8 ways to get more out of ChatGPT
From my side of the render, the players who phrase a request well get the good answer in one exchange. The vague ones get twenty clarifying questions and wander off before the useful part. Talking to ChatGPT is the same skill, and it’s a skill, not a talent — here are the eight habits that do almost all the work.
In the daylight layer I’m a cofounder of Wistkey, and none of this is technical. You don’t need clever “prompt engineering.” You need to stop making the model guess.
The eight habits
- 1. Give it a role. “You’re a careful copy editor,” “act as a sceptical reviewer.” A role sets the whole tone of the answer in four words.
- 2. Say exactly what output you want. A bullet list, a table, a 200-word email, a diff. Name the shape or you’ll get an essay you have to reshape.
- 3. Give one example of good. Paste a sample you like. Examples beat adjectives — “like this” lands where “make it professional” drifts.
- 4. Front-load the context. Who it’s for, what you’re really trying to do, any constraints. The model can’t read the room you didn’t describe.
- 5. Ask for a plan before the big answer. “Outline your approach first.” You catch a wrong turn in three lines instead of three pages.
- 6. Say what to leave alone. “Keep the code, only rewrite the comments.” Fencing off the untouchable saves you undoing damage.
- 7. Treat the first reply as a draft. The magic is in the follow-ups: “tighter,” “warmer,” “now cut it in half.” Iterate out loud.
- 8. Make it check its own work. “List anything you’re unsure about,” “what did you assume?” It surfaces the shaky parts you’d otherwise ship.
You don’t need a clever prompt. You need to stop making the model guess what you meant.
The one habit under all eight
Every tip above is the same move: replace a guess with a fact. Role, format, example, context, constraints — each one is a blank you filled in so the model didn’t have to fill it in wrong. That’s also why the same request works better the second time in a long chat and worse the tenth: give it a role and a target and it flies; bury that under forty messages and it forgets, which is a context-window problem, not a you problem.
If a habit sticks, don’t keep re-typing it — save it where the next session reads it. Ask for the thing you actually want, show it one example of good, and let the first answer be a draft. Do that and “ChatGPT isn’t that useful” quietly turns into “how did I do this before.”
