These are the prompting lines worth keeping by your side. Paste one into a chat, swap the [brackets] for your own words, and that's the whole move. None of them are tied to a single model; they work on any modern reasoning model, with Claude as my reference point.
1. Give it the "why" (context over detail)
One line of "here's why I'm asking" beats piling on step-by-step instructions. The context lets the model connect your task to the right information instead of guessing what you meant.
Weak: "Write me an email to a client about the delay."
Paste this:
I'm working on [the bigger task] for [who it's for]. They need [what the output enables]. With that in mind: [your request].
2. Tell it what NOT to do (negative prompting)
A more capable model acts on its own more readily: it will draft, edit, or refactor around a one-line fix without being asked. The sentence that saves you starts with "don't."
Weak: "Take a look at this problem and handle it."
Paste this:
When I'm describing a problem or asking a question, the deliverable is your assessment. Report what you find and stop. Don't fix, send, edit, or delete anything until I say go. Do the simplest thing that works, and skip cleanup I didn't ask for.
3. Let it act once it has enough (don't over-plan)
More deliberation isn't better. On a model that can run for minutes, endless option-surveying just burns time on choices it will never use.
Weak: "Research everything and make a full plan before you do anything."
Paste this:
When you have enough information to act, act. Don't re-derive what we've already settled or narrate options you won't pursue. If you're weighing a choice, give a recommendation, not an exhaustive survey.
4. Make it prove it
"Done and working" is a claim, not a fact. One line asking for the receipt keeps the model honest about what it actually finished.
Weak: "Is it done and working? Great, thanks."
Paste this:
Before you tell me something is done, point to the result that proves it. Only report work you can show evidence for. If something isn't verified, say so plainly instead of guessing.
5. Say less, not more
A short instruction steers as well as a long rulebook, and reusing last year's over-detailed prompt can make the answer worse. No contradiction with habit 1: add the why, which is real signal the model can't guess, and cut the rules, which are procedure it already follows.
Weak: "Rule 1: be concise. Rule 2: use bullets. Rule 3: no jargon. Rule 4:…"
Paste this:
Lead with the outcome, keep it simple, and pause only when the work truly needs me.
Two settings, one dial
Habits 2 and 3 pull in opposite directions on purpose. Habit 2 holds the model back; habit 3 pushes it forward. Treat them as one dial rather than two contradictory rules, match the setting to the moment, and never paste both into the same prompt.
- Too eager (it wants to fix, send, or refactor before you asked): reach for habit 2 so it assesses, then stops.
- Stalling (it keeps re-checking and surveying options it won't use): reach for habit 3 so it acts once it has enough.
Two honest caveats
"Prove it" is an honesty habit, not a guarantee. In a plain chat with no tools, treat it as a good habit rather than proof. On high-stakes work, verify it yourself, or open a fresh chat and have it check the result against your original ask.
Keep a human yes on anything irreversible. These lines cut the babysitting on reversible, internal work. Keep your own approval on anything external or permanent: client emails, spending, publishing, deletions, legal, financial.