What is prompt engineering?

In plain English

Prompt engineering is the craft of getting better answers out of an AI model by changing what you ask and how you ask it. In practice it means adding context, stating the audience and format, showing examples of a good answer, breaking a big task into steps, and testing variations to see which wording works most reliably.

Despite the word “engineering”, it mostly involves ordinary writing skills: being specific, anticipating misunderstandings and saying what “good” looks like. The engineering part comes in when prompts are reused at scale — inside a company’s software, for instance — and need to be tested like any other component.

Why it matters

Many complaints about AI tools being vague or wrong come down to thin instructions. Better prompts can cut errors, enforce a house style and save rewriting time, without any change to the underlying model. For teams, a shared library of tested prompts makes results consistent across colleagues rather than dependent on individual knack.

An example

A recruiter who asks for “a job ad for an accountant” gets filler. A prompt that supplies the salary band, the team size, three must-have qualifications, the company’s tone (“warm, no buzzwords”) and a past advert as a model gets a draft that needs only light editing.

What to watch out for

Prompting cannot fix everything. If a model lacks the information — a private document, a recent event, a specialist standard — no clever wording will conjure it, and the model may invent something instead. Prompts that work well with one model may work poorly with another, so techniques need re-testing when tools change. Beware of “magic prompt” lists sold as secrets; clear, specific instructions beat incantations.