AI

Prompt

What it means

A prompt is the instruction, question or information you give an artificial intelligence system to tell it what you want it to do.

If you have ever typed a question into ChatGPT, you have already written a prompt. The prompt is the starting point of the conversation — the message that sets the task. Everything the AI produces is shaped by what you asked and how you asked it, which is why the same model can give a vague, unhelpful answer to one request and a precise, useful answer to a slightly different one.

Prompts range from the very simple to the quite detailed. A simple prompt might be a single line — 'summarise this article' — while a more useful one adds context and constraints: 'summarise this article for a busy manager who needs the three key decisions and any deadlines, in under 150 words, in a neutral tone.' The second version gives the model enough to aim at, and the result is usually closer to what you actually wanted.

What makes a prompt useful is not a secret formula but a few clear ingredients: enough context for the model to know the situation, a clear statement of the desired output, any constraints (length, tone, format, what to include or avoid), and, often, the role you want it to take ('explain this as if to a curious beginner'). Follow-up prompts matter too — you can refine a result by asking the model to shorten, rephrase, add a section or correct something, and a short back-and-forth frequently beats one perfect opening request.

It is worth resisting the idea that there is one 'perfect' prompt. In practice, good prompting is iterative: you try, see what comes back, and adjust. Clear, specific, well-contextualised requests tend to work better than vague ones, but the best prompt for a task is usually discovered by working with the model rather than copied from a list.

One quiet but important habit: because an LLM's AI hallucination risk rises when you ask it for specifics it cannot reliably know, a good prompt often includes an instruction to say when it is unsure — for example, 'if you're not certain, tell me rather than guessing.' The prompt is not just a task; it is also where you set the expectations that keep the result trustworthy.

Why it matters

Prompts are the everyday lever for getting value from generative AI. A small change in how you ask — adding context, a desired format, a tone — can turn a mediocre answer into a genuinely useful one. Learning to write a clear prompt is less about memorising formulas and more about describing what you actually want, which is a skill most people already have. The better you communicate the task, the more the tool can do for you. For a plain-English walkthrough, see How to Write Better ChatGPT Prompts Without Complicated Formulas.

A practical example

Compare two prompts. Unclear: 'write about email.' This could mean almost anything, and the model will guess. More useful: 'write a friendly 120-word email to a colleague acknowledging a delayed project, explaining the reason briefly, and proposing a new deadline of Friday.' The second leaves little to chance — the model knows the audience, length, tone, content and the specific outcome you want. The difference in the result is usually dramatic, and it comes entirely from how the request was phrased.

Related terms

Go Deeper

This term is explored fully in Get Better Answers From ChatGPT.