# Prompting for journalists

> Short, practical guide to writing instructions models can follow — with a longer warning about over-trusting outputs.

**Format:** Guide · **Length:** 16 pp · **Languages:** EN
**Licence:** CC BY-SA 4.0
**Published by:** Good AI for Journalism — a programme of Icebauhaus, Labtek Indie, and BandungBergerak.

---

Prompting is a small skill with a big surface area. This guide focuses on the subset of that surface that actually matters in editorial work: how to instruct a model to summarise, extract, translate, or rewrite — and how to tell when the result is not trustworthy.

The core advice is short. State the role, the input, the task, and the constraints. Ask for structured output when you need to verify quickly. Never let a model speak in the first person of a real source. Treat every named quote as suspect until you have checked the recording.

The longer warning is also short, and worth repeating: a fluent answer is not a correct answer. The model has no idea whether it is wrong.


---

This document is released under CC BY-SA 4.0. If you adapt or republish it, please credit the Good AI for Journalism programme and link back to the source.