Prompt Engineering

20 Prompts You Can Copy Into Claude, ChatGPT, or Gemini

Good structure beats the right brand.

These prompt patterns produce high-quality output on any frontier model (Claude, ChatGPT, or Gemini) across writing, coding, analysis, research, and strategy. People search for "Claude prompts," but what actually makes them work is structure, and structure travels between models. Each includes the template and why it works.

Prompt Engineering CoursePractice Free

Try every prompt in the live ChatGPT sandbox, no copy-paste needed

The best prompts follow one pattern regardless of which model you use: define a role, specify the output format, set constraints, and include one or two examples. Claude, ChatGPT, and Gemini all reward this structure. Which is why a prompt tuned for one usually works on the others with little or no change. The skill is portable, not brand-locked.

Writing

Writing prompts

1
Voice-matching editor
Below are three samples of how I write: [paste samples]. Study the rhythm, word choice, and tone, then redraft the text that follows so it reads like I wrote it. No em dashes, no clichés, and no wrap-up paragraph.

Text to redraft: [paste draft]
Why it works: Stops any model from defaulting to generic AI prose. The examples anchor the output to your actual style; this works identically on Claude, ChatGPT, and Gemini.
2
Long-form outline
I need an outline for a [article/report/essay] on [topic], aimed at [audience]. Give me a working headline, five to seven sections, the single point each section should land, and for each one a note on the evidence or example to back it up.
Why it works: A structured output request makes the model organize before it writes, which saves real revision time no matter which assistant you paste it into.
3
Email with a clear ask
Draft a [professional/friendly/direct] email to [recipient] that asks them to [action]. Background: [one or two sentences]. Keep it under 150 words and close with one concrete next step, not a vague open question.
Why it works: The word limit and CTA constraint cut the vague, padded emails every model tends to produce when left unconstrained.
4
Content repurpose
Here's a [blog post / transcript / report]: [paste content]. Turn it into three things while keeping the original tone: a LinkedIn post under 200 words, three one-line takeaways for social, and a short summary paragraph I can drop into an email.
Why it works: One source, three formats. Saves about an hour of reformatting per piece, on whichever model you already have open.

Coding

Coding prompts

5
Code review with criteria
Do a production-grade review of this [language] code. Check for: security holes (injection, auth, exposed data), slow paths, missing error handling, and type problems. For each issue, point to the line and say exactly how to fix it.

[paste code]
Why it works: Explicit criteria prevent vague 'looks good' reviews. Any capable model surfaces real issues once you tell it exactly what to look for.
6
Bug hunt with reproduction
There's a bug in my [framework/language] code. Actual behavior: [symptom]. Expected behavior: [what should happen]. Relevant code: [paste code]. Reason through the likely cause step by step before you propose the fix.
Why it works: Expected-vs-actual framing forces systematic diagnosis instead of guessing, a technique that pays off on every model.
7
Function with tests
Write a [language] function that [what it should do]. Requirements: [list them]. Ship it with input validation, error handling, doc comments, and a [Jest/Pytest/etc] test suite covering the normal path, the edge cases, and the failure cases.
Why it works: Asking for tests alongside the function pushes the model toward testable code, not just code that runs. True across ChatGPT, Claude, and Gemini.
8
Refactor for readability
Refactor this code for readability without changing what it does. Pull repeated logic into named functions, swap magic numbers for named constants, rename unclear variables, and add a comment only where the reason isn't obvious from the code.

[paste code]
Why it works: The 'without changing behavior' constraint is critical: it stops the model from helpfully 'improving' your logic while it tidies the code.

Analysis

Analysis prompts

9
Document summary with key decisions
Summarize this [document type] in three tiers: a one-line takeaway, five key points each with its supporting evidence, and the decisions or actions it demands from the reader. Skip filler openers like 'This document explores.'

[paste document]
Why it works: A three-layer summary forces the model to separate signal from noise and surface the actionable part instead of paraphrasing everything.
10
Competitive analysis
Put [Company A] and [Company B] side by side in a table across pricing, target customer, standout features, positioning, known weak spots, and recent moves. Then call it: which one wins for [my use case], and why.
Why it works: A table plus a required winner forces the model to make a judgment call rather than hedging with 'both have pros and cons.'
11
Data interpretation
Here's a dataset: [paste or describe it]. Give me the single most important trend or anomaly, the missing data that would change how I read it, and which decisions this data can and cannot justify. Be concrete.
Why it works: Asking for what the data doesn't support is the critical question; it prevents overconfident conclusions.
12
Risk analysis
Map the risks of [decision / project / plan] in a table: each risk, its probability (High/Med/Low), its impact (High/Med/Low), and the trigger that would set it off. For every High-impact one, give me a single concrete mitigation.
Why it works: Probability × impact matrix forces prioritization instead of a flat list of every possible concern.

Research

Research prompts

13
Explain a complex topic
Explain [topic] to me as [a smart 12-year-old / a non-technical business person / an expert in a nearby field]. Lean on analogies, define any jargon you use, and finish with the three things people at my level most often get wrong about it.
Why it works: The misunderstanding section is the most valuable part: it surfaces the non-obvious gaps in most explanations.
14
Research synthesis
I'm researching [topic]. My sources: [paste excerpts or notes]. Pull together where they agree, where they clash and why, the questions none of them answer, and what the strongest current consensus appears to be.
Why it works: Synthesis across sources is where these tools earn their keep; the value is in the connections, not in summarizing each source alone.
15
Devil's advocate
My position is [position]. Argue the other side as well as a smart, well-informed opponent would: real arguments, no strawmen. Then tell me which of those counterpoints I should lose the most sleep over.
Why it works: Explicitly asking for the strongest counterargument stops the model from sliding into agreeable, flattering agreement, a failure mode all of them share.
16
Literature overview
As of your knowledge cutoff, walk me through what's known about [topic]: the main schools of thought, the strongest evidence behind each, the questions still in dispute, and the researchers or sources worth knowing. Flag anything likely to be out of date.
Why it works: Naming the knowledge cutoff keeps any model from presenting stale training data as current, especially useful when the model can't browse.

Strategy

Strategy prompts

17
Decision framework
I'm choosing between [option A] and [option B]. My context: [constraints, goals, resources]. Score both on short-term impact, long-term impact, how reversible they are, resource cost, and fit with [goal]. Pick one and name the trade-off I'm accepting by choosing it.
Why it works: The reversibility and trade-off framing are what separate useful strategic advice from generic pros/cons lists.
18
First-principles breakdown
Take [problem / challenge / question] down to first principles. What's the real goal here? Which assumptions does the standard approach quietly rely on? And what would a fix look like if we ignored how it's normally done?
Why it works: Forces the model to challenge the frame of the question instead of answering it exactly as posed.
19
Pre-mortem
Fast-forward to [one year from now]: [project / plan] has failed outright. Working backward, list the five most probable reasons it collapsed (specific to my situation, not boilerplate) and for each, the move I could make today to head it off.
Why it works: Pre-mortem framing surfaces implementation risks that forward-looking planning systematically misses.

Go deeper with the Prompt Engineering track

Learn to GPT's prompt engineering track covers chain-of-thought reasoning, structured output, few-shot examples, and how to evaluate a prompt. Everything is taught model-agnostically, so the patterns transfer whether you end up on ChatGPT, Claude, or Gemini. Every lesson has a live sandbox so you practice, not just read.

Prompt Engineering CourseStart Free

Frequently Asked Questions

What makes a prompt effective on any model?

Specificity (clear role, audience, output format), structure (examples and constraints), and context (relevant background). The pattern of role + format + constraints + examples produces the best results on Claude, ChatGPT, and Gemini alike.

Can I use the same prompt on Claude, ChatGPT, and Gemini?

Usually, yes. A well-structured prompt is portable because all three respond to the same signals. You'll occasionally tweak wording for a model's quirks, but the bones carry over, which is exactly why we teach prompting model-agnostically.

How do I write a good system prompt?

Start with a role, specify the output format, add constraints (word count, tone, what to avoid), and include one or two examples of ideal output. This structure works whether the field is called a system prompt, custom instructions, or a gem.

What are the most common prompt mistakes?

Being too vague (no role, format, or constraints), giving no context (no model can infer your industry or goals), and asking for everything at once. Breaking a complex task into steps beats one giant prompt every time.

Practice every prompt in the live sandbox

Free forever. No credit card. Start in 60 seconds.

Try Learn to GPT Free