Steal the good ones
AI prompt examples: steal them, then make them yours
A good example prompt is the fastest way past the blank box — but a copied prompt gives you someone else’s audience, someone else’s constraints, and someone else’s idea of “good.” The result lands one notch off, and most people blame the model. Below is a curated set of prompt examples worth stealing, each with the reason it works — plus a free tool that does the one step that actually matters: adapting the example to your situation. Load one, drop in your specifics, watch it become yours.
Pick an example, add your specifics, and remix it into a prompt that’s actually about you — the bracket gets filled with your words. That last step is the whole skill.
No sign-up, nothing sent anywhere — the remix runs in your browser. Here is the thinking behind it.
Why copied prompt examples usually disappoint
Almost everyone has felt this small heartbreak. You find a prompt example online — upvoted, screenshotted, promised to “change how you use AI” — you paste it in, and… the answer is fine. Grammatical, plausible, and somehow not for you. The instinct is to blame the example or the model. Both miss the cause: the example worked beautifully — it just worked for the person who wrote it.
Here is what every polished example quietly hides: it was built around a specific situation. When someone shares a “perfect cold email prompt,” it was perfect because it encoded their product, their buyer, and their sense of a good email. Strip those out to make it shareable and you are left with a scaffold — useful, but generic by construction. Paste the scaffold and the model does what it always does with an open gap: fills it with the statistical average. You did not get a bad prompt. You got a prompt with your details missing.
Watch the mechanism up close. An example that reads “Write a compelling product description for my item” is a template pretending to be a prompt. “My item” is a hole. “Compelling” is a hole. “Product description” could be for a spec sheet or a gift guide. The model guesses all three, and it guesses average. Now fill the holes: “Write a 60-word product description for a $34 refillable steel water bottle, for commuters who hate single-use plastic, in a plain, slightly witty voice.” Same example, same model — but now there is one reasonable answer instead of a thousand. You stopped making it guess.
So here is the reframe the rest of this page is built on: an example prompt is not a finished product, it is a starting point one honest edit away from great. The people who get astonishing results from AI are not hoarding secret prompts — they treat every example as a draft to adapt, and that adapting, not the finding, is the actual skill. Conveniently, it is one you can learn in about the time it takes to read the next three sections.
A curated set of prompt examples that work — and why
Below are twelve examples across the six things people most often ask AI to do. Each one is a real, paste-able prompt, and each comes with the reason it works, because a prompt you understand is a prompt you can adapt. Read them less as scripts to memorize and more as a pattern language: once you see why the blog outline locks its format or why the debug prompt forbids a rewrite, you start reaching for the same moves everywhere. The brackets are the seams — they are where your situation goes.
Writing
Writing prompts fail in the most boring way: the model gives you correct, forgettable prose because you asked for a topic instead of a brief. The fix is to hand it a role and a shape.
Blog-post outline
You are a seasoned content strategist. Create a detailed blog-post outline about [topic]. Return a working title, a one-line hook, 5–7 H2 sections (each with 2–3 bullets), and a closing CTA. Reader is a smart beginner; keep it clear and jargon-free.
Why it works — The persona raises the ceiling and the exact output shape (title → hook → sections → CTA) means you get a skeleton to draft from, not a paragraph about drafting.
Rewrite in a voice
Here is the voice I want: “[one short sample]”. Rewrite the passage below in that same voice — same rhythm, same plain confidence. Keep the meaning; change only how it sounds. Text: [paste].
Why it works — One concrete sample teaches the model more than ten adjectives. Show the voice, do not describe it — the highest-leverage move in the set.
Business
Business prompts live or die on constraints. The difference between a cold email that gets deleted and one that gets a reply is almost never the offer — it is the clichés you told the model to avoid.
Cold email that gets replies
You are a veteran B2B copywriter. Write a 90-word cold email about [offer and audience]. Open with one specific observation about their world, make one clear offer, end with a low-friction question. No buzzwords, no “hope this finds you well,” no fake urgency.
Why it works — Naming the banned phrases is the trick. The model defaults to the average cold email; the negative space drags it back toward something a human would actually send.
Meeting notes → action items
You are a sharp chief of staff. From the notes below, give (1) a 3-bullet summary of decisions, (2) an action table with Owner, Task, Due, and (3) open questions. If an owner or date is missing, write “unassigned,” do not guess. Notes: [paste].
Why it works — One transform, a locked table, and an explicit rule for missing data. That last instruction is what stops the model from inventing a deadline nobody agreed to.
Coding
The best coding prompts do not ask for a rewrite — they ask for a diagnosis. Telling the model what NOT to do (don’t rebuild everything) is often more valuable than telling it what to do.
Explain this code
You are a patient senior engineer. Explain the code below in three passes: one sentence on what it does overall, a short line-by-line of only the tricky parts, then one edge case it might hit. I know the language, not this codebase. Skip the obvious. Code: [paste].
Why it works — The three-pass structure stops both failure modes at once — over-explaining basics and hand-waving the hard part. You control the altitude.
Debug helper
Act as a calm pair-programmer. Bug: [describe + error]. Do not rewrite everything. List the 3 most likely causes ranked by probability, the fastest check for each, then the smallest fix. Ask before guessing.
Why it works — “Do not rewrite everything” and “ask before guessing” convert a shotgun rewrite into a ranked hypothesis list — the way a good engineer actually debugs.
Learning
A learning prompt should make the model teach, not lecture. The magic words are an analogy requirement and a comprehension check — and, for real retention, forcing one question at a time.
Explain it like I’m new
You are a patient tutor. Explain [concept] as if I’m curious but brand new. Use one everyday analogy, define any jargon the moment you use it, and finish by asking me one question to check I understood. Under 200 words.
Why it works — The analogy forces the model to build a bridge from something you already know, and the check-question turns passive reading into a tiny active-recall loop.
Quiz me
You are a study coach. I need to learn [subject and goal]. Give a 5-day plan, then quiz me: ask ONE question at a time, wait for my answer, tell me why I’m right or wrong, and adapt the next question to my mistakes. Start now.
Why it works — One question at a time is the whole trick. It turns an info-dump you would skim into a conversation you cannot skim, which is where learning actually happens.
Image
Image prompts reward the vocabulary of a photographer or art director. Vague adjectives (“beautiful,” “high quality”) get you the average of everything; concrete nouns and light direction get you a picture.
Photo-real scene
Write a photorealistic image prompt for [subject and setting]. Brief it like an art director: subject and pose, environment, time of day and quality of light, lens and depth of field, color palette, mood. One paragraph, concrete nouns over adjectives, and name the one detail that should draw the eye first.
Why it works — Naming the focal point is what saves the image. Without it you get a flat, evenly-lit scene where nothing matters; with it the model knows where to spend its attention.
Logo concepts
You are a brand-identity designer. Propose three logo concepts for [brand and what it does]. For each: a one-line idea, the core shape, a two-color palette with hex codes, and the feeling. Keep them readable as a 16px favicon. No gradients, no clip-art clichés.
Why it works — The favicon constraint quietly forces simplicity — anything that survives at 16 pixels is a real mark, not decoration. Constraints do the design thinking for you.
Personal
The most underused prompts are the ones for ordinary life. They work because your constraints are specific and real — the people, the time, the budget — which is exactly what a generic template can never know.
Weekly meal plan
You are a practical cooking coach. Build a 5-dinner plan for [diet, constraints, who you’re feeding]. Each night: a dish, a 5-ingredient-max list, a rough time. Reuse ingredients to cut waste, keep weeknights under 30 minutes, and end with one shopping list grouped by aisle.
Why it works — The reuse-ingredients rule and the consolidated list are what make it usable on a Tuesday. A recipe dump is trivia; a plan you can shop from is a tool.
Tough-conversation script
You are a calm communication coach. Help me prepare for: [situation and who]. Give one opening line that lowers the temperature, three points as sentences I could say out loud, one thing to avoid, and a graceful way to close if it gets tense. Honest, not manipulative.
Why it works — “Sentences I could actually say out loud” and “honest, not manipulative” keep it a rehearsal instead of a manipulation script — a rare place where a guardrail is also an ethic.
Notice what these twelve share. None is long and none is clever. Each just refuses to leave an important decision to the model: it names a role, states one clear task, hands over the real specifics, sets a constraint or two, and locks the output shape. That is the entire anatomy of a prompt that works — every example above is the same five moves wearing different clothes. Learn to see the moves and you stop collecting prompts and start writing them.
The remix method: adapting an example to your situation
So you have found an example that is close. Here is the reliable way to close the last gap — three edits, in order of how much they matter, so if you only do one you do the right one.
1. Swap the slot — put your real subject in
Every example above has a bracket: [topic], [paste the code], [offer and audience]. That bracket is not decoration — it is the one place the prompt is deliberately blank, waiting for you. Replacing it with your actual subject is the highest-return edit, and it is the one the tool at the top of this page does for you: load an example, type your specifics, get the filled-in version. Do only this and you are already ahead of anyone who pasted the raw template.
2. Add one line of context — who it’s for and what “great” looks like
This is the edit that separates a decent result from an uncanny one — and the one people skip. The model does not know your reader, your stakes, or your taste unless you tell it, so add a sentence: “This is for skeptical enterprise buyers who have been burned before.” A single concrete example of the outcome you admire does more work than a paragraph of adjectives, because you are showing the target instead of describing it.
3. Tighten one guardrail — a length, a tone, a word to avoid
Finally, claim the output as yours by adding a constraint the original author could not have known. A word count. A banned phrase. A “no preamble, just the answer.” Constraints are not limits on the model; they are how you stop it from handing you the average and make it hand you your version. The negative space — what to avoid — is especially potent, because it teaches you your own taste as much as it steers the model.
That is the whole method, and it takes under a minute: swap the slot, add your context, tighten one rule. What makes it powerful is that it is repeatable — the same three edits turn any example, in any category, from someone else’s prompt into one calibrated to you. And the moment a prompt is calibrated to you, it becomes worth keeping. That is where the real leverage begins.
From one example to a system: find, copy, fork
Here is the part nobody tells you: adapting one example is a skill, but never having to adapt it twice is a superpower. The people who get the most out of AI are not remixing from scratch every morning — they keep a library of prompts they have already adapted, and start every task by reaching for the closest one. That is the entire idea behind PromptFork, and it maps onto three moves.
Find
Start from an example that already works — search the library by goal, model, or task instead of the blank page.
Copy
One click puts a community-tested prompt on your clipboard, ready to paste and run. Great for a one-off.
Fork
Make it yours — swap the slot, add your context, tighten a rule — and save your version to your library forever.
Forking is the remix method with a memory. Copying gives you the example; forking gives you the example plus a saved home for your adapted version, so next time you start from your calibrated prompt, not the generic one. Do that a few dozen times and you have built something no single clever prompt could be: a personal toolkit where every entry is already shaped to your work.
And when the closest example still is not close enough — a brand-new task, a weird edge case — that is what Studio is for. You describe your goal in plain language and it builds a precision prompt from scratch to refine and publish back to the library. It is the “Supercharge with AI” button on the remixer above: rough idea in, forged prompt out, five free every day. The examples teach you the moves; forking means you make each move only once; Studio covers the cases the library has not met yet. Same quiet edge — you never start from zero again.
Examples worth forking right now
Enough theory. Here are real, community-tested prompts you can copy or fork this minute — each one an example already shaped and proven in the wild. Find the closest, fork it, and make it yours.
Turn ChatGPT into a gentle shadow work guide
A prompt that makes AI lead a real shadow-work session — one probing question at a time, reflecting patterns back, ending with an integration practice.
Runway image-to-video — camera moves on your own photos and art
Runway's image-to-video mode adds a camera move to any still — a photo, painting, or product shot — without distorting what's in it. The key is using Runway's camera presets correctly and describing only the camera, not the scene.
Prompt: an endless creative writing prompt generator
Give it your genre and tone; get original story sparks on demand — each with a character, a situation, and built-in conflict.
YouTube first 30 seconds engineered for maximum retention (with re-hooks by content type)
Script the first 30 seconds using curiosity gap theory and pattern interrupts — with the critical first-3-second visual hook, B-roll direction, and re-hook lines tailored to tutorials vs commentary vs storytelling.
Midjourney logo & brand mark — scalable marks that pass the favicon test
Vector-ready logos built on negative space and geometric precision — includes the favicon scalability test, two style examples (monoline vs emblem), and the Ideogram workflow for adding text that doesn't look garbled.
Open-world (GTA-style) game build prompt
Scopes a 3D open-world prototype realistically — character controller + drivable vehicle + map first, bigger systems phased.
Questions about AI prompt examples
What are the best AI prompt examples to start with?+
Start with the examples that match a task you do weekly, because you will get instant feedback on whether the output is any good. For most people that means a blog-post outline, a cold email, a "explain this code" walkthrough, a "teach me this concept" tutor prompt, or a meal plan. The best starting example is not the cleverest one — it is the one you can judge, because you already know what a great result looks like. Every example on this page is grouped by category so you can jump straight to yours.
Can I just copy and paste prompt examples?+
You can, and it is a fine first move — a strong example beats a blank box every time. But a copy-pasted prompt gives you the example author’s audience, their constraints, and their idea of "good," not yours, so the output usually lands one notch off. The examples here are written to be adapted: each has a bracketed slot you replace with your own specifics. Copy to get moving; remix to make it fit. The remixer tool on this page does exactly that swap for you.
How do I adapt a prompt example to my own situation?+
Three moves. First, swap the bracketed slot for your real subject — the topic, the code, the audience. Second, add one line of your context: who it is for and what "great" looks like to you (an example of the outcome you admire is worth more than a paragraph of adjectives). Third, tighten one guardrail — a length, a tone, a word to avoid. That is the entire remix method, and the tool above does the first move automatically so you can focus on the two that carry the most weight.
Do these prompt examples work in ChatGPT, Claude, and Gemini?+
Yes. Every example here is built from model-agnostic fundamentals — a role, a clear task, your specifics, constraints, and an output format — which raise quality on every major model. Each model has its own quirks worth tuning for (some reward very explicit structure, others lean on long context or images), which is why PromptFork keeps per-platform hubs for ChatGPT, Claude, and more. But you can paste any example on this page into any of them today and see the difference immediately.
Where can I find more AI prompt examples?+
The examples on this page are a curated starter set; the full PromptFork library is the deep well. Browse it by goal, model, or task, then copy or fork the ones that fit — forking saves your adapted version so you never rebuild it from scratch. And when no example is close enough, Studio turns a plain-language description of your goal into a precision prompt you can refine and publish back to the library. Find, copy, fork, and generate — four ways to never start from a blank page again.
An example is a starting point. Make it yours.
Fork a prompt that already works, or describe your goal and let Studio build one. Your first adapted prompt is thirty seconds away.