied to MCP. Here is what it is, and why it matters for how your data becomes available to AI.
Read moreHold deg oppdatert
Få fagartikler, produktnyheter og analyser rett i innboksen.