How to write a ChatGPT prompt that doesn't waste your time

A great ChatGPT prompt has three parts: instruction + context + format. Skip any one and the output gets generic fast. That's the whole formula. Most people miss two of the three and then conclude AI is "just OK." It isn't — your prompts are.

This is a practical guide. Examples, not theory. By the end you'll know exactly why your prompts have been underperforming and what to do about it.

The 3-part formula

Instruction — what you want done. Be specific. "Write an email" is too vague. "Write a 3-paragraph follow-up email" tells ChatGPT what shape the answer should take.

Context — who you are, who it's for, what came before. ChatGPT knows nothing about your situation by default. Spell it out. "I'm a sales rep emailing a CFO who didn't reply to my proposal last week" gives ChatGPT something to work with.

Format — how you want it back. Bullet list, table, single paragraph, Slack message, JSON. Tell it. Otherwise you get whatever it feels like, which is usually a wall of prose.

Before / after

Watch what happens when you add the missing pieces.

❌ Bad: "Write me an email about scheduling a meeting."

You'll get a generic, formal-sounding email with too many adjectives and a "I hope this email finds you well" opener.

✅ Better: "Write a 60-word follow-up email to a CFO I met at a conference last month who said she'd 'circle back next quarter.' I want to nudge for a 15-minute call this week. Tone: warm but direct. No 'hope this finds you well.' End with a yes/no question."

You'll get something you can actually paste and send.

The difference is 60 seconds of typing context. The output difference is 10x.

The role frame: the cheapest 1-line upgrade

Adding "You are a [role]" at the start of a prompt anchors ChatGPT in a specific voice and depth. It's the single biggest cheap upgrade most people aren't using.

• "Help me with my pitch deck" → generic startup advice.
• "You are a Series A pitch coach who's seen 500 decks. Critique mine." → actual pointed feedback.

• "Explain compound interest" → textbook definition.
• "You are a finance teacher for 8th graders. Explain compound interest using a sneaker example." → a kid-comprehensible analogy you can actually use.

Common mistakes to stop making

Too vague

"Climate change." "Help with my essay." "Make this better." These aren't prompts; they're topics. ChatGPT writes a Wikipedia stub and you walk away thinking AI is overrated.

Too long

Walls of text without paragraph breaks. ChatGPT misses your actual ask. Keep prompts scannable. If your prompt is over 200 words, split it: set context first, then ask the question.

Contradictory

"Make it formal but casual." "Brief but thorough." "Professional but funny." Pick one or specify the tradeoff.

No context

You know your situation. ChatGPT doesn't. Describe it. Two sentences are usually enough.

Asking for opinions like facts

"What is the best CRM?" forces ChatGPT into a confident-sounding guess. Better: "What are the tradeoffs between Salesforce, HubSpot, and Pipedrive for a 5-person B2B startup?" Now you get something you can actually act on.

Iterate — don't restart

First output is rarely the final output. The skill isn't writing the perfect first prompt; it's the follow-up. Common iteration moves:

• "Cut to half the length." Tightens prose dramatically.
• "More specific examples in section 2." Fills in vague spots.
• "Less corporate. Sound like a human." Fixes tone.
• "Where am I being weak here?" Switches into critique mode on your own work.
• "Try a totally different angle." Restart within the same context, no need to start a new chat.
• "What did you assume that I didn't tell you? Surface those assumptions." Catches misinterpretations.

Three rounds of "tighter" + "more specific" + "different angle" almost always beats one perfect prompt. You're directing, not authoring.

The chain-of-thought prompt

For anything that requires reasoning — analysis, decisions, multi-step math, complex code — add "Let's think step by step" or "Reason carefully before answering."

It costs nothing. It dramatically improves quality on hard questions. ChatGPT will show its work, which means you can spot where it went sideways and correct course. Without it, you get a confident answer with no audit trail.

Few-shot examples (when you want a specific style)

If you want output in your voice, give it 2–3 examples of your past output. "Here are three subject lines I've written that I like: [paste 3]. Now write 5 more in the same voice for [topic]."

Wildly more reliable than describing your style in words. Show, don't tell — works for AI exactly like it does for human collaborators.

The pattern that compounds

Once you find a prompt that works for a recurring task, save it. Reuse it. Tweak it. Build a personal library. The 5–10 prompts you use weekly will be worth more than every "ultimate prompt list" on the internet, because they're tuned to your context.

If you use ChatGPT daily, set up Custom Instructions (in Settings → Personalization). One-time setup: tell it your role, your tone preferences, your formatting defaults. Every chat going forward inherits that. Massive ongoing time saver.

The 60-second test

Before you send any prompt, ask yourself: does it have all three? Instruction, context, format? If no, add the missing piece. The 30 seconds spent will save you 5 minutes of regenerating bad output.

Climer's Unit 2 walks through this with hands-on examples — the formula, the iteration loop, the role frame, advanced patterns like chain-of-thought and few-shot. 10 climbs, 5–15 minutes each, free during early access.

Related: ChatGPT vs Claude vs Gemini covers when to use each tool — the prompts above work for all three.