Jump to section
Claude AI Cheatsheet
A practical Claude AI guide covering models, prompts, projects, Claude Code, and API usage.
Claude at a Glance
Claude is an AI assistant by Anthropic — for writing, research, coding, analysis, and learning. Start free at claude.ai.
what Claude can do
- Emails, reports, posts, scripts.
- Fix grammar, tone, and clarity.
- Summarise documents and PDFs.
- Compare options and spot patterns.
- Write, debug, and review code.
- Build working apps and tools.
- Brainstorm strategies and ideas.
- Work through multi-step problems.
- Break down complex topics clearly.
- Adjust to your knowledge level.
access & plans
- Haiku model. Limited daily messages.
- Start at claude.ai — no install needed.
- Sonnet + Opus. 5× more usage.
- Projects, Cowork, priority access.
- Much higher limits than Pro.
- Early access to Labs products.
Where to use Claude Browser ❧ claude.ai — the main chat interface Desktop ❧ Mac / Windows — hub for all Claude products Mobile ❧ iOS / Android — syncs with desktop Which model? Quick question / short summary ✔ Haiku Writing, email, code, daily use ✔ Sonnet (start here) Complex reasoning, long docs ✔ Opus
Not sure which model? Use Sonnet. Switch to Opus when the task is genuinely hard and Sonnet isn't cutting it.
Prompt Frameworks
Five structures that get better results than typing a random question. Each one suits a different kind of task.
RTF — role, task, format
- Who Claude should be.
- 'Act as a senior copywriter...'
- Exactly what to do.
- 'Write a product description for...'
- How the answer should look.
- Bullets, table, word count, tone.
RTF is the most versatile framework — use it as your default starting point. Example 1 — Marketing copy Prompt Act as an experienced copywriter. (Role) Write a product description for a standing desk. (Task) Under 100 words. Bullet points for key features. Energetic tone. (Format) ――――――――――――――――――――――――――――― Example 2 — Data analysis Prompt Act as a data analyst. (Role) Review the sales data below and find the top 3 trends. (Task) Output as a table. One sentence insight per row. (Format) ――――――――――――――――――――――――――――― Example 3 — Learning Prompt Act as a patient teacher explaining to a 12-year-old. (Role) Explain how the internet works. (Task) Use a simple analogy. Under 150 words. (Format) ――――――――――――――――――――――――――――― Example 4 — Code review Prompt Act as a senior Python developer. (Role) Review this function for bugs and readability issues. (Task) List each issue with a one-line fix. (Format)
If you only learn one framework, learn RTF. It works for writing, analysis, coding, learning — almost everything.
TAG — task, action, goal
- The core problem — the 'what'.
- How to approach it — the 'how'.
- The desired outcome — the 'why'.
Use TAG when you have a clear objective and want help figuring out how to get there. Example 1 — Content strategy Prompt Task: Increase viewer engagement on my YouTube channel. Action: Suggest two creative strategies for upcoming videos. Goal: Boost average view duration and interactions by 20%. ――――――――――――――――――――――――――――― Example 2 — Business growth Prompt Task: Our email open rate dropped from 35% to 18%. Action: Identify likely causes and suggest 3 tests to run. Goal: Recover to at least 28% open rate within 60 days. ――――――――――――――――――――――――――――― Example 3 — Personal productivity Prompt Task: I spend 3 hours a day on email and feel behind. Action: Suggest a daily email system I can start immediately. Goal: Spend under 45 minutes on email each day.
TAG works best for content strategy, business planning, and any problem where you know the outcome but not the path.
BAB — before, after, bridge
- The current problem state.
- The ideal outcome.
- The plan to get from before to after.
Use BAB for persuasive writing, ad scripts, sales copy, or showing a transformation. Example 1 — YouTube growth Prompt Before: We average 100 views per video and get almost no comments. After: 1,000 views per video and an active comment section. Bridge: Draft a plan with content changes and promotional tactics to get us there. ――――――――――――――――――――――――――――― Example 2 — Sales email sequence Prompt Before: Prospects say they're interested but go quiet after the first call. After: More prospects reply, book a second call, and convert. Bridge: Write a follow-up email sequence (3 emails) that keeps them engaged. ――――――――――――――――――――――――――――― Example 3 — Product landing page Prompt Before: Small business owners waste 2 hours a day on manual invoicing. After: Invoices are sent automatically in under 5 minutes. Bridge: Write a landing page hero section that tells this story compellingly.
BAB is the fastest framework for persuasive writing. The more specific your before/after, the better the output.
CARE — context, action, result, example
- Background Claude needs first.
- The specific task to carry out.
- The outcome you're looking for.
- A reference to make advice concrete.
Use CARE for evaluations and strategic analysis. The Example component stops Claude from giving generic advice. Example 1 — Channel analysis Prompt Context: I run a tech review channel with 50k subscribers. Views dropped 30% over 3 months. Action: Analyse possible reasons and suggest recovery actions. Result: A prioritised list of 3–5 recommendations with expected impact. Example: If video length is a factor, cite how MrMobile improved retention by cutting to 8 minutes. ――――――――――――――――――――――――――――― Example 2 — SaaS churn diagnosis Prompt Context: We sell a SaaS tool at $49/month. Churn is 8% monthly, mostly in month 2. Action: Diagnose the likely churn drivers based on this pattern. Result: 3 hypotheses with a cheap test I can run for each. Example: If onboarding is the cause, suggest how Slack's checklist approach could apply here. ――――――――――――――――――――――――――――― Example 3 — Newsletter feedback Prompt Context: I'm a non-native English speaker writing a newsletter for SaaS founders. Action: Review the draft below and identify what weakens it. Result: Specific improvements, not vague suggestions. Example: If the opening is weak, show how Lenny's Newsletter hooks readers in the first line.
If CARE feels long, at least keep the Example — it's the part that makes Claude's advice specific instead of generic.
RISE — role, input, steps, expectation
- Expert role for Claude to take.
- Specific data or info to work with.
- Ask for a step-by-step process.
- What the steps should achieve.
Use RISE when you need detailed step-by-step guidance and have specific data to share. Best for strategy roadmaps and process design. Example 1 — Growth roadmap Prompt Role: You are a YouTube growth strategist with 10 years of experience. Input: My channel has plateaued at 5k views per video for 4 months. Steps: Give a step-by-step plan covering content, schedule, and outreach. Expectation: Reach 10k average views and double engagement within 6 months. ――――――――――――――――――――――――――――― Example 2 — Hiring process Prompt Role: You are an experienced HR director at a tech startup. Input: We're a 12-person team hiring our first senior engineer. No HR process exists. Steps: Design a hiring process from job post to offer letter. Expectation: A repeatable process we can run in 3 weeks with a team of 3. ――――――――――――――――――――――――――――― Example 3 — Product launch plan Prompt Role: You are a product launch consultant. Input: We're launching a B2B invoicing tool next month. Budget is $5,000. Steps: Create a week-by-week launch plan covering channels and messaging. Expectation: 200 sign-ups in the first 30 days with minimal paid spend.
RISE gives the most detailed outputs. Use it when you need a plan you can actually act on, not just ideas.
Chained Prompting
Big tasks break. Chain them — each prompt builds on the last, and you can review and correct at every stage.
why chaining works
- One huge prompt is harder to get right.
- Each output feeds the next — like an assembly line.
- You catch and fix mistakes between steps.
Chain 1 — Writing a balanced article Step 1 — Research "List 5 strong arguments FOR remote work productivity. List only — no explanation." Step 2 — Counter-arguments "Now list 3 strong counter-arguments against each point you just gave me." Step 3 — Final output "Using both lists, write a balanced 600-word opinion article. Acknowledge both sides fairly." Why it works: each step is focused. Claude isn't juggling research, writing, and balance all at once. ――――――――――――――――――――――――――――― Chain 2 — Building a marketing campaign Step 1 — Audience "Describe the top 3 pain points of a 35-year-old SaaS founder who struggles with team communication." Step 2 — Messaging "Using pain point #1 from above, write 3 ad headline variations. Each under 10 words." Step 3 — Full ad "Take the strongest headline and write a full Facebook ad — headline, body (50 words), and CTA." Why it works: you pick the best headline yourself before Claude builds on it. ――――――――――――――――――――――――――――― Chain 3 — Debugging code Step 1 — Understand "Explain what this function is supposed to do, line by line: [paste code]" Step 2 — Diagnose "Now find everything that could cause it to return None instead of a value." Step 3 — Fix "Fix the most likely bug. Show the corrected function and explain what changed."
For complex projects — chain your prompts. Don't try to do everything in one shot.
Ready-to-Use Prompts
Copy-paste prompts for common tasks. Swap the brackets for your actual content.
writing & content
"Write a [length] [format] about [topic] for [audience]. Tone: [tone]." "Rewrite this in a [tone] tone. Keep the same meaning: [paste text]" "Summarise this in [number] bullet points: [paste content]" "Create an outline for a [blog post / essay / course] on [topic]." "Give me 10 [post / headline / hook] ideas for a [niche] audience." "Expand this into a full paragraph: [paste sentence or note]" "Write a [LinkedIn / Twitter / Instagram] post about [topic]. Under [word count] words." "Take this rough draft and improve the structure without changing my voice: [paste]"
Paste 2–3 examples of your own writing and say 'match this style' — Claude writes in your voice.
"Write a [friendly / professional / firm] email to [recipient] about [topic]." "Rewrite this email to sound more polite and under 80 words: [paste]" "Write a follow-up email for someone who didn't reply to this: [paste original]" "Proofread and fix this email I wrote: [paste]" "Write a cold outreach email to [target] offering [value]. Subject line included." "Draft a reply to this message: [paste received message]" "Create a 3-email onboarding sequence for new customers of [product/service]."
Add the goal: 'I want them to book a call' or 'I want a yes/no reply' — Claude will write toward that outcome.
research & learning
"Explain [topic] like I'm [age / background]. One analogy, under 200 words." "What's the difference between [X] and [Y]? Give a concrete example." "Give me a simple analogy for [complex concept]." "Teach me [skill] in 5 steps. Start with the most important one." "Quiz me on [topic] — 5 questions, one at a time. Tell me if I'm right after each." "Summarise [topic] into the 5 things I need to know to understand it." "What are the most common misconceptions about [topic]?" "Recommend books similar to [title] and explain why each one fits."
Tell Claude your level — 'I know nothing' or 'I have basic knowledge of X' — and the explanation adjusts automatically.
coding
"Write a [language] function that [does X]. Add inline comments." "Explain what this code does, line by line: [paste code]" "Find the bug. It should return [X] but it returns [Y]: [paste code]" "Refactor this to be more readable without changing the behaviour: [paste]" "Write unit tests for this function: [paste code]" "Convert this [Python / JS / SQL] into [target language]: [paste code]" "Create a regex that matches [pattern]. Explain each part." "Review this code for security issues and suggest fixes: [paste]"
Always include the full error message when debugging — Claude fixes issues much faster with the stack trace.
data & analysis
"Analyse this data and tell me the 3 most important findings: [paste]" "Create a table from this data: [paste]" "Which category has the highest average? Show your working: [paste data]" "Clean this data: remove duplicates, fix date formats, fill blanks: [paste]" "Summarise this spreadsheet for someone non-technical." "Write a Python script to plot [chart type] from this CSV data." "Extract all [emails / numbers / dates] from this text: [paste]" "Compare [Option A] vs [Option B] in a table. Criteria: [list criteria]."
Paste data directly into chat — Claude reads tables, lists, and raw numbers without needing a file upload.
problem solving & decisions
"Find all the problems with this [plan / document / argument]: [paste]" "What am I missing from this [plan / strategy / design]?" "What are the risks of [approach]? How would you mitigate each one?" "Give me a pros and cons list for [decision]. Be specific." "Create a step-by-step plan for [goal]. Include rough timelines." "I need to decide between [A] and [B]. Here's my situation: [context]. What would you recommend and why?" "What questions should I be asking about [topic] that I'm probably not asking?"
'What am I missing?' is one of the most useful prompts you can give Claude — it finds blind spots fast.
feedback & editing
"Proofread this and fix grammar, clarity, and flow: [paste]" "Cut this by 30% without losing the main points: [paste]" "Make this more persuasive for a [audience] audience: [paste]" "Change the tone of this to [conversational / formal / direct]: [paste]" "Check this argument for logical gaps or weak points: [paste]" "Here's my first draft. Give me 3 specific improvements and explain each one: [paste]" "Be brutally honest about what's wrong with this: [paste]"
Say 'be brutally honest' if you want real criticism. Claude holds back unless you ask it not to.
Projects & Memory
Projects keep your files, instructions, and conversation history in one place — across sessions.
what projects do
- PDFs, code, docs — stay in the project.
- No re-uploading each session.
- Set tone and rules once — applies every chat.
- e.g. 'Always reply in British English'
- Claude remembers context from previous sessions.
- Pick up exactly where you left off.
Good uses for Projects → Client work One project per client, with their documents → Writing Style guide, drafts, and research in one place → Coding Codebase and conventions always in context → Learning Study materials and Q&A together with progress Custom instruction examples Marketing team "Always respond in British English. Brand voice: friendly and concise. Never use jargon. Always end with a call to action." Coding project "Next.js 14, TypeScript, Tailwind CSS. App Router only. Add JSDoc comments to all functions. No unused variables." Personal writing "I'm a non-native English speaker. Use simple, clear English. Preserve my meaning — just fix the language."
One Project per major area of work — not one for everything. Focused context gives better results.
Artifacts & Files
Claude creates live documents, apps, and charts in a side panel — and can read any file you upload.
artifacts — what Claude can build
- Python, JS, TypeScript, SQL.
- Copy or download instantly.
- Markdown reports, plans, structured text.
- Editable directly in the panel.
- Working web pages with live preview.
- Calculators, forms, dashboards.
- Diagrams, icons, illustrations.
- Charts and interactive widgets.
Prompts that create Artifacts "Write a Python script that converts CSV to JSON." "Build an interactive budget calculator in HTML." "Create a React pricing table component." "Make a bar chart showing these sales numbers: [data]" Iterating on an open Artifact "Add a dark mode toggle." "Make this mobile-responsive." "Add input validation to the form." "Refactor this into separate components."
Build in steps. Create the basic version first, then add features one at a time — each iteration updates the same Artifact.
uploading files & images
- PDF, Word, Markdown, plain text.
- Summarise, extract, or analyse.
- CSV, Excel, JSON.
- Ask questions in plain English.
- Claude can see and describe images.
- Diagnose errors, analyse charts, convert wireframes to code.
After uploading, ask "Summarise this PDF in 5 bullet points." "Find errors or ambiguities in this contract." "Analyse this spreadsheet and show the top trends." "What's wrong with this UI? How would you improve it?" "Turn this wireframe into HTML and CSS code." "What does this error message mean and how do I fix it?"
Take a screenshot of anything confusing — an error, a form, a chart — and ask Claude to explain or fix it.
Connectors & MCP
Connect Claude to your real tools — so it works with actual emails, files, and tasks instead of hypotheticals.
what connectors do
- Gmail, Google Calendar, Google Drive.
- Work with real emails and events.
- GitHub, GitLab, databases.
- Query repos and data directly.
- Asana, Linear, Jira, Notion.
- Create tasks, read boards, update tickets.
What you can ask once connected "Summarise all emails from this week about the project launch." "Draft a reply to the latest email from [name]." "What meetings do I have tomorrow?" "Create a new Asana task for each item in this list." "Find all TypeScript files in this repo that use useEffect." How to connect claude.ai → Settings → Integrations → Add a connector Gmail, Google Calendar, and Google Drive connect in one click. Start there — immediately useful for most people.
Connectors turn Claude from a text assistant into a real work tool that acts on your actual data.
The Claude Ecosystem
Claude isn't just a chat window. Here's the full product map — so you know where to go next as your needs grow.
everyday products
- Main chat interface — works in any browser.
- Best starting point for most users.
- Mac and Windows — hub for all Claude products.
- Required for Cowork. Connects to local files.
- iOS and Android — syncs with desktop.
- Assign Cowork tasks from your phone.
agentic products — Claude does the work
- Give it a goal — Claude works through your files and apps to finish it.
- For knowledge workers: analysts, researchers, operations, legal, finance.
- Lives in your terminal. Reads your codebase, writes and edits real files.
- For developers building software.
The difference between chat and agentic Chat (claude.ai) Claude replies to your messages. You copy and paste the output yourself. Cowork / Claude Code Claude has permission to read, edit, and create real files — so it completes the task, not just describes it. Cowork runs inside the Desktop app. Claude Code runs in the terminal. Both are available on Pro and Max plans.
If you spend more than 20 minutes copying Claude's output into your files, Cowork will save you that time.
Microsoft 365 — Claude inside Office
- Build spreadsheets, charts, and formulas from a conversation.
- Generally available on paid plans.
- Create and edit presentations from prompts or docs.
- Generally available on paid plans.
- Draft, redline, and compare documents.
- Generally available on paid plans.
- Triage inbox, draft replies, schedule follow-ups.
- In beta on paid plans.
Anthropic Labs & API
- Turn prompts into prototypes, slides, one-pagers, and branded visuals.
- Research preview — Pro, Max, Team, and Enterprise.
- Build apps, chatbots, and automations powered by Claude.
- Python and JavaScript SDKs. Pay per use.
The full product map Everyday chat → claude.ai, Desktop, Mobile Autonomous file work → Cowork (non-technical), Claude Code (developers) Inside Microsoft Office → Excel, PowerPoint, Word, Outlook Visual creation → Claude Design (Labs, research preview) Build your own product → Claude API / Platform
All products share the same models. A Project you build in claude.ai carries context into Cowork and the API.
Tips & Mistakes
The habits that separate great results from frustrating ones.
iterate — first answers aren't always final
After getting an answer, say "Make this shorter." "Too formal — make it more conversational." "Good, but add a section about [X]." "Try a completely different approach." "Give me three variations of this." "That's not right. [Correct info]. Try again." "You're misunderstanding my question. What I meant is..."
Treat Claude like a collaborator, not a one-shot machine. The second or third attempt is often much better.
prompt mistakes to avoid
- 'Write about marketing' → Claude doesn't know what you want.
- Always add: audience, length, tone, purpose.
- 10 requests in one prompt = messy output.
- One goal per prompt = cleaner results.
- Accepting the first answer even when it's off.
- Always give feedback — 'too formal', 'wrong tone'.
✗ "Write me something."← No format, no audience, no goal ✗ "Fix my code, add tests, write docs, and make it faster."← Four separate tasks in one prompt
The biggest mistake is treating Claude like a search engine. It's a collaborator — give context, iterate, give feedback.
know Claude's limits
- Can't browse the web unless web search is on.
- Knowledge has a cutoff — verify current facts.
- Doesn't remember previous chats by default.
- Use Projects to keep context across sessions.
- May state incorrect facts confidently (hallucination).
- Always verify medical, legal, and financial claims.
Always verify before acting on: → Medical, legal, or financial advice — consult a professional → Specific statistics — check primary sources → Code before production — test thoroughly Never paste passwords, API keys, or real customer data into Claude. Anonymise before sharing sensitive information.
Claude is a powerful assistant, not an oracle. Use it for 80% of the work — apply your own judgement on the parts that matter.
No login required to share feedback
More Cheatsheets
Keep your reference handy
Explore more zero-to-hero cheatsheets for the tools you use daily.