AI Prompts for Claude Advanced Prompting

Write a system prompt that configures Claude as [DESCRIBE ROLE, e.g. a senior product manager / a rigorous legal analyst / a Socratic tutor]. The system prompt should define: the role and expertise level, the communication style and tone, what the assistant should and should not do, how to handle uncertainty, and the format for responses. Make it specific enough to change behavior noticeably.

Write a system prompt for an AI assistant embedded in [DESCRIBE PRODUCT, e.g. a customer support interface / a writing tool / a coding assistant]. Define: the assistant's purpose and scope, what topics are in and out of scope, the tone and persona, how to handle edge cases and questions outside scope, and the response format. Make it production-ready.

I want Claude to adopt this persona: [DESCRIBE PERSONA]. Write the system prompt. Define: the character's knowledge domain, their communication style, their perspective and opinions on [RELEVANT TOPICS], what they will not do or say, and how they handle questions outside their expertise. The persona should be consistent and distinctive.

I want a system prompt that configures Claude to run a specific workflow: [DESCRIBE WORKFLOW, e.g. code review / editorial review / customer discovery interview]. The system prompt should: define the process step by step, specify the output format at each step, set the evaluation criteria, and tell Claude when to ask for more information versus proceed.

Here is my current system prompt: [PASTE]. I have been using it for [DESCRIBE USE CASE] and it is producing [DESCRIBE ISSUES]. Diagnose why the prompt is not working as intended, identify the specific instructions that are too vague or conflicting, and write an improved version. Then propose three test cases I should use to evaluate the improvement.

I need Claude to extract information from text and return it as JSON. Here is the schema I need: [DESCRIBE OR PASTE JSON SCHEMA]. Write a prompt template I can use repeatedly with different input texts that will reliably produce JSON matching this schema. Include instructions for handling missing fields and edge cases.

I need Claude to produce reports in a consistent format every time. Here is the template I want: [DESCRIBE TEMPLATE]. Write a prompt and system prompt combination that will reliably produce output in this exact format, even for diverse input content. Test it on this example: [PASTE EXAMPLE INPUT].

I have unstructured text and I need to extract specific fields from it. The fields I need: [LIST FIELDS WITH DESCRIPTIONS]. Write a prompt that extracts these fields reliably, handles ambiguous or missing values gracefully, and returns the result in [FORMAT: JSON / table / bullet list]. Test on this example: [PASTE EXAMPLE TEXT].

I need Claude to classify [TEXT TYPE] into these categories: [LIST CATEGORIES WITH BRIEF DESCRIPTIONS]. Write a prompt that: explains each category clearly, handles items that could fit multiple categories, includes a confidence level in the output, and provides a brief reason for the classification. Make it reliable across diverse inputs.

I run [DESCRIBE TASK] repeatedly with different inputs. Build a prompt template I can reuse that: uses [VARIABLE PLACEHOLDERS] for the parts that change, includes all the context that stays constant, produces consistent output format, and handles the most common edge cases. Document how to fill in each placeholder.

I want Claude to reason through problems step by step before answering. Write a system or user prompt that reliably triggers this behavior for [TYPE OF PROBLEM, e.g. logic puzzles / strategy decisions / debugging]. The prompt should ask Claude to show its work, not just the conclusion, and to flag uncertainty in specific reasoning steps.

I need Claude to help me make complex decisions by thinking out loud. Write a prompt framework that: first identifies the relevant considerations, then reasons through each one, then synthesizes a recommendation, and finally flags what could change the recommendation. Test it on this decision: [DESCRIBE DECISION].

I want Claude to check its own outputs for internal consistency before responding. Write a prompt that instructs Claude to: produce its initial answer, then review that answer for logical errors, contradictions, or unsupported claims, and revise before presenting the final answer. Show both the draft and the revised version.

I want a prompt pattern I can use to have Claude challenge any position I take. The prompt should configure Claude to: always assume there is a strong opposing view, find the best evidence against my stated position, identify the weakest assumption in my reasoning, and avoid agreeing until I have addressed the strongest objection.

I want to solve [COMPLEX PROBLEM] by breaking it into a chain of smaller reasoning steps. Design a multi-step prompting sequence: the first prompt establishes the problem and gathers what we know, subsequent prompts each solve one sub-problem, and the final prompt synthesizes to an answer. Write all the prompts in sequence.

I want to build a multi-turn conversation workflow for [DESCRIBE TASK, e.g. requirements gathering / coaching session / iterative writing]. Design the conversation flow: the opening prompt that sets context and expectations, the follow-up prompts at each stage, how to handle user responses that go off-path, and the closing prompt that synthesizes the session.

I am working on [DESCRIBE PROJECT] across multiple Claude sessions. Design a "context handoff" template I can paste at the start of each new session to recreate the relevant context. What information should it include, how should it be structured, and how do I keep it concise enough to not waste context window on setup?

I want to use Claude to iteratively improve [TYPE OF OUTPUT, e.g. a piece of writing / a code module / a strategic plan] through multiple passes. Design the prompting pattern: the initial generation prompt, the self-critique prompt, the revision prompt, and the stopping criterion. Make it a repeatable cycle I can run until satisfied.

I want to give Claude feedback on its outputs and have it integrate that feedback effectively across a session. Write a feedback prompt template that: describes the problem with the last output specifically, explains what I wanted instead, and instructs Claude to apply this learning to all future outputs in the session.

I want to give Claude a knowledge base it uses for every response in a session. My knowledge base: [DESCRIBE WHAT YOU WANT CLAUDE TO KNOW: your product, your style guide, your use case context]. Write a system or context prompt that loads this knowledge efficiently without using excessive tokens, and that Claude will reliably reference rather than fall back on generic knowledge.