Claude Prompts

Example: Upload a 60 page vendor proposal and replace {DOCUMENT TYPE} with “vendor proposal for migrating our CRM from Salesforce to HubSpot” and {LENGTH} with “60 pages.”

Why this works with Claude: Claude’s 1 million token context window means it actually reads the entire document rather than summarizing from fragments. The “be direct” instruction leverages Claude’s willingness to give honest critical feedback when explicitly invited. Most AI tools default to positive framing; Claude will flag genuine weaknesses when asked.

Output format: Five numbered sections with increasing specificity.

Example: Replace {CONTENT TYPE} with “board update memo,” {AUDIENCE} with “the board of directors who have 5 minutes to read this,” {TOPIC} with “Q1 results and revised Q2 forecast,” {TONE} with “confident but candid about the miss in EMEA,” and {WORD COUNT} with “400 words.”

Why this works with Claude: Claude’s writing naturally avoids the most common AI filler, but the explicit ban list eliminates the remaining patterns. The “every paragraph must contain a specific fact” constraint plays to Claude’s strength in producing substantive prose rather than fluffy generalizations. The result typically needs less editing than any other AI tool’s output.

Output format: Formatted document ready to send.

Example: Replace {CONTENT TYPE} with “product launch email,” {DEADLINE OR AUDIENCE} with “it goes to 50,000 subscribers tomorrow,” and paste your draft.

Why this works with Claude: Most AI tools default to positive feedback regardless of quality. Claude responds well to explicit permission to be critical. The “do not tell me what I did well” instruction overrides Claude’s politeness defaults. The structured format (quote, problem, rewrite) ensures feedback is actionable rather than vague. This prompt consistently produces the most useful editorial feedback available from any AI tool.

Output format: Numbered issues, each with quote, diagnosis, and rewrite.

Example: Replace {LANGUAGE} with “Python” and paste your authentication middleware, API endpoint handler, or data processing pipeline.

Why this works with Claude: Claude Code leads coding benchmarks for a reason: the underlying models excel at systematic code analysis. The five dimension framework gives Claude structure to work through methodically. The “do not invent problems” instruction prevents the false positive issue common with AI code reviews. The severity ratings help you prioritize fixes for real production impact.

Output format: Five sections with severity rated issues and corrected code blocks.

Example: Replace {DECISION} with “whether to accept a job offer at 15% more pay but at an earlier stage startup” and {SITUATION} with “I have been at my current company for 3 years, just got promoted, and have good equity. The startup has Series A funding and 30 employees.”

Why this works with Claude: Claude’s extended thinking capability means it actually works through multi-step reasoning rather than pattern matching to a generic framework. The “be honest, not diplomatic” instruction and the “advising a friend” reframe both push Claude past its cautious defaults into genuinely useful analysis. The cognitive bias check is something Claude handles particularly well compared to other models.

Output format: Five numbered sections with concrete analysis.

Example: Replace {MEETING TYPE} with “weekly product sync” and {DATE} with the meeting date, then paste raw meeting notes, transcript, or a combination of both.

Why this works with Claude: Claude’s context window handles even long meeting transcripts in full. The five category structure forces extraction of actionable output rather than a narrative summary. The “do not include discussion” instruction prevents Claude from padding the summary with conversational back and forth. The “flag ambiguity” instruction leverages Claude’s honesty about uncertainty rather than forcing confident but wrong attributions.

Output format: Five sections with structured, actionable items.

Example: Replace {CONTENT TYPE} with “blog post,” {NUMBER} with “5,” {OUTPUT FORMAT} with “LinkedIn posts,” {PLATFORM} with “LinkedIn,” {SPECIFIC CONVENTIONS} with “no hashtag spam, use line breaks for readability, end with a discussion prompt not a CTA,” and {LENGTH} with “150 to 200 words each.”

Why this works with Claude: The “no two pieces should make the same point” constraint is where Claude outperforms. Other models tend to rephrase the same core message. Claude actually extracts distinct angles because its reasoning evaluates overlap before generating each piece. The platform conventions field lets you teach Claude the specific norms of any platform without relying on its training data assumptions.

Output format: Numbered posts, each with a hook and standalone message.

Example: Replace {CONTRACT TYPE} with “SaaS vendor agreement” and paste the full contract text.

Why this works with Claude: This prompt is only practical on Claude because it requires processing entire contracts (often 20 to 50 pages) in a single conversation. Claude’s 1 million token context window handles this natively. The “organize by section” instruction means the output maps directly to the contract structure, making it easy to follow. The explicit “not my lawyer” framing gets more practical, actionable analysis than generic legal caution.

Output format: Section by section commentary following the contract’s structure.

Example: Upload a spreadsheet and replace {DATA TYPE} with “CSV” and {DESCRIPTION} with “monthly revenue by product line and region for the last 18 months, columns are Month, Product, Region, Revenue, and Units Sold.”

Why this works with Claude: The “what this data does NOT tell us” section leverages Claude’s strength in epistemic honesty. Most AI tools will overstate conclusions from limited data. Claude, when explicitly asked, will flag when sample sizes are too small, when correlation does not imply causation, and when confounding variables exist. The Slack summary at the end gives you an immediately usable deliverable.

Output format: Five sections with data backed findings and a ready to send summary.

Example: Replace {PROCESS NAME} with “handling a customer escalation,” {WHO DOES IT} with “our customer success team of 4 people,” and {FREQUENCY} with “3 to 5 times per week.”

Why this works with Claude: This is a multi-turn conversational prompt that uses Claude’s strength in asking focused follow up questions. The “one at a time” instruction prevents Claude from dumping 20 questions at once (a common AI pattern). The structured interview approach extracts tacit knowledge that the user often does not realize they have. Claude’s writing quality then produces documentation that actually reads well rather than feeling like a robotic checklist.

Output format: Interview questions (interactive), then a formatted SOP document.

Example: Replace {OPTION A} with “building an internal tool,” {OPTION B} with “buying a SaaS solution,” and {USE CASE} with “managing our 200 person company’s expense reporting workflow.”

Why this works with Claude: The “criteria that actually matter” instruction prevents the generic comparison trap where AI lists the same 5 criteria regardless of context. Claude’s reasoning ability shines in the weighting and scoring phase because it can articulate why a criterion matters more or less for the specific use case. The conditional recommendation format (“choose A if, choose B if”) produces more useful output than a single winner declaration.

Output format: Weighted comparison table followed by a conditional recommendation.

Example: Replace {PROJECT TYPE} with “product launch,” {PROJECT NAME} with “Q3 Mobile App Redesign,” {YOUR ROLE} with “Product Manager,” {TEAM SIZE AND COMPOSITION} with “12 people across engineering, design, and marketing,” {TIMELINE} with “design freeze June 15, beta July 1, launch August 1,” {WHERE THINGS STAND} with “wireframes approved, engineering sprint 1 starts Monday,” and {KEY DOCUMENTS} with “PRD, wireframes, competitive analysis, launch checklist.”

Why this works with Claude: This is a meta prompt that sets up a Claude Project for ongoing use. The custom instruction set Claude writes becomes the persistent context for every future conversation in that Project. This leverages Claude’s Project feature so you build context once and reuse it across dozens of conversations without repeating your situation each time. The recurring tasks list helps you anticipate the prompts you will use most.

Output format: Project instructions, task list, naming conventions, and upload priorities.