Practice
Prompting for journalists
A short guide to writing instructions the model can follow — and a longer warning about over-trusting the result.
A useful prompt names four things: the role the model should take, the input it is working from, the task it is performing, and the constraints on the output. Skip any one of them and the model fills in a default you did not choose.
Ask for structured output — JSON, a table, a numbered list — whenever you intend to verify the result. Free-flowing prose is harder to spot-check than a list of named claims.
The longer warning is the one to print on the wall: a fluent answer is not a correct answer. The model has no internal signal for "I don't know." Treat every named entity, every quoted phrase, every numeric claim as unverified until you have checked it.
Related articles