Quick answer
Good prompting isn't a magic phrase collection — it's clear task delegation. The pattern that covers most work: give the AI a role, the relevant context, the exact output you want, and any constraints. Then iterate once or twice instead of rewriting from scratch. Tools like ChatGPT, Claude, and Gemini all respond to the same fundamentals; no engineering background required.
Pattern 1: Role + Context + Task + Format
The workhorse structure for nearly any request:
- Role: "You are an experienced editor" — sets vocabulary and depth.
- Context: who the audience is, what happened before, what you already tried.
- Task: one specific action verb — draft, summarize, compare, rewrite.
- Format: length, structure, tone ("300 words, three bullets, plain language").
Weak: "Write something about our product launch."
Strong: "You are a B2B marketing writer. Our company launched a scheduling tool for clinics (context below). Draft a 150-word announcement email to existing customers. Warm, no hype words, end with one clear call to action."
Pattern 2: Give it your material
Paste real inputs — your draft, the meeting transcript, last quarter's report — and ask questions about it. AI responses grounded in your actual material beat anything generated from thin air. This is how non-engineers get analyst-grade help: "Here is our Q3 support ticket log. What are the five most common complaints, in order?"
Pattern 3: Show an example
If style matters, include one sample of writing you like: "Match this voice: [paste your best paragraph]." One good example outperforms paragraphs of adjectives like "professional but friendly."
Pattern 4: Ask for options, not verdicts
"Give me three subject lines with different angles" beats "what's the best subject line?" Options keep you the decision-maker and reveal the model's range cheaply.
Pattern 5: Iterate; don't restart
Treat the first response as a first draft of the conversation. Useful follow-ups:
- "Tighten this by half without losing the specifics."
- "You assumed X — actually, here's the correction; redo section two."
- "What would a skeptical CFO object to in this? Then fix those objections."
If a response is badly off after two rounds, the prompt is usually missing context rather than phrasing. Add information instead of rewording.
Pattern 6: Make it show its sources and doubts
For factual work: "List which claims you're confident about and which you'd need to verify." Models are more useful when invited to flag uncertainty — though you should still verify important facts independently regardless.
Common mistakes
- Vague verbs: "help me with marketing." Help how? Name the deliverable.
- One giant ask: splitting a big job into sequential small requests produces far better results than one mega-prompt.
- Accepting fluent nonsense: confident tone is not evidence. Verify numbers, names, quotes, and citations before they reach anyone else.
- Pasting secrets: don't put customer data, credentials, or confidential material into consumer tools without checking your organization's policy and the tool's data-use terms.
- Prompt hoarding: a saved prompt that no longer fits your task costs more time than writing a fresh one.
A five-minute daily practice
Pick one recurring task this week — status updates, email triage, meeting prep — and write one solid prompt for it using Role + Context + Task + Format. Reuse it daily, refining once per day. After a week you'll have a personal prompt library built on your real work, which beats any downloaded list of "100 magic prompts."
Frequently asked questions
Do different AI tools need different prompting?
The fundamentals transfer across ChatGPT, Claude, Microsoft Copilot, and others. Minor differences exist in default tone and instruction-following style, but clear roles, context, and format specs work everywhere.
Are long prompts better?
Longer only when it adds context the model lacks. Padding ("I want you to act as a world-class expert with 20 years of experience…") rarely helps; specific facts about your situation always do.
How do I keep my data private?
Check each tool's data-use settings (many offer training opt-outs), avoid pasting identifiable client data into consumer tiers, and follow your employer's policy. For sensitive material, look at business plans with stronger data controls — verify current terms on official pages.