ChatGPT Cheat Sheet: Prompts, Tricks and Shortcuts

Keep this page open next to ChatGPT. A ChatGPT cheat sheet isn't theory — it's the handful of patterns that reliably turn a vague reply into the one you keep. Below you'll find the prompt formula, the phrases that actually steer output, formatting tricks, quick iteration moves, and copy-ready templates for the tasks you do every week. Skim it, grab what you need, drop your details where the [BRACKETS] are. No fluff, just the parts that work.

Bookmark it. The point of a cheat sheet is that you stop memorising and start reaching — glance, copy, adapt, send.

The prompt formula

Almost every strong prompt fills four boxes. Miss one and the reply drifts.

• Role — who ChatGPT acts as. "You are a hiring manager" pulls in the right judgement.
• Task — one clear verb: write, compare, plan, debug. One sharp ask beats three fuzzy ones.
• Context — the background only you hold: audience, constraints, the source text.
• Format — the shape you want back: five bullets, a table, a 120-word email, valid JSON.

That's the whole ChatGPT prompt formula: role + task + context + format. Here it is assembled.

When a reply misses, check which box was empty — it's usually context or format. You rarely need all four to be long; you need all four to be present. A one-line role and a one-line format often do more than a paragraph of backstory. Want the model to build the prompt for you? The ChatGPT prompt generator assembles one from a plain description, and the prompt optimizer tightens a draft you already have. The rest of this ChatGPT cheat sheet is variations on those four boxes.

Power phrases that steer output

Short phrases do most of the steering. Keep these on hand.

• "Act as [role]" — sets vocabulary and standards in three words.
• "Think step by step" — forces reasoning before the answer, fewer sloppy jumps.
• "In the style of [X]" — borrows a tone or structure without a paragraph of description.
• "Ask me questions first" — stops the model guessing when it's missing key facts.
• "Return only [X]" — kills the preamble and the "Sure, here's..." wrapper.
• "No preamble, no summary" — same idea, blunter.
• "If you're unsure, say so" — cuts confident nonsense.

Front-loading questions beats a wrong first draft every time. These are the same ChatGPT tricks the prompt analyzer flags when a request is too vague to answer well.

Stack two or three phrases when you need to. "Act as a skeptical editor, think step by step, and return only the three weakest lines" does real work in one sentence.

Formatting tricks

Tell ChatGPT the container and it fills it cleanly. Ask after the fact and you're reformatting by hand.

• Bullets: "Answer as 5 bullets, one line each."
• Table: "Compare these in a table: columns [A], [B], [C]."
• JSON: "Return valid JSON only, keys: title, summary, tags."
• Word limit: "Under 80 words." Numbers work; "short" doesn't.
• Tone: "Warm but direct." / "Plain English, no jargon."

For structured output you reuse, the JSON prompt tool locks the shape so the model can't wander. One rule holds across all of these: state the format before the content, not after. The model commits to a container in its first tokens, so telling it the shape up front is cheaper than reformatting a finished wall of prose.

Iteration phrases

The fastest fix is your second message, not a rewritten prompt. Short corrections land better than starting over.

• "Shorter — cut it in half."
• "More concrete — add a real example."
• "Less formal, drop the hedging."
• "Keep 2 and 4, redo the rest."
• "Same idea, three different angles."
• "Now do it in a table."

Naming what to keep saves the parts you liked from a full rewrite.

Copy-ready templates

Six workhorses for common tasks. Fill the brackets and go.

Save the ones that work in a prompt library so you're not rebuilding them each time. For more by scenario, see our best ChatGPT prompts and the broader AI prompt examples.

Custom instructions

Anything you repeat in every chat belongs in ChatGPT's Custom Instructions instead. Set your role, tone, and defaults once and every reply follows them.

Put "who I am and what I do" in the first box, "how I want answers" in the second. It's the closest thing to a personal system prompt without retyping context each session. Keep it short — three or four firm rules beat a wishlist the model waters down. Review it every few weeks and cut whatever you stopped needing.

Common mistakes to avoid

• Asking three things at once — the model half-answers all three. Split them.
• Saying "make it better" — better how? Name the axis: shorter, warmer, more specific.
• No format — you get a wall of prose. Ask for the shape up front.
• No role or audience — generic in, generic out.
• Accepting the first draft — it's a starting point, not the finish line.
• Vague length — "short" is a guess; "under 80 words" is an instruction.
• Burying the ask — put the real request in the first line, not after three paragraphs of setup.

Fix the format and the role first; those two account for most weak replies. The rest is iteration, and iteration is fast once the frame is right.

New to the tool entirely? Start with how to use ChatGPT, then come back and bookmark this page. Keep the ChatGPT cheat sheet one tab away, lean on the formula, and you'll spend less time coaxing and more time using what comes back.

References

• OpenAI — Prompt engineering guide
• OpenAI