What is prompt engineering?

Prompt engineering is the more deliberate side of writing prompts. It means working out what instructions, context, and examples an AI model needs to do a task well, then testing and refining them until the output is reliable. For a one-off question this is rarely needed, but it matters when the same prompt will be reused many times, shared with a team, or built into an automation.

For example, a support team that uses AI to draft replies might write one carefully designed prompt that sets the tone, lists the company's refund rules, tells the model to ask for an order number when one is missing, and shows two examples of good replies. The team then tests it against a batch of real past emails and adjusts the wording wherever the drafts go wrong.

Common techniques include giving the model a role or an audience, breaking a task into clear steps, supplying examples of good output, specifying the format, and telling the model what to do when it is unsure, such as saying that it does not know. Testing across varied examples, including awkward ones, matters more than finding clever wording.

Prompt engineering improves results but cannot make a model more capable than it is. A prompt that works well with one model may behave differently with another, or after the same model is updated, so important prompts are worth retesting from time to time. It also cannot fix missing information; if the model needs facts it does not have, they have to be supplied.

An example

An HR team refines a prompt for summarizing job applications until it reliably lists skills, experience, and missing information in the same format.