A prompt is the instruction you give an AI – the text that tells the model what to deliver, for whom and how.

The word means a cue or a request. In practice, a prompt is a task description. It can be a single line or several pages with background, examples and requirements for the format.

How a prompt works

A language model does not know your strategy, your customers or the meeting the task came from. It only has what is in the prompt and the patterns it has learned from its training data. Anything missing from the prompt, the model fills in with what is most likely. That is why the answer to a vague prompt is average.

A good prompt resembles good delegation. It describes the purpose, the situation, the sources to be used, the form the result should take and the tone it should be written in. Those are the five elements of the FOCUS model.

Why the prompt goes beyond technique

When AI becomes part of the work, part of the work shifts from doing the task to describing it. That applies in every role. The more precisely purpose, audience and quality requirements are described, the more useful the result.

That has three consequences. First, a task can only be described precisely when the goal is clear. Unclear goals in the organisation turn into unclear prompts. Second, it has to be decided which tasks AI may handle and where a person must assess the result. Third, the ability to describe tasks has to be trained. Otherwise the licences do not turn into better work.

The common misunderstanding

Many organisations treat prompting as a technical skill that can be solved with a course in phrasing or a prompt library. The phrasing is the smallest part. The hard part is knowing what you want from the task. Anyone who knows that can write a good prompt with the FOCUS model in a few minutes.

In practice

A CEO asks an AI for 'a board paper on our AI strategy'. The result is generic text that could be about any organisation. Next time she writes: 'Write a one-page draft for the board. The purpose is to get three pilot projects approved. The board is concerned about data security. Use the attached memo as the only source. Write in a calm and factual tone.' Now she gets a draft she can work with.

What has changed is that the CEO has worked out what the paper should achieve and who it is for.

Useful moves

  • A good first question, before more tools are bought, is whether the most important tasks can be described precisely enough for a colleague to carry them out.
  • The ability to describe tasks is best trained on real work, for example with the FOCUS model, rather than with ready-made prompts.
  • Experience with AI in your own tasks makes it easier to judge what you can hand over to AI, and what you are asking others to do.
  • It has to be decided which tasks AI may help with and where a person must assess the result. The AI role atlas is a tool for that.

Related terms: Prompt library, System prompt, Context window.

The concepts in this glossary run through my AI courses and workshops, where we work with them on your own tasks.