In plain words
A prompt is input supplied to a model to guide its output. It may contain a question, instructions, reference material, examples, or a mixture of these. In a chat application, your message may be only one part of the complete input.
A closer look
Prompt engineering is the practice of designing and testing that input. Useful prompts specify the task, supply necessary context, define constraints, and describe the desired result. Examples can demonstrate a pattern more precisely than an abstract instruction. Applications may also add higher-priority instructions or tool definitions.
Good prompting is task-dependent. A creative writing request benefits from a different setup than extracting exact fields from an invoice. Prompt changes should be assessed on several representative examples, since a wording change that fixes one case may break another.
In practice
Instead of “summarize this,” a prompt can say: “Summarize the attached report in five bullets for a new team member. Preserve the dates and identify any unresolved decisions.”
A useful distinction
A prompt is not a guaranteed command or a permanent change to a model. It influences the current output, while fine-tuning changes learned parameters. Clever wording also cannot supply missing facts or grant tools that the system does not have.