Prompting, properly: a field guide to getting what you asked for
You do not need tricks or magic words. You need context, a clear goal and a way to check the answer. Six habits that work with every model.
Most disappointing AI answers have the same cause: the model had to guess. It did not know who the answer was for, what good looks like, or what you already know. Better prompting is mostly about removing guesswork.
1. Say who it is for
"Explain inflation" and "explain inflation to a 12-year-old who loves football" produce very different answers. Naming the audience sets the vocabulary, the depth and the examples in one go.
2. Give it the material
Paste the email, attach the report, share the numbers. A model working from your actual material will outperform one working from memory every time, and it is far less likely to make things up.
3. Describe the finished result
Length, format and tone are easy to state and easy to forget. "Three bullet points, plain language, no jargon" takes five seconds to write and saves two rounds of revision.
4. Show an example
If you have a piece of writing you like, include it and say what to keep. Examples communicate style faster than any adjective.
5. Ask for the reasoning when it matters
For decisions, ask the model to lay out the options and the trade-offs before it recommends one. You can then check the logic, not just the conclusion.
6. Treat the first answer as a draft
The best results come from conversation. Point at what is wrong, say what to keep, and iterate. Two short follow-ups usually beat one enormous prompt.
A good prompt reads like a good brief to a capable colleague.
None of this is specific to one assistant. Claude, ChatGPT, Gemini and the rest all respond to the same thing: clear context and a clear goal.
