50 Best Claude Prompts: Copy-Paste Templates (2026)

50 Best Claude Prompts: Copy-Paste Tested Templates for 2026

Most people use Claude the way they use a search engine - type a vague question, get a vague answer, repeat. That approach gets you 20 percent of what Claude can do. The other 80 percent unlocks when you stop asking and start instructing. This guide gives you 50 copy-paste-ready prompts organized by use case - writing, coding, data analysis, strategy, research, marketing, productivity, and advanced patterns. Every prompt has been tested on Claude Opus 4.8 and Sonnet 4.6. Each one includes a brief note on why the structure works, so you can adapt it to your own context rather than just copying blindly.

Quick navigation note: if you want to go deeper on any category after reading this guide, the AI Prompts hub at Build Fast with AI has expanded prompt libraries, prompt engineering deep dives, and model-specific optimization guides across Claude, ChatGPT, Gemini, and the developer prompt patterns used in production agentic systems.

Writing Prompts (Prompts 1-10)

Claude is the strongest AI model for writing tasks that require consistency across long outputs, precise instruction-following, and nuanced tone control. The prompts below work significantly better than generic 'write me a blog post about X' requests because they specify the output structure, the reader, and what to avoid.

Pro tip for all writing prompts: always include your target reader in the prompt. 'Write for early-career marketers at B2B SaaS companies' produces radically different output than 'write for marketers.' The more specific the audience, the more usable the result.

Prompt 1 - Long-Form Article With Full Specs

Prompt 1

Task: Write a 1,200-word article arguing that [TOPIC].

Audience: [describe your specific reader].

Format: One strong hook sentence, then 4 H2 sections, then a punchy 2-sentence close.

Do NOT include: passive voice, "in today's landscape," generic CTAs, or unsupported claims.

Every section must open with a direct statement of the point, not a question.

Why it works: the negative constraints are as important as the positive ones. Claude follows explicit 'do not' instructions very precisely. Banning 'in today's landscape' eliminates the most common AI writing cliche in a single instruction.

Prompt 2 - Draft Feedback With Clarifying Questions

Prompt 2

I am giving you a draft. Your job:

1. Identify the 3 biggest structural weaknesses (not grammar).

2. Ask me 2 clarifying questions before revising.

3. After I answer, produce a revised version.

Do not revise until I answer the questions.

[PASTE DRAFT]

Why it works: asking Claude to ask clarifying questions before acting is one of the highest-leverage prompt patterns. It forces the model to surface assumptions instead of guessing. Use this whenever you have a draft that feels almost right but you're not sure why it isn't working.

Prompt 3 - Repurpose a Piece Into Multiple Formats

Prompt 3

Here is a [article / transcript / report]: [PASTE CONTENT]

Create the following from this single source:

1. A 280-character X/Twitter post (no hashtags)

2. A 3-bullet LinkedIn summary (each bullet under 20 words)

3. An email subject line (under 50 characters) and 3-sentence preview

4. A 60-second spoken summary (approx 150 words)

Keep the same core message across all formats. Adjust tone for each channel.

Why it works: giving Claude multiple output targets in one prompt produces more consistent messaging than running four separate prompts, because the model balances all four formats against the same source simultaneously.

Prompt 4 - Brand Voice Reverse Engineering

Prompt 4

Here are 5 pieces of our existing content: [attach or paste examples]

 Reverse-engineer a one-page brand style guide:
- Tone (3 adjectives with examples)
- Vocabulary: 10 words/phrases to use, 10 to avoid
- Sentence length pattern (short, long, or mixed)
- Formatting conventions we consistently use
A new writer should be able to match our style without seeing the originals.

Why it works: Claude excels at pattern extraction from examples. This prompt converts implicit brand style into explicit, teachable rules - useful for onboarding writers, briefing agencies, or maintaining consistency across a large content operation.

Prompt 5 - Contrarian Argument Generator

Prompt 5

Conventional wisdom in [INDUSTRY/FIELD] says: [COMMON BELIEF].

Make the strongest possible case AGAINST this view.
Requirements:
- Use real examples or data where possible
- Acknowledge where the conventional view is partially correct
- End with the single most actionable implication of the contrarian position
Length: 400-500 words.

Why it works: asking Claude to argue against a given position while requiring it to acknowledge where that position has merit produces more intellectually honest output than a pure contrarian take. The 'single most actionable implication' instruction forces a landing point rather than open-ended critique.

Prompt 6 - Edit for Clarity Without Changing Voice

Prompt 6

Edit this text for clarity and concision only.

Rules:
- Do not change the author's voice or distinctive phrasin
- Do not add information not in the original
- Cut passive voice and filler words
- Target 20% fewer words while keeping all key ideas
[PASTE TEXT]

Why it works: the 'do not add information' instruction prevents Claude from hallucinating or expanding the text. The 20% reduction target gives the model a concrete goal rather than editing arbitrarily.

Prompt 7 - Executive Summary Generator

Prompt 7

Summarize the following [report / document / meeting transcript] for a senior executive who:
- Has 3 minutes to read this
- Will make a resource allocation decision based on it
- Does not need background context
Format: 3 bullet points of key findings, 1 recommended action, 1 open risk.
Total length: under 150 words.
[PASTE CONTENT]

Why it works: framing the reader's decision context (resource allocation) changes what Claude chooses to include. This prompt reliably filters out background context and focuses output on decision-relevant information.

Prompt 8 - Email Thread Response Drafter

Prompt 8

Here is an email thread I need to respond to: [PASTE THREAD]
My goal in responding: [e.g., push back on the timeline without damaging the relationship]
My tone should be: [e.g., direct but collegial]
Things I must NOT say or imply: [e.g., we don't have the budget]

Draft 2 response options: one shorter (3 sentences max), one longer (1 paragraph).
Do not use: "I hope this finds you well," "Please let me know," or "As per my last email."

Why it works: giving Claude the strategic goal (push back without damaging relationship) produces output that balances the competing objectives. The negative constraints on clichéd email openers and closers improve output quality more than almost any positive instruction.

Prompt 9 - Technical Documentation to Plain English

Prompt 9

Translate this technical documentation into plain English for a non-technical stakeholder.
The reader is: [describe - e.g., a marketing manager who will present this to clients]
They need to understand: what it does, why it matters, and what they need to do (if anything)
They do not need to understand: how it works technically
Avoid all jargon. If a technical term is unavoidable, explain it in parentheses.
Length: match the original length, compressed by 30%.

[PASTE DOCUMENTATION]

Why it works: specifying what the reader needs to understand versus what they don't need to understand allows Claude to make principled omission decisions rather than trying to translate everything.

Prompt 10 - Argument Steelman

Prompt 10

Here is a position I hold: [STATE YOUR POSITION]
Steelman this position: construct the strongest possible version of the argument IN FAVOR of my view,
using evidence and reasoning I may have missed. Do not include weak arguments.
Then: give me the single strongest counterargument I need to be able to answer.

Why it works: steelmanning your own position (rather than having Claude argue against it) helps you find evidence and framings you may have overlooked. The 'single strongest counterargument' instruction prevents Claude from producing a laundry list and forces it to prioritize.

Coding and Development Prompts (Prompts 11-20)

Claude Opus 4.8 is the top-ranked coding model on SWE-Bench Pro as of June 2026, with Claude Fable 5 (currently offline) holding the DeepSWE #1 position at 70% PASS@1. For production coding tasks, the prompts below extract performance closer to the model's ceiling than default 'write me code to do X' requests. For a full 200-prompt developer library organized by task type, the Claude prompts for developers guide covers code review, debugging, security audits, migration planning, and more.

Prompt 11 - Code Review With Severity Classification

Prompt 11

Review this [Python/TypeScript/etc.] code. Classify every issue you find as:

- CRITICAL: security vulnerability or data loss risk

- HIGH: bug that will surface in production

- MEDIUM: performance or maintainability issue

- LOW: style or minor improvement

 

For each issue: line number, classification, description, and fix.

If you find no issues in a category, say so explicitly.

[PASTE CODE]

Why it works: severity classification forces Claude to prioritize and prevents it from treating a style issue the same as a security vulnerability. The explicit 'if none, say so' instruction reduces false positives.

Prompt 12 - Bug Root Cause Diagnosis

Prompt 12

Here is a bug I am experiencing:

Expected behavior: [describe]

Actual behavior: [describe]

Error message or stack trace: [paste]

Relevant code: [paste]

Your task:

1. Diagnose the root cause (not just the symptom)

2. Explain WHY this happens in 2-3 sentences

3. Provide the fix with commented code

4. List one thing I should check to prevent this class of bug in future

Why it works: asking for root cause versus symptom diagnosis changes the depth of the answer. The 'prevent this class of bug' instruction extracts a generalizable lesson rather than just fixing the specific instance.

Prompt 13 - Function Documentation Generator

Prompt 13

Write complete documentation for this function:

[PASTE FUNCTION]

Include:

- One-sentence description

- Parameters with types and constraints

- Return value with type and description

- Edge cases and error conditions

- One usage example


Use [Google / NumPy / JSDoc] docstring format.

Why it works: specifying the docstring format standard (Google, NumPy, JSDoc) ensures compatibility with your existing documentation tooling without requiring post-processing.

Prompt 14 - Code Migration Planner

Prompt 14

I need to migrate this codebase from [FROM] to [TO].

Example: from Boto3 API v2 to v3, or from Python 3.9 to 3.12, or from Express to Fastify.

 

Codebase summary: [briefly describe what it does and its approximate size]

 

Produce:

1. Migration risk assessment (what is most likely to break)

2. Step-by-step migration order (what to migrate first)

3. Automated vs manual changes breakdown

4. Testing strategy for validating the migration

Do not write migration code yet. Focus on planning.

Why it works: 'do not write code yet' is one of the highest-leverage instructions for complex tasks. It forces Claude to invest in planning before generating output, which prevents the most common failure mode: generating code that solves the wrong problem.

Prompt 15 - API Integration Scaffolding

Prompt 15

Write a [Python/TypeScript] client for this API: [PASTE API DOCUMENTATION OR ENDPOINT LIST]

Requirements:

- Retry logic with exponential backoff (max 3 attempts)

- Rate limit handling with automatic waiting

- Request/response logging at DEBUG level

- Type annotations on all public methods

- Error handling that distinguishes: network errors, auth errors, rate limits, and API errors

Include a usage example for the 3 most common endpoints.

Why it works: production-grade requirements (retry logic, rate limit handling, structured error types) produce code that actually works in production rather than a hello-world wrapper that fails the first time it hits an API limit.

Prompt 16 - SQL Query Optimizer

Prompt 16

Here is a SQL query that is running slowly:

[PASTE QUERY]

Table schemas: [paste or describe]

Approximate row counts: [e.g., users: 2M, orders: 50M]

Current execution time: [e.g., 8 seconds]

Identify: the top 3 performance problems in priority order.

For each: explain why it is slow and provide the optimized version.

If indexes would help, specify exactly which columns and why.

Why it works: providing approximate row counts gives Claude the data it needs to reason about index selectivity and join order. Without this context, query optimization advice is generic. With it, the model can prioritize meaningfully.

Prompt 17 - Security Vulnerability Audit

Prompt 17

Audit this code for security vulnerabilities.

Check specifically for:

- SQL/NoSQL injection

- XSS vulnerabilities

- Authentication and authorization flaws

- Secrets or credentials hardcoded or logged
- Dependency vulnerabilities (flag any libraries you recognize as historically vulnerable)

- Input validation gaps

For each finding: severity (Critical/High/Medium/Low), location, and remediation.

[PASTE CODE]

Why it works: explicit vulnerability category checklists prevent Claude from stopping after finding the first issue. The severity classification keeps output actionable for the security triage process.

Prompt 18 - Unit Test Generator

Prompt 18

Write unit tests for this function: [PASTE FUNCTION]

Test coverage requirements:

- Happy path (at least 2 variants)

- Edge cases (empty input, max values, boundary conditions)

- Error conditions (each error type the function can throw)

- At least one test with mocked dependencies if applicable

Use [pytest / jest / mocha / your framework] syntax.

Test names should describe the scenario, not the function.

Why it works: specifying that test names should describe the scenario ('returns empty list when input is None') rather than the function name ('test_process_items_3') produces tests that are readable as documentation.

Prompt 19 - Architecture Decision Record

Prompt 19

Write an Architecture Decision Record (ADR) for the following technical decision:

Decision: [what you decided]

Context: [what problem you were solving]

Options considered: [list the alternatives you evaluated]

Why you chose this option: [your reasoning] 

Format: Status / Context / Decision / Consequences / Alternatives Considered

Length: 300-400 words.

Tone: factual, future team members are the audience.

Why it works: ADRs are frequently written poorly because they omit context and rejected alternatives. This prompt structure forces documentation of both, making the decision meaningful to engineers who join the team six months later.

Prompt 20 - Pull Request Description Generator

Prompt 20

Write a pull request description for these changes:

[PASTE GIT DIFF or DESCRIBE CHANGES]

Include:

- What changed and why (not how)

- What was NOT changed that reviewers might expect (to prevent confusion)

- Testing approach

- Any deployment considerations or feature flags

- One-line summary for the merge commit message

Audience: engineers who did not write this code.

Why it works: 'what was NOT changed that reviewers might expect' is the most underused element of good PR descriptions. It prevents reviewers from spending review time asking questions about scope that the author already answered implicitly.

Data Analysis and Research Prompts (Prompts 21-28)

Claude's long-context window (1 million tokens on Opus 4.8) makes it uniquely capable for data analysis tasks that require ingesting entire datasets, reports, or research papers in a single session. The prompts below leverage that capability. For more structured data analysis templates, the gen-ai-experiments cookbook repository has working Python notebooks for Claude-powered data analysis workflows.

Prompt 21 - Exploratory Data Analysis Summary

Prompt 21

I am sharing a CSV dataset. Perform an initial exploratory data analysis.

[PASTE CSV HEADERS AND SAMPLE ROWS, or attach file]

Report:

1. Data quality issues (missing values, duplicates, type inconsistencies)

2. Top 3 distributions worth examining (with reasoning)

3. Top 3 correlations worth testing (with reasoning)

4. Questions this data can answer vs. questions it cannot

5. Recommended next analysis steps

Do not produce visualizations. Produce text findings only.

Why it works: asking Claude to explicitly state what the data cannot answer prevents the most common analytical error - overstating conclusions. The 'no visualizations' instruction keeps output in a format Claude can actually produce reliably.

Prompt 22 - Research Paper Summary for Non-Experts

Prompt 22

Summarize this research paper for a practitioner audience (not academics).

[PASTE PAPER or ABSTRACT]

Format:

- What problem did they study? (1 sentence)

- What did they find? (3 bullet points, plain English)

- How confident should we be in the results? (mention sample size, methodology limits)

- What can I actually do with this finding? (1 actionable recommendation)

- What is the biggest limitation of this study?

Why it works: the 'how confident should we be' instruction extracts methodological caveats that popular science summaries typically omit. This prevents practitioners from overcorrecting on preliminary or underpowered research.

Prompt 23 - Competitor Analysis From Public Sources

Prompt 23

Analyze [COMPETITOR NAME] based on these public sources:

[PASTE: job listings, LinkedIn page, website copy, press releases, reviews]

Extract:

1. Their apparent strategic priorities (inferred from hiring and investment signals)

2. Their positioning vs. our product: where they are stronger, where they are weaker

3. Which customer segments they appear to be targeting most aggressively

4. One move they could make in the next 12 months that would hurt us most

Distinguish clearly between what is stated vs. what is inferred.

Why it works: 'distinguish between stated vs. inferred' is the most important instruction for competitive intelligence. It forces Claude to label its deductions clearly, making the output far more useful for decision-making than a confident-sounding summary that blends fact and interpretation.

Prompt 24 - Survey Data Interpreter

Prompt 24

I have survey results I need to interpret:

[PASTE RESULTS]

Questions to answer:

1. What is the headline finding - the single most important thing these results show?

2. What segments of respondents differ most from the overall average and why might that be?

3. What does this data NOT tell us that we might assume it does?

4. If you had to recommend one action based on these results, what would it be and why?

Note: sample size is [N]. Treat findings accordingly.

Why it works: providing the sample size upfront allows Claude to calibrate its confidence levels appropriately. Without this, the model typically treats all survey data with the same authority regardless of whether the N is 18 or 18,000.

Prompt 25 - Literature Review Synthesis

Prompt 25

I have assembled notes or abstracts from [N] papers on [TOPIC].

[PASTE NOTES / ABSTRACTS]

Produce a structured literature review:

1. Main themes and what evidence supports each (cite paper numbers)

2. Areas of consensus across papers

3. Areas of disagreement or conflicting findings

4. The most significant gap in the existing research

5. One sentence conclusion synthesizing all the evidence


Do not introduce findings from outside the provided sources.

Why it works: 'do not introduce findings from outside the provided sources' is essential for research contexts where hallucination would corrupt the work. This instruction confines Claude to synthesis rather than generation.

Prompt 26 - A/B Test Results Interpreter

Prompt 26

Interpret these A/B test results:

Variant A: [conversion rate, sample size, time period]

Variant B: [conversion rate, sample size, time period]

Statistical significance: [p-value or confidence level if known]

Secondary metrics: [list any]

1. What do the results show?

2. Are the results statistically meaningful given the sample sizes?

3. What confounders could have influenced the outcome?

4. Should we ship Variant B? What would you need to be more confident?

Why it works: including secondary metrics and asking about confounders prevents the classic A/B test mistake of celebrating a statistically significant but practically meaningless win, or ignoring a result because it lacks p < 0.05 despite strong directional evidence.

Prompt 27 - Financial Report to Key Insights

Prompt 27

Analyze this financial report or earnings call transcript:

[PASTE CONTENT]

I need:

1. Revenue growth trend (last 3 periods if available)

2. Margin trajectory (improving, declining, or stable)

3. The biggest risk factor mentioned or implied

4. The biggest opportunity mentioned or implied

5. One thing management emphasized that I should be skeptical about and why

Audience: I am a [investor / competitor / potential employee] - frame accordingly.

Why it works: asking Claude to identify something management emphasized that you should be skeptical about is the prompt instruction that produces the most underrated analysis. It specifically targets spin and narrative management rather than taking executive framing at face value.

Prompt 28 - Meeting Notes to Action Items

Prompt 28

Convert these meeting notes or transcript into a structured action item list.

[PASTE NOTES / TRANSCRIPT]

Output format:

- Owner: [name or role]

- Task: [specific, measurable action]

- Due date: [if mentioned, or flag as "not specified"]

- Dependent on: [other action items this cannot start until completed]

Flag any commitments that were made but lacked a clear owner.

Flag any decisions that were discussed but not actually made.

Why it works: flagging commitments without owners and discussions without decisions surfaces the two most common meeting failure modes. These flags are more valuable than the action items themselves because they represent follow-up conversations that need to happen.

Strategy and Business Prompts (Prompts 29-35)

Claude handles multi-stakeholder strategic reasoning better than almost any other model when you give it named perspectives and specific constraints. The prompts below exploit this by giving Claude a defined decision framework rather than an open-ended 'what should I do?' question. For model-specific optimization on strategy tasks and the complete Claude AI guide with full capability breakdown, that resource covers how to calibrate prompt depth by use case.

Prompt 29 - Pre-Mortem Analysis

Prompt 29

I am about to launch [PROJECT / PRODUCT / DECISION].

Imagine it is 12 months from now and this has failed badly.

Work backwards: what are the 5 most likely reasons it failed?

For each failure mode:

- Probability (High / Medium / Low)

- Early warning signs I should watch for in the first 30 days

- One action I can take NOW to reduce this risk

Why it works: the pre-mortem technique is one of the highest-ROI planning exercises in decision research. Framing it as future retrospection unlocks Claude's ability to surface risks that present-tense 'what could go wrong?' prompts miss because they feel hypothetical rather than concrete.

Prompt 30 - Devil's Advocate Review

Prompt 30

Here is a business decision or strategy I am planning to execute:

[DESCRIBE PLAN]

Play devil's advocate. Make the strongest possible case against this plan.

Rules:

- Only use arguments that are actually credible (no strawmen)

- Include at least one risk I probably have not thought of

- End with: what would have to be true for this plan to succeed despite your objections?

Why it works: the final question ('what would have to be true for this to succeed') converts a critique into a conditional endorsement. It forces Claude to state the assumptions that underlie the plan rather than just attacking it, which is more useful for stress-testing than pure criticism.

Prompt 31 - Market Sizing Estimate

Prompt 31

Help me build a bottom-up market size estimate for [MARKET].

Known inputs:

- [any data points you have: population, price, frequency, etc.]

Walk through each assumption explicitly.

Flag which assumptions are weakest (most uncertain).

Give a range: conservative, base, and optimistic case.

Show the math at each step.

Do not give me a top-down TAM/SAM/SOM framework. Use bottom-up calculation only.

Why it works: explicitly banning the top-down TAM/SAM/SOM framework forces the model into a more rigorous calculation. Top-down market sizing is typically circular (it uses the same sources as the rest of the pitch deck). Bottom-up sizing forces actual assumption work.

Prompt 32 - OKR Writing Assistant

Prompt 32

Write OKRs for the following team goal for [QUARTER].

Context:

- Team: [name and function]

- Company priority this OKR should serve: [describe]

- Biggest constraint we face: [resource, time, dependency]

- What does success look like in plain English: [describe
 

Format: 1 Objective (aspirational, not a metric) + 3-4 Key Results (measurable, specific numbers).

Each Key Result should be a stretch but achievable. Do not write milestones as Key Results.

Why it works: the 'do not write milestones as Key Results' instruction addresses the most common OKR failure mode. Milestones (ship feature X by date Y) are outputs. Key Results should be outcomes (user retention increases by 15%). This distinction makes the difference between OKRs that drive accountability and OKRs that become checkbox exercises.

Prompt 33 - Job Description Quality Check

Prompt 33

Review this job description for the following problems:

[PASTE JD
 

1. Biased language that may discourage qualified candidates from applying

2. Requirements that are "nice to have" listed as "must have"

3. Vague phrases that tell candidates nothing about the actual job

4. Missing information candidates need to decide whether to apply

5. Compensation or growth path (is it mentioned? should it be?)

For each issue found: quote the problem text and suggest the fix.

Why it works: quoting the specific problem text rather than describing it generically makes this prompt immediately actionable. The output becomes a tracked-changes-style edit rather than a list of abstract suggestions.

Prompt 34 - Negotiation Preparation

Prompt 34

I am about to negotiate [SITUATION: salary, contract, partnership, etc.].

My position: [what I want]

Their likely position: [what you expect they want]

My BATNA (best alternative to no agreement): [describe]

Their likely BATNA: [your guess]

Prepare me:

1. What is my realistic zone of possible agreement?

2. What 3 concessions can I offer that cost me little but matter to them?

3. What should I ask for that I don't strictly need but can trade away?

4. The 2 most likely hardball tactics they will use and how to respond

Why it works: including BATNA analysis for both sides grounds the preparation in negotiation theory rather than advice. Asking for concessions that cost you little but matter to them is the most practical single negotiation technique, and Claude generates surprisingly good analysis when prompted this specifically.

Prompt 35 - Product Positioning Statement

Prompt 35

Write a product positioning statement for:

Product: [name and what it does]

Target customer: [specific segment, not "everyone"]

Category: [what category does this compete in?]

Key differentiator: [the ONE thing we do better than all alternatives]

Proof point: [one specific, credible claim that supports the differentiator]

Format: Use the Geoffrey Moore positioning template:

"For [target customer] who [pain point], [product] is a [category] that [key benefit]. Unlike [alternative], our product [differentiator]."
Then write 3 variations: corporate, casual, and technical.

Why it works: the Moore template forces the writer to make specific choices about category, customer, and differentiation. Three tone variations makes the positioning useful across different channels without requiring additional prompts.

Marketing and Content Prompts (Prompts 36-42)

These prompts produce marketing copy that avoids the two most common AI marketing failures: generic benefit claims and the passive, corporate tone that signals AI-generated content to readers. For context engineering patterns that make these prompts work at scale, the 150 Best Claude Prompts guide covers the advanced patterns including persona anchoring and constraint-first prompting for marketing content.

Prompt 36 - Cold Email Sequence

Prompt 36

Write a 3-email cold outreach sequence for [PRODUCT/SERVICE].

Prospect: [role and company type - e.g., Head of Engineering at Series B SaaS]

Pain point we solve: [specific problem, not category]

Social proof available: [customer logo, stat, or case study]


Email 1 (day 0): Lead with the pain, not the product. Under 100 words.

Email 2 (day 3): Share one piece of value (stat, case study, or insight). Under 80 words.

Email 3 (day 7): Direct close with one specific CTA. Under 60 words.


Do not use: "I hope this email finds you well," "quick question," or "circle back."

Why it works: the explicit per-email length targets and the specific banned phrases together constrain Claude toward the compact, direct style that cold email research shows generates the best reply rates. The 'lead with the pain, not the product' instruction for email 1 is the single most important cold email improvement available.

Prompt 37 - Landing Page Copy

Prompt 37

Write landing page copy for [PRODUCT].
 
Target visitor: [who is arriving on this page and from where?]

Primary CTA: [what action do you want them to take?]

Biggest objection to that CTA: [what stops most visitors from converting?]

Sections needed:

1. Hero headline (under 8 words) + subheadline (1 sentence)

2. 3 benefit statements (outcome-focused, not feature-focused)

3. Social proof block (adapt to: [customers/reviews/logos you have])

4. FAQ that addresses the biggest objection

5. CTA with urgency or specificity

Tone: [e.g., confident and direct, not hyped or fluffy]

Why it works: naming the biggest objection upfront allows Claude to thread it through the entire page rather than adding an FAQ at the end as an afterthought. Landing pages that address objections proactively (especially in the hero section) consistently convert better than those that lead only with benefits.

Prompt 38 - LinkedIn Post Generator

Prompt 38

Write a LinkedIn post about: [TOPIC or KEY INSIGHT]


Goal: [e.g., build credibility as an expert in X, drive traffic to Y, spark engagement]

My voice: [paste 1-2 previous posts you like, or describe your style]


Structure:

- Hook (first line, standalone, must stop the scroll)

- 3-5 short paragraphs (each under 2 sentences, lots of white space)

- One specific, counterintuitive or surprising statement

- Soft CTA that invites comments (not a hard link push)


Do not start with "I" or with a rhetorical question.

No emojis unless I explicitly confirm I use them.

Why it works: the two most common LinkedIn writing mistakes are opening with 'I' (which signals a self-promotional post before the reader has a reason to care) and rhetorical questions (which signal that the writer has nothing surprising to say). Banning both forces a stronger hook.

Prompt 39 - Case Study Draft

Prompt 39

Write a customer case study based on this information:

Customer: [name and brief description]

Problem before: [what they were struggling with]

Solution: [what they used our product/service for]

Results: [specific metrics - numbers, percentages, timeframes]

Customer quote: [if available]

Format: Situation / Complication / Solution / Results (SCSR)

Length: 500-700 words

Audience: similar prospects evaluating whether to buy

Tone: factual, credible, let the numbers speak

Why it works: the Situation/Complication/Solution/Results framework is a narrative structure that works as well for B2B case studies as the hero's journey works for fiction. The 'let the numbers speak' tone instruction prevents Claude from producing hyperbolic praise that reduces credibility.

Prompt 40 - Newsletter Issue Writer

Prompt 40

Write a newsletter issue on [TOPIC] for [AUDIENCE].


This week's angle or hook: [your unique perspective or news peg]

Subscriber context: [what do they already know? what do they care about?
 

Structure:

1. Opening hook (1 paragraph, personal or surprising)

2. Main insight or story (300-400 words)

3. 3-5 actionable takeaways for readers

4. One resource or link with a sentence on why it's worth their time

5. Closing line that makes readers look forward to next issue

Voice: [describe your existing newsletter tone or paste a past issue]

Why it works: the 'what do they already know' instruction prevents Claude from writing 101-level content for an expert audience. The 'closing line that makes readers look forward to next issue' instruction is the most underused element of good newsletters - it builds forward momentum rather than just closing.

Prompt 41 - Ad Copy Variants

Prompt 41

Write 5 ad copy variants for [PRODUCT/OFFER].

Funnel stage: [awareness / consideration / conversion]

Platform: [Google Search / Facebook/Instagram / LinkedIn / YouTube]

Character limits: [headline: X chars, body: Y chars if applicable]

Each variant must use a DIFFERENT hook angle:

- Fear of missing out

- Social proof / authority

- Curiosity gap

- Direct benefit claim

- Contrast / before and after

Label each variant with its hook angle. Do not repeat angles.

Why it works: forcing five different hook angles in the same prompt produces genuinely diverse creative options rather than five variations on the same approach. Labeling each with its angle also makes ad testing more systematic - you're testing the angle, not just the execution.

Prompt 42 - YouTube Script Outline

Prompt 42

Create a YouTube video script outline for this topic: [TOPIC]

Channel context: [your niche and typical video length]

Target viewer: [who is watching and why]

Desired outcome: [what should viewers do or think after watching?]

Outline sections:

1. Hook (0-15 seconds): why they MUST keep watching - specific, not "in this video..."

2. Problem setup (15-45s): the problem or question this video answers

3. Main content (section titles only, with suggested timing per section)

4. Key visual or demo moments to plan for

5. CTA (last 30s): one specific action

Do not write the full script. Outline only.

Why it works: 'do not write the full script' combined with an outline-level request produces something far more useful than a 3,000-word generated script that doesn't match your voice. The outline gives you structure to fill in with your own delivery rather than reading AI-written content on camera.

Productivity and Personal Use Prompts (Prompts 43-47)

These prompts are for everyday personal and professional use. Claude's precise instruction-following and long-context capability make it especially effective for these tasks. For the full advanced prompt patterns including KAIROS memory management and the multi-step chaining patterns that work best with Claude Code, the Claude AI Prompt Codes That Actually Work guide covers the Claude-specific prompt architecture in depth.

Prompt 43 - Weekly Plan From Chaos

Prompt 43

Help me create a prioritized weekly plan.

Inputs:

- My task list (brain dump): [paste everything]

- Non-negotiable commitments this week: [meetings, deadlines]

- My most productive hours are: [e.g., 9am-12pm]

- My goal for this week: [single most important outcome]

Output:

1. The 3 high-priority tasks I MUST complete (everything else is secondary)

2. Tasks to delegate or drop entirely

3. A suggested daily structure for Monday-Friday

4. What I am not doing this week that could cause problems later

Why it works: asking Claude to name what you are NOT doing this week converts a prioritization exercise into a risk awareness exercise. This is the step most productivity systems miss.

Prompt 44 - Difficult Conversation Prep

Prompt 44

Help me prepare for this difficult conversation:

Situation: [what happened or what needs to be addressed]

My goal: [what outcome do I want from this conversation?]

The other person's likely perspective: [your best guess at how they see it]

What I am afraid will happen: [name the feared outcome]

Provide:

1. A suggested opening statement (2-3 sentences, non-accusatory)

2. 3 questions I can ask to understand their perspective better

3. How to respond if they get defensive

4. A clear articulation of what I need from this conversation

Why it works: naming the feared outcome explicitly allows Claude to help you prepare for it specifically rather than giving generic 'active listening' advice. The opening statement output is immediately rehearseable.

Prompt 45 - Learning Plan Generator

Prompt 45

Create a 30-day learning plan to get me to [SKILL LEVEL] in [SKILL].

Starting point: [your current level - be honest]

Time available: [hours per week]

Learning style: [reading, videos, projects, courses, or mix]

Goal: [what do you want to be able to DO at the end of 30 days?]

Format:

- Week 1: foundations (what to cover)

- Week 2: core skills (what to build)

- Week 3: applied practice (what project to attempt)

- Week 4: consolidation and gaps (what to review)

One resource recommendation per week (be specific about title/URL if you know it).

Why it works: the week-by-week structure with specific project requirements forces Claude to connect learning to application, rather than producing a reading list with no practice component - which is what most AI-generated learning plans default to.

Prompt 46 - Performance Review Self-Assessment

Prompt 46

Help me write a self-assessment for my performance review.

Context:

- My role: [title and main responsibilities]

- Review period: [dates]

- 3 things I accomplished that I am proud of: [list with specifics]

- 1-2 things that did not go well and what I learned: [honest reflection]

- What I want to work on next year: [specific, not vague]

Tone: confident but not boastful, honest about growth areas.

Length: 400-500 words.

Do NOT include: buzzwords, "synergy," "bandwidth," or passive constructions like "was involved in."

Why it works: asking Claude to help structure your own accomplishments and honest reflections produces a more authentic document than asking it to write a performance review from scratch. You provide the substance; it provides the structure and phrasing.

Prompt 47 - Decision Framework Builder

Prompt 47

I am trying to decide: [DECISION]
What I know:

[list the relevant facts, constraints, and preferences you have

What I do not know:
[list the key uncertainties]
Help me think through this using a structured approach:

1. What type of decision is this? (reversible vs. irreversible, time-sensitive or not)

2. What is the single most important factor?

3. What would a person I respect do in this situation and why?

4. What would I regret more: making this choice or not making it?

5. Your recommendation and the key assumption it rests on.

Why it works: the 'regret minimization' question in step 4 is Jeff Bezos's famous decision framework and is one of the most reliable ways to cut through analysis paralysis for high-stakes personal decisions. Combining it with structural and analytical questions gives Claude both the rational and emotional decision context it needs to give a useful recommendation.

Advanced Prompt Engineering Patterns (Prompts 48-50)

These three prompts are not content-type prompts but structural techniques that make every other prompt on this list more powerful. Master these patterns and you can adapt any of the previous 47 prompts to produce even better results. For the complete advanced prompt engineering guide including persona anchoring, few-shot chaining, and context engineering for agentic Claude Code workflows, the gen-ai-experiments repository has worked examples for each pattern.

Prompt 48 - Role + Objective + Format + Constraints (ROFC) Template

Prompt 48

ROLE: You are a [specific expert with a relevant perspective].

OBJECTIVE: Your task is to [specific output you need].

FORMAT: Structure your response as [exact format].

CONSTRAINTS: Do not [list of things to avoid].
[PASTE YOUR CONTENT OR QUESTION HERE]
Before answering, restate in one sentence what you understand my goal to be.

If my goal is unclear, ask me one clarifying question before proceeding.

Why it works: the ROFC structure (Role, Objective, Format, Constraints) is the most consistently effective four-element prompt architecture across all content types. The 'restate my goal' instruction is a calibration step that catches misalignments before Claude generates a long output in the wrong direction. The single clarifying question instruction (not multiple questions) prevents the model from paralysing you with a questionnaire when a reasonable assumption would suffice.

Prompt 49 - Chain of Thought Activation

Prompt 49

Before giving me the answer, work through this problem step by step

Problem: [PASTE YOUR PROBLEM]

Step 1: Identify what type of problem this is.

Step 2: List what you know and what you need to figure out.

Step 3: Work through the solution or reasoning.

Step 4: Check your work - does your answer make sense?

Step 5: Give me your final answer in one clear sentence.

Why it works: explicit chain-of-thought prompting consistently improves Claude's performance on multi-step reasoning problems. The key is asking for visible reasoning steps (not just an answer), which allows Claude to catch and correct its own errors during generation. On complex mathematical, logical, or planning problems, this pattern can double output quality versus asking for the direct answer.

Prompt 50 - Iterative Refinement Loop

Prompt 50

This is a multi-turn refinement task

Round 1: Produce [INITIAL OUTPUT] based on this context: [CONTEXT].

After I see your Round 1 output, I will respond with one or more of:

- STRONGER: push this section further in the same direction

- SOFTER: dial this back

- DIFFERENT: try a completely different approach to this section

- KEEP: this section is right, move to the next

We will iterate until I say FINAL.

Do not revise sections I have not commented on.

In each round, tell me what you changed and why.

Why it works: establishing the refinement protocol in the first prompt creates a shared vocabulary for iteration that dramatically speeds up the editing process. 'Do not revise sections I have not commented on' prevents the most frustrating AI behavior: the model 'helping' by changing things you already got right. The 'what you changed and why' requirement maintains a decision trail through the editing process.

Frequently Asked Questions

What makes Claude different from ChatGPT when it comes to prompting?

Claude follows precise, structured instructions more reliably than GPT-4o for most task types. Where GPT-4o tends toward elaboration, Claude tends toward precision and adherence to specified formats. This makes negative constraints ('do not include', 'do not revise sections I have not commented on') particularly effective with Claude - the model takes them literally in a way that actually helps. Claude also tends to be more honest about uncertainty, which makes it better for analysis tasks where you need the model to tell you what it does not know rather than filling gaps with confident-sounding content.

How long should a Claude prompt be for the best results?

Prompt length should match task complexity. For simple reformatting tasks (bullet to prose, prose to bullet), 2-3 sentences are sufficient. For complex analytical or creative tasks, a well-structured 150-300 word prompt typically produces dramatically better results than a 20-word request. The highest-leverage addition is not more words but more specificity: a 50-word prompt with specific audience, format, and negative constraints often outperforms a 200-word prompt with general guidance. Do not pad prompts with explanations of why you need the output - Claude does not need that context and it dilutes the instruction signal.

Can I use the same prompts I use for ChatGPT on Claude?

Most prompts transfer, but Claude-specific optimizations exist. Claude responds especially well to: explicit format specifications, negative constraints ('do not include X'), named roles, and staged multi-turn instructions. Claude responds less well than GPT-4o to: vague emotional appeals for 'better' output, personas that ask it to pretend to be unconstrained (it will not comply and will tell you so), and single-word or very short prompts on complex tasks. Claude also tends to take 'never do X' instructions more literally, which is useful for enforcing constraints across long outputs.

What is prompt engineering and do I need it for Claude?

Prompt engineering is the practice of designing instructions to extract specific, high-quality outputs from AI models. For simple tasks, basic prompting is fine. For professional use cases where output quality directly affects business outcomes - client deliverables, code that ships to production, analysis that informs decisions - structured prompting techniques like the ROFC pattern (Prompt 48), chain-of-thought activation (Prompt 49), and iterative refinement protocols (Prompt 50) produce meaningfully better results. The return on prompt engineering investment is highest for tasks you do repeatedly.

How do I adapt these prompts to my specific use case?

The prompts in this guide are templates with placeholders in [brackets]. For one-time use, fill in the brackets and run. For recurring use, turn them into saved system prompts in Claude.ai (Settings > Saved Prompts) so you can call them with a keyword. For team use, build a prompt library in a shared Notion page or Claude's team workspace. The most valuable adaptation is adding your specific voice, audience, and tone preferences to the beginning of each prompt as a persistent context block - Claude will apply this context across every subsequent instruction.

Which Claude model should I use for these prompts?

For most prompts in this guide, Claude Sonnet 4.6 is the right default: it delivers the best cost-performance balance for writing, analysis, and strategy tasks. Use Claude Opus 4.8 for prompts that require deep reasoning (Prompts 31-35 strategy work, complex data analysis, architecture decisions) where response quality justifies higher cost. Claude Haiku 4.5 works for high-volume, structured formatting tasks where speed and cost matter more than nuance. Claude Fable 5 (currently offline due to the June 12 export control ban) was the best model for the coding prompts in Section 2 by a significant margin - watch for its restoration announcement.

Recommended Blogs

• 150 Best Claude Prompts That Work in 2026 - The Expanded Library With 8 Advanced Patterns
• 200 Claude Prompts for Developers: Code Review, Debugging, Security, and More
• 25 Claude Fable 5 Prompts to Test Every Capability (2026)
• Claude AI Prompt Codes That Actually Work - Secret Shortcuts Explained (2026)
• Claude AI 2026: Complete Guide to Models, Features, and Best Use Cases
• AI Prompts Hub - Complete BFWAI Prompt Library for Claude, ChatGPT, and Gemini
• Claude AI Complete Hub - Everything on Anthropic Models, Claude Code, and Featur

Start with the 5 prompts that match your most common tasks this week. Once you have tested and adapted them, you will have a personal prompt library that consistently outperforms generic requests. If you want the full prompt library with 150+ tested templates across every category, subscribe to Build Fast with AI for daily AI tools guides, prompt libraries, and model reviews.

References

• Anthropic - Claude Model Overview and Capabilities (2026)
• Anthropic - Prompt Engineering Guide (Official Documentation)
• Build Fast with AI - 150 Best Claude Prompts That Work in 2026
• Build Fast with AI - 200 Claude Prompts for Developers
• Build Fast with AI - Gen-AI Experiments Cookbook Repository
• Anthropic - Claude Fable 5 Launch and Benchmark Details (June 9, 2026)
• Build Fast with AI - Claude AI Complete Hub