Anthropic Claude · Updated October 2026
The best Claude prompts for writing, coding, research, business, learning and Claude Code — each tagged with the Claude model it suits best. Copy, fill in the brackets, paste. No signup.
These 50 Claude prompts are written for the current Claude models — Sonnet 5.5, Opus 5.5, Haiku 5.5 and Fable 5.1 — not the Claude 3 and Claude 4 models many older prompt lists were built on. Every prompt follows the same structure: a clear goal, the context Claude needs, a few constraints, and the exact output format.
Each card shows the model that handles the task best for the cost. If you are not sure, use Sonnet 5.5 — it is Claude's everyday default. Filter by category below, replace the [bracketed] parts, and paste the prompt into Claude, Claude Code or the API.
Filter by category. Every prompt works on the free Claude plan unless it needs Claude Code.
Edit this draft for clarity and flow. Keep my voice, my opinions and my examples. Cut repetition and filler, fix grammar, and tighten long sentences. Do not add new ideas. Return the edited text, then a short list of the 5 biggest changes you made and why. [Paste draft]
Turn my notes into a [1,200]-word blog post for [audience]. Open with the problem in the first two sentences, use 4 sections with descriptive H2 headings, and end with one concrete next step. Plain, direct English — no phrases like "in today's fast-paced world". [Paste notes]
Rewrite this text for [new audience, e.g. 'a 12-year-old' / 'a busy executive' / 'a technical reviewer']. Keep every fact, change the vocabulary, length and examples to suit them. Target length: [words]. [Paste text]
Write 10 headline options for this article, each under 65 characters, in 5 styles (2 each): how-to, number list, question, bold claim, curiosity gap. Then write 3 alternative opening sentences. Mark your top pick and say why in one line. [Paste article or summary]
Write a [2,000]-word short story from this premise: [premise]. Genre: [genre]. Point of view: [first / close third]. Show character through action and dialogue rather than description, keep one clear turning point, and end on an image rather than a moral. Avoid clichés like waking from a dream.
Write this week's issue of my newsletter about [topic] for [audience]. Sections: a 2-sentence intro, 3 short items (each with a bold lead line, 60–90 words, and why it matters), one recommendation, and a sign-off in my tone: [describe tone]. Under 600 words total. [Paste links or notes]
Act as a developmental editor. Read this manuscript excerpt and give feedback on: structure and pacing, character motivation, clarity of stakes, and where a reader is likely to lose interest. Quote the exact passages you mean. Do not rewrite the text — tell me what to fix and why. [Paste excerpt]
Explain what this code does to a developer who is new to the codebase. Cover: its purpose in one sentence, the inputs and outputs, the main steps in order, and anything surprising or risky. Keep it under 250 words. ``` [Paste code] ```
Review this diff like a senior engineer. Group findings into Bugs (will break), Risks (could break under some condition) and Suggestions (optional). For each: file and line, what is wrong, and the smallest fix. Skip anything a linter or formatter would catch. What the PR is for: [context] [Paste diff]
Find the cause of this error. Error: ``` [Paste stack trace] ``` Code: ``` [Paste relevant code] ``` Expected: [expected behavior]. Already tried: [attempts]. Give the most likely cause with the evidence for it, the fix as a code change, and a test that would have caught it.
Write tests for this module with [Jest / pytest / Go test]. Cover the normal path, edge cases (empty, very large, invalid input) and error handling. Name each test after the behavior it checks. Afterwards, list any behavior that looks like a bug rather than a design choice. [Paste module]
Write a [PostgreSQL / MySQL / BigQuery] query that answers: [question]. Schema: [Paste table definitions] Return only the query, then one line on any assumption you made about the data.
Rebuild the UI in this screenshot as a [React + Tailwind / HTML + CSS] component. Match layout, spacing and hierarchy; use semantic HTML and accessible labels; use neutral placeholders for images. After the code, list anything you could not read clearly and what you assumed. [Attach screenshot]
Review the architecture of this system before we scale it to [target, e.g. 10× traffic]. Identify the 3 parts most likely to fail first, explain why, and propose a change for each with its trade-off. Point out anything that will be hard to change once clients depend on it. [Paste architecture description, diagrams or key files]
Review this code for security problems: authentication and sessions, input validation, injection, secrets handling, access control and rate limiting. For each issue give severity (Critical / High / Medium / Low), how it could be exploited in one sentence, and the fix as a code change. Also list what you checked and found no problem with. [Paste code]
Summarise this document for someone who has 2 minutes. Give: the main point in one sentence, 5 key findings with the page or section each comes from, anything the author is uncertain about, and 3 questions I should ask next. Do not add information that is not in the document. [Paste or attach document]
Read these sources and write a 1-page brief answering: [question]. Include the answer in 3 sentences, the key evidence with its source, where sources disagree and which is more reliable and why, and what is still unknown. Use only the sources provided. [Paste or attach sources]
Give me the strongest honest case for and against this position: [position]. For each side: the 3 best arguments, the best evidence for each, and the weakest point an opponent would attack. Finish with what evidence would change the answer. Do not pick a winner.
Analyse this data to answer: [question]. Give the answer in 3 sentences, the 3 numbers that support it, one chart I should build (type, x-axis, y-axis), and any data-quality problem that could change the conclusion. [Paste CSV or table]
I am researching [topic]. From the papers below, build a map of the field: the main schools of thought, the key paper for each, the methods they use, where they agree and disagree, and the 3 most promising open questions. Cite each claim to a paper by title. [Paste abstracts or attach papers]
List every factual claim in this draft as a numbered table with columns: Claim, Type (number / date / quote / general fact), How confident you are it is correct (High / Medium / Low), and What I should check it against. Do not rewrite the draft. [Paste draft]
Compare [option A], [option B] and [option C] for [use case]. Build a table with rows for [criteria, e.g. price, setup time, key features, limits] and one column per option. Below it, say which option fits best for [my situation] in 2 sentences, and note anything you are unsure about.
Write a reply to this email. Goal: [what I want, e.g. 'decline politely but keep the relationship']. Tone: [warm / neutral / firm]. Under 120 words, with a clear next step in the last sentence. [Paste email]
Turn these notes into a table: Action, Owner, Due date. Write "unassigned" or "no date" where they were not stated — do not guess. Below the table, list decisions made and questions left open. [Paste notes or transcript]
Write a one-page memo recommending what we should do about [decision]. Structure: Recommendation (2 sentences), Context, Options considered (3, with pros and cons), Risks and how we reduce them, What we need to decide by [date]. Write for [audience, e.g. the leadership team]. [Paste background]
Synthesise these customer interviews. Give: the top 5 problems ranked by how many people raised them (with counts), a representative quote for each, surprises that contradict our assumptions ([list assumptions]), and 3 product changes worth testing. [Paste transcripts]
Write landing page copy for [product] aimed at [customer]. Sections: headline and subheadline, 3 benefits (outcome first, feature second), how it works in 3 steps, 4 FAQs that handle objections, and a call to action. Use plain language and no unverifiable claims.
Write a job description for a [role] at [company type]. Include a 3-sentence summary, 6 responsibilities, must-have and nice-to-have skills (kept separate), and what the first 90 days look like. Use inclusive language and avoid jargon like "rockstar" or "ninja".
Read this contract and list the clauses a [buyer / supplier / freelancer] should worry about: liability, termination, payment terms, IP ownership, renewal and anything unusual. For each, quote the clause, explain the risk in plain English, and suggest what to ask for instead. This is not legal advice — flag anything a lawyer should review. [Paste contract]
Explain [concept] to someone with no background in [field]. Use one everyday analogy, then the accurate explanation, then a common misconception and why it is wrong. Under 200 words.
Teach me [topic] by asking me questions one at a time instead of explaining. Start with what I probably already know, wait for my answer, then build from it. If I am wrong, give a hint rather than the answer. After 8 questions, summarise what I have learned and what to study next.
Build a 30-day plan to learn [skill] with [time] per day, starting from [current level]. Group it into 4 weeks with a goal for each, list daily tasks, include free resources by name, and add a small project at the end of each week to prove progress.
Write a 10-question quiz on [topic] for [level]. Mix multiple choice, short answer and one applied problem. Put the answers and a one-line explanation for each at the end, after a line that says ANSWERS, so I can test myself first.
Here is an exam question and my answer. Grade it against [mark scheme / level], tell me what earned marks and what lost them, and show what a full-marks answer would include. Do not rewrite my whole answer. Question: [paste] My answer: [paste]
I just joined a team that owns this codebase. Give me a tour: what the system does, the main modules and how data flows between them, the 5 files I should read first and why, and the conventions I should follow. End with 5 questions to ask my teammates. [Paste file tree and key files]
Implement [feature]. Before writing any code, give me a plan: files to change, new files, data model changes and risks. Wait for my OK. Then build it in small steps that each leave the app working, with tests for the new behavior.
Get the test suite passing without skipping or disabling any test. Run the tests, find the root cause of each failure before changing code, make the smallest fix, and re-run. Ask me before any change to public behavior. Finish with a list of changes: file, reason, test fixed.
Explore this repository and write a CLAUDE.md for it: what the project does, how to install, build, test and lint (exact commands), the folder structure, coding conventions you can see in the code, and gotchas a new contributor would hit. Keep it under 150 lines and only include things you verified.
Refactor [module or pattern] to [goal, e.g. 'remove the duplicated validation logic']. Do not change behavior. First add tests that pin the current behavior if they are missing. Then refactor in small commits, running the tests after each one. Summarise what changed and anything you left alone on purpose.
Upgrade this project from [library + version] to [target version]. Read the migration notes below, list each breaking change that affects this codebase with the files it touches, apply them one area at a time with tests passing between steps, and flag anything that needs my decision. [Paste migration notes]
Build [describe the app] with [stack]. Work in checkpoints: 1) setup and a running empty app, 2) data model and storage, 3) core feature, 4) UI, 5) tests and README. After each checkpoint, run the app and tests, report what works, and list choices I might want to change. Never move on with failing tests.
[Endpoint / page / job] is slow: [symptoms, e.g. 'p95 of 4s']. Profile it, find where the time goes, rank causes by cost, and propose a fix for the top 3 with the expected improvement. Implement the top fix only after showing me the measurement, then measure again.
Plan 5 dinners for [number] people this week. Constraints: [diet, allergies, budget, max cooking time]. Reuse ingredients across meals to cut waste. Give each recipe in 6 steps or fewer, then one combined shopping list grouped by supermarket section.
Plan [number] days in [place] for [who is travelling] in [month]. Budget: [budget]. We like [interests] and dislike [dislikes]. For each day: morning, afternoon and evening with travel time between stops, one food suggestion, and a rainy-day backup. Flag anything that needs booking ahead.
Help me decide between [option A] and [option B]. My priorities, in order: [list]. Ask me up to 5 questions first if you need more information. Then score each option against each priority, explain the trade-off in plain language, and tell me which way you lean and what would change your mind.
I need to talk to [person] about [issue]. My goal is [goal] and I want to keep the relationship [good / professional]. Write an opening line, 3 key points in calm "I" statements, likely responses and how to answer each, and a way to close the conversation.
Here are my monthly income and expenses. Group spending into categories, show each as a percentage of income, point out the 3 easiest places to save [amount], and suggest a simple rule I can follow. Not financial advice — just a clear picture. [Paste list]
Suggest 10 gift ideas for [person, age, relationship] who likes [interests]. Budget: [budget]. Mix 3 practical, 3 experiences, 2 handmade and 2 surprising picks. One line each on why it suits them.
Look at this photo of [my fridge / my desk / a room / a whiteboard]. Describe what you see, then turn it into [recipes I can make / a decluttering plan / a furniture layout / a task list]. Tell me if anything in the image was unclear. [Attach photo]
Here is something that went wrong: [describe what happened]. Help me run a blameless review: the timeline, what I knew at each decision point, the root cause (not just the trigger), what I would do differently, and one habit that would prevent it next time.
API prices are per million input / output tokens. Each model has its own page of 20 dedicated prompts.
| Model | Best for | API price | Prompts |
|---|---|---|---|
| Claude Haiku 5.5 | Quick replies, rewrites, summaries, high-volume batch work | $0.10 / $0.50 | Haiku 5.5 prompts |
| Claude Sonnet 5.5 | Everyday default — writing, coding, analysis, Claude Code | $2 / $10 | Sonnet 5.5 prompts |
| Claude Opus 5.5 | Long documents, deep reasoning, architecture and refactors | About 2× Sonnet 5.5 | Opus 5.5 prompts |
| Claude Fable 5.1 | The hardest research, security and long agent tasks | $10 / $50 | Fable 5.1 prompts |
| Claude Sonnet 5 | Previous Sonnet — still available, same price as 5.5 | $2 / $10 | Sonnet 5 prompts |
| Claude Opus 5 | Previous Opus — strong reasoning and analysis | Higher than Sonnet | Opus 5 prompts |
| Claude Fable 5 | Original Fable — long-context research and writing | Highest tier | Fable 5 prompts |
The best Claude prompts state the goal, give the context Claude needs, set a few hard constraints, and describe the exact output format. The 50 Claude prompts on this page follow that pattern for writing, coding, research, business, learning, Claude Code and everyday tasks. Replace the [bracketed] parts with your own details and paste the prompt into Claude.
Yes. Every prompt on this page works in the free Claude app on the web, desktop and mobile. The Claude Code prompts need Claude Code, and very long documents or codebases work best on a paid plan because of usage limits. The prompts are free to copy, edit and reuse without attribution.
Start with Claude Sonnet 5.5 — it is the default for most tasks and the best balance of speed, quality and cost. Use Claude Haiku 5.5 for quick, simple or high-volume work. Move up to Claude Opus 5.5 for very long documents and deep reasoning, and to Claude Fable 5.1 for the hardest research, security and long agent tasks. Each prompt here is tagged with the model it suits best.
Be clear and direct, and explain why you want something, not just what. Give Claude the material to work from (paste the document, code or data), say who the output is for, and describe the format you want — headings, a table, a word count. For bigger tasks, ask Claude to plan first and wait for your OK, and ask it to list any assumptions it made so you can check them.
Most good prompts work on both, because the basics — clear goal, context, constraints and format — are the same. Claude tends to follow detailed instructions and long pasted documents especially well, so you can give it full source material instead of a summary. Claude Code prompts are Claude-specific because they rely on Claude working directly in your repository.
Claude Code is Anthropic's coding agent that works in your terminal and repository. Claude Code prompts give it a goal plus rules for how to work: plan before coding, run the tests after each change, never skip a failing test, and ask before changing public behavior. The Claude Code section on this page has 7 ready-made prompts for features, failing builds, refactors, upgrades and long tasks.
Quick replies, rewrites, summaries, high-volume batch work — 20 prompts
Everyday default — writing, coding, analysis, Claude Code — 20 prompts
Long documents, deep reasoning, architecture and refactors — 20 prompts
The hardest research, security and long agent tasks — 20 prompts
Previous Sonnet — still available, same price as 5.5 — 20 prompts
Previous Opus — strong reasoning and analysis — 20 prompts
Original Fable — long-context research and writing — 20 prompts
Turn a short idea into a detailed prompt for Claude or ChatGPT
How to pick the right effort level for every task
More free generators in this collection — no signup, unlimited use.
20 copy-paste prompts for Meta Muse Wonder 1.0 — frontier multimodal AI with 1M context, text/image/video/audio input.
20 copy-paste prompts for Mistral Large 4 — cybersecurity, finance, engineering and multimodal analysis.
20 copy-paste prompts for Claude Haiku 5.5 — high-volume classification, summarization and batch tasks at $0.10/M tokens.
20 copy-paste prompts for Claude Sonnet 5.5 by effort level — fast agentic coding and visual tasks.
20 free prompts for Claude Opus 5.5 — deep reasoning for long documents and audits.
20 copy-paste prompts for OpenAI GPT-6 Sol — coding, agents and long documents.