Home / Prompt engineering basics
Fig. 02 — Foundations

Prompt engineering basics

A prompt is not a random question: it is a specification. The Role · Task · Context · Format structure remains the most reliable starting point, whatever the model.

01 — Role: who answers?

A realistic stance that serves the task: “financial analyst”, “technical reviewer”. A theatrical role only adds noise.

02 — Task: what, exactly?

One clear action verb, one main task per prompt. Vague goals cause most off-topic answers.

03 — Context: with which data?

Provide the necessary information, and only that. Delimit it (tags, quotes) so the model knows where data starts and ends.

04 — Format: in what shape?

Table, JSON, bullets, maximum length: the most underused and most profitable lever.

2026 note: recent models infer intent better, which reduces the examples you need — not the need to structure your request.

What is a prompt?

A prompt bundles the instructions, data and examples sent to a model. The field now extends to context engineering: deciding which information to load into the context window, and in what form.

Before / after

BEFORE: Summarise this text.

AFTER: You are a science editor. Summarise the text below in 5 bullets for a non-expert audience, no jargon. If a passage is ambiguous, flag it instead of interpreting it.