The agentic marketing glossary

Coordinating multiple AI agents so their work composes: who runs when, what each is allowed to touch, and how their outputs feed each other.

The cycle an AI agent runs continuously: observe the data, notice what changed, investigate why, act within its permissions, and verify the action worked.

An autonomous system that uses an LLM to plan, execute, and iterate on tasks with minimal human intervention.

The discipline of structuring content so answer engines, from featured snippets to AI assistants, can extract and attribute a correct answer to you.

Numerical vector representations of text that capture semantic meaning, used for search, clustering, and recommendations.

Providing a small number of examples in a prompt to guide an LLM's output format and behavior.

Training a pre-existing model on a specific dataset to improve its performance on a narrow task or domain.

Optimising content to be read, cited and recommended by AI answer engines such as ChatGPT, Perplexity and Google's AI Overviews, rather than only ranked by classic search.

When an AI model generates information that sounds plausible but is factually incorrect or fabricated.

Large Language Model. A neural network trained on massive text data that can generate, summarize, and reason about language.

An open protocol that standardises how AI models connect to external tools and data sources, so any compliant agent can use any compliant tool.

The practice of crafting inputs to an LLM to reliably produce desired outputs, including system prompts and few-shot examples.

Retrieval-Augmented Generation. A technique that feeds relevant external data into an LLM at query time to ground its responses in facts.

The basic unit of text processed by an LLM, roughly equivalent to 3/4 of a word. Models have token limits for input and output.

The mechanism by which an AI model invokes external functions, APIs or systems: the difference between an agent that can only talk and one that can act.

Asking an LLM to perform a task with no examples, relying entirely on the model's pre-trained knowledge and instructions.