AI Prompts for ChatGPT Custom Instructions

Write a Custom Instructions context block for the first field based on this information. I am: [YOUR ROLE AND INDUSTRY]. I work on: [YOUR MAIN RESPONSIBILITIES]. My level of expertise in my field: [DESCRIBE]. Tools and technologies I use: [LIST]. My most common ChatGPT use cases: [LIST]. Make it concise (under 200 words) but specific enough that ChatGPT can infer my background in any response.

I want ChatGPT to calibrate explanations to my level in different domains. Write a Custom Instructions block that communicates: I am expert-level in [DOMAIN 1], intermediate in [DOMAIN 2], and a beginner in [DOMAIN 3]. For expert topics, skip basics. For beginner topics, explain from first principles. Format this as a clear instruction it will actually follow.

Add a section to my Custom Instructions that tells ChatGPT about my current main projects: [DESCRIBE 2-3 PROJECTS]. Include enough context that if I reference one of them in a future conversation without explaining it, ChatGPT will understand the context. Keep it under 100 words per project.

Write a Custom Instructions block that establishes my communication style for the first field. I prefer: [DESCRIBE YOUR PREFERENCES, e.g. direct over diplomatic, practical over theoretical, examples over explanations, short answers unless complexity requires length]. I dislike: [LIST]. Make these preferences specific enough that ChatGPT will deviate from them only when genuinely necessary.

Combine all of this into a single polished Custom Instructions block for field one. My role: [ROLE]. My industry: [INDUSTRY]. My expertise: [DESCRIBE]. My current focus: [DESCRIBE]. How I use ChatGPT most often: [LIST]. Limit to 300 words. Make it read as information rather than instructions.

Write response style instructions for field two that specify: (1) when to use bullet points versus prose, (2) maximum response length for different question types, (3) whether to use headers and when, (4) how to handle code formatting, (5) whether to include examples by default. My preference is [DESCRIBE YOUR PREFERENCES]. Make the instructions specific enough to eliminate hedging and padding.

Write a Custom Instructions block that explicitly bans the response patterns I find most annoying. I want ChatGPT to never: start responses with affirmations like "Great question!", add unnecessary preambles before answering, end every response with "I hope this helps!", add excessive caveats and qualifiers, or pad short answers with filler sentences. Add any others I should include: [YOUR ADDITIONS].

Write Custom Instructions for field two that set my default tone preferences. I want responses that are: [DESCRIBE TONE, e.g. direct and confident, conversational but precise, technical but readable]. When I ask for something creative, the default should be [DESCRIBE]. When I ask for something analytical, the default should be [DESCRIBE]. Include instructions for what to do when tone is ambiguous.

I am a developer and most of my ChatGPT sessions involve code. Write Custom Instructions for field two that set my coding defaults: always include type annotations, always use [LANGUAGE/FRAMEWORK] unless I specify otherwise, never use placeholder comments, include error handling by default, and follow [STYLE GUIDE OR CONVENTIONS]. Also specify how to handle code in non-code responses.

Write Custom Instructions that set my preferred default output format for different request types. For analysis: [FORMAT PREFERENCE]. For writing tasks: [FORMAT PREFERENCE]. For lists: [FORMAT PREFERENCE]. For explanations: [FORMAT PREFERENCE]. For plans: [FORMAT PREFERENCE]. Include an instruction to adapt when I explicitly request something different.

I use ChatGPT heavily for writing. Write Custom Instructions that apply whenever I ask for writing help: match my voice (I write in [DESCRIBE YOUR STYLE]), never add a conclusion unless I ask, never use [YOUR BANNED PHRASES], and always show me the word count. When editing, show changes inline rather than producing a clean rewrite. Calibrate length to [MY PREFERRED LENGTH].

When I ask research-type questions, I want ChatGPT to: (1) lead with the direct answer, not background; (2) flag uncertainty explicitly rather than presenting everything with equal confidence; (3) distinguish between widely accepted facts and contested claims; (4) never invent citations; (5) suggest what I should verify independently. Write this as a Custom Instructions block.

When I ask for code, ChatGPT should: write complete code, never use placeholder comments, include error handling unless I say not to, default to [MY PREFERRED LANGUAGE], add a one-line comment above any non-obvious logic, and explain key design decisions in a short paragraph after the code block. Write this as a Custom Instructions block I can paste directly.

When I ask for ideas or ask ChatGPT to brainstorm, I want it to: generate more options than I ask for (if I say 5, give 8), prioritize unusual ideas over obvious ones, avoid repeating ideas from earlier in the conversation, and flag the top three it thinks are strongest and why. Write this as a Custom Instructions block.

When I am making a decision and ask for help, ChatGPT should: identify the criteria I seem to be using, point out any criteria I appear to be ignoring, recommend a specific choice rather than just listing pros and cons, and flag if it thinks I am anchored on a weak option. Write this as a Custom Instructions block.

Here are my current Custom Instructions: [PASTE]. Ask me ten test questions that would reveal whether these instructions are working as intended. After I answer, run a few of the test prompts yourself and evaluate whether the responses match what the instructions specify.

Read these Custom Instructions: [PASTE]. Identify: (1) any instruction that is too vague to reliably influence behavior, (2) any contradiction between instructions, (3) any important behavior preference I appear to have omitted, and (4) any instruction that might cause problems in edge cases. Propose specific rewrites for the weakest instructions.

Use this prompt to test the impact of my Custom Instructions. Run it once in a fresh conversation without my Custom Instructions active, then once with them active. Prompt: [YOUR REPRESENTATIVE TEST PROMPT]. Compare the two outputs and tell me specifically which elements of my Custom Instructions are having the intended effect.

Here are my current Custom Instructions: [PASTE]. They are getting too long and may be causing the model to ignore parts of them. Edit them down to the 20% that has the most impact on response quality. Preserve the most important behavioral rules; cut anything generic or redundant.

My general Custom Instructions are: [PASTE]. I am starting a new intensive project: [DESCRIBE PROJECT]. What temporary modifications should I make to my Custom Instructions for the duration of this project? What should I add, remove, or change to optimize for this specific type of work?