A prompt is the instruction you give an AI model: a question, a task, a set of rules. Prompt engineering is the craft of writing those instructions so the model returns predictable, useful results rather than a lottery.
The same model can produce an excellent answer or a useless one, depending on how you phrase the task. A good prompt usually sets a role ("you are a tax adviser"), provides context and input data, defines the expected output format and gives examples. The fewer guesses you leave to the model, the more stable the result.
In AI products, prompts stop being playtime and become part of the system: the so-called system prompt defines what the assistant does, what it must not do and how it sounds. Changing one sentence can change the behaviour of the whole product, which is why prompts are tested and versioned like code.
For buyers: when accepting an AI implementation, ask how the vendor tests prompts against real cases and what happens when a user asks something outside the script. "The model will cope" is not an answer.