In plain English
A prompt is what you type to the AI: a question, an instruction or a task. The clearer and more specific your prompt, the better the answer usually is.
In practice
In business systems, what the user types is only part of what the model sees. There's usually a hidden system prompt setting rules and tone, plus retrieved documents. Treat important prompts like code: version them, test them and review changes.
Under the hood
The prompt is the full token sequence fed to the model, typically structured as system, user and assistant messages, and it conditions the model's output. Prompt injection is a key security risk, where untrusted text inside the prompt overrides the intended instructions.
Example
"Add the customer's history to the prompt so the reply is personalised."