The text given to a large language model to condition its response: the system instruction, context, retrieved documents, and user message the model reads before it generates.
Prompt is the text given to a large language model to condition its response: the instruction, question, or material the model reads before it generates anything. It spans more than what a user types. A full prompt usually bundles a system instruction, the conversation so far, any retrieved documents, and the user's message, all of it broken into tokens the model processes together within its context window.
Prompts differ from search queries in form and range. A query is a few keywords; a prompt is often a full sentence or paragraph, phrased countless ways for the same underlying intent. That variety is why we anchor optimisation to an entity and study the query fan-out of prompts that stem from it, rather than chasing individual wordings.
Because the whole prompt conditions the output, what sits alongside the user's words matters as much as the words themselves. Deliberately shaping that surrounding context is context engineering, and probing how a model responds across many controlled prompts is how we measure a brand's AI visibility.
