# Prompt engineering

> The practice of writing model inputs so the output is reliably what you need — structure, examples, constraints and tone rather than hope.

A prompt is the entire input the model conditions on: the system prompt that sets role and rules, the context pulled in for this request, and the user's words. Engineering it means being explicit about format, showing examples of the desired output, and stating what must not happen — 'answer from the attached price list, and say you do not know when it is not there'.

Its reputation as magic wording is wrong: the durable gains come from structure, not incantations, and a prompt that matters gets tested like code — versioned, eval'd against cases, rolled back when a change regresses.

## Related terms

- https://dfieldsolutions.com/en/glossary/system-prompt.md
- https://dfieldsolutions.com/en/glossary/llm-evaluation.md
- https://dfieldsolutions.com/en/glossary/chain-of-thought.md

---

Source: https://dfieldsolutions.com/en/glossary/prompt-engineering
DField Solutions — Dunakeszi, Hungary — dezso@dfieldsolutions.com
Booking: see https://dfieldsolutions.com/en/contact