What Is an AI Prompt? Definition and How It Works

The word "prompt" has become common, but it stays fuzzy for a lot of people. It is also the single most useful concept to master to get anything out of generative AI. Here is what a prompt actually is, what separates a good prompt from a bad one, and how to practise, particularly in a teaching context.

What is a prompt, exactly?

A prompt is the instruction, written or spoken, that you give a generative AI to get a result. It is the text you type into the input box: a question, an instruction, a request to draft or edit something. In short, it is how you talk to the tool.

That instruction plays a central role. The model does not guess your intent, it reacts to what you give it. The same AI can produce a mediocre or an excellent answer depending on how the request is phrased. The prompt is therefore the interface between what is in your head and what the machine produces.

How does a good prompt work?

An effective prompt is not longer, it is more precise. You can break it down into useful ingredients, dosed according to the task.

Role and context

Telling the AI which point of view to answer from, and in what setting, immediately shapes the output. "You are a hospitality management trainer speaking to first-year students" frames the answer far better than a neutral request. Context cuts down on generic answers.

Task and format

You need to say what you want, and in what form. "Write a five-point outline", "give three examples", "answer in a table": specifying the format saves you from having to redo the work later. A request with no expected format often produces an answer in an unpredictable one.

The example, when it helps

Showing an example of what you want is often better than describing it. Pasting in a sample email to imitate, or a model exercise to reproduce, aligns the output with your expectation. It is a simple, underused lever.

Iteration

The first result is rarely the right one, and that is not a problem. You read the answer, adjust the instruction, run it again. "Too long, cut it in half", "the tone is too formal". This correction loop is part of the work, not a failure of the original prompt. These principles match the ones set out in Anthropic's prompt engineering guide.

Practical examples for teaching and training

For a teacher or a trainer, knowing how to write a prompt changes day-to-day work.

• A weak prompt: "Write me a course on marketing." The result is vague, with no level, no duration, no audience specified.
• A strong prompt: "You are a trainer. Propose the outline of a two-hour session on the basics of digital marketing, for students on a vocational business course with no prior knowledge. Five parts, each with a learning objective and an exercise."

The difference is not the subject, it is the framing. The same habits work for generating a quiz calibrated to a level, rephrasing an explanation for a struggling student, or producing several variants of the same exercise. This is exactly the kind of skill you build when you train your team on AI: learning to ask, then to check.

Common mistakes to avoid

The first mistake is staying too vague and expecting the AI to guess. A short, fuzzy prompt gets a generic answer. Adding a role, an audience and a format is often enough to transform the result.

The second is expecting everything on the first try. An effective prompt gets built through iteration. Wanting the perfect result without ever adjusting leads to frustration.

The third is taking the answer at face value. A well-crafted prompt improves quality, it does not guarantee accuracy. AI remains a system that produces plausible output, as our article on generative AI explains. Human checking stays essential, especially for facts, figures and sources.

The link to real-world business use

In business, prompt skill is the first real step in AI adoption. Before talking about sophisticated tools, a team that knows how to phrase its requests is already saving time on drafting, summarising and preparing documents.

It is a transferable, durable skill. Models change, interfaces evolve, but the habit of framing an intention clearly stays useful everywhere. A team that shares the same framing habits produces results that are more consistent, easier to review and easier to reuse from one person to the next.

The best starting point, then, is not a tool but a method: a handful of prompt templates, matched to your recurring tasks, that anyone can pick up and adjust. Establishing a shared vocabulary first, starting with the definition of generative AI, helps an organisation move from scattered individual use to a shared, reliable practice, then build a genuinely tailored upskilling plan.