Custom Instructions: Set Up ChatGPT, Claude & Gemini to Work Your Way

Master ChatGPT custom instructions, Claude Projects, and Gemini Gems. Learn the RODES framework and get 10 ready-to-use templates for developers, writers, and marketers.

Table of Contents

Ever feel like you’re repeating yourself every time you start a new AI chat? “I’m a developer working with Python…” “Write in a friendly tone…” “Keep responses under 500 words…”

You’re not alone. Most people waste 5-10 minutes per session re-explaining their preferences to ChatGPT, Claude, or Gemini. But here’s the thing: you can set up your AI assistant once and have it remember your preferences forever.

This guide shows you exactly how to use custom instructions (ChatGPT), Projects (Claude), and Gems (Gemini) to make AI work the way you want—automatically. No more repetitive prompts. No more inconsistent responses. Just AI that gets you.

By the end of this post, you’ll have:

  • Custom instructions set up on all three major AI platforms
  • 10 profession-specific templates you can copy and use immediately
  • The RODES framework for writing bulletproof instructions
  • Common mistakes to avoid (and how to fix them)

Let’s dive in.

What Are Custom Instructions, Projects, and Gems?

Think of these features as persistent memory for your AI assistant. Instead of explaining your preferences every single time you chat, you set them once and the AI remembers.

Here’s how each platform handles it:

ChatGPT: Custom Instructions

Custom instructions appear in every new chat you start with ChatGPT. They’re split into two sections:

  1. What would you like ChatGPT to know about you? — Your role, preferences, context
  2. How would you like ChatGPT to respond? — Tone, format, style preferences

Claude: Projects

Projects are workspaces where you can:

  • Set custom instructions that apply to all chats in that Project
  • Upload reference documents (up to 200,000 tokens worth)
  • Organize related conversations by topic

Projects are perfect for different work contexts (e.g., “Web Development,” “Content Writing,” “Data Analysis”).

Gemini: Gems

Gems are pre-configured AI assistants you create for specific tasks. Each Gem has:

  • A name and description
  • Custom instructions that define its behavior
  • Optional starter prompts

Gems live in your sidebar and can be activated with one click.

Bottom line: All three do the same thing—they remember your preferences so you don’t have to repeat yourself. The implementation differs, but the outcome is the same: faster, more consistent AI interactions.

ChatGPT Custom Instructions: Step-by-Step Setup

How to Access Custom Instructions

For ChatGPT Plus/Pro users:

  1. Click your profile icon (bottom-left corner)
  2. Select Settings
  3. Click Personalization in the sidebar
  4. Toggle Custom instructions ON
  5. Click Custom instructions to edit

For ChatGPT Free users: Custom instructions are available! Follow the same steps above.

The Two Input Fields Explained

ChatGPT splits custom instructions into two sections. Here’s what to put in each:

Section 1: “What would you like ChatGPT to know about you?”

This is your context section. Include:

  • Your profession or role
  • Your experience level
  • Your goals or what you’re working on
  • Any relevant background information

Example for a developer:

I'm a senior full-stack developer with 8 years of experience. I primarily work with TypeScript, React, Node.js, and PostgreSQL. I'm building SaaS products and care deeply about clean code, performance, and user experience.

I prefer modern best practices, functional programming patterns, and pragmatic solutions over over-engineering.

Section 2: “How would you like ChatGPT to respond?”

This is your style section. Include:

  • Preferred tone (formal, casual, technical)
  • Response length preferences
  • Format preferences (bullet points, code examples, explanations)
  • What to avoid

Example for a developer:

- Be direct and technical. Skip unnecessary explanations.
- Provide code examples when relevant, with inline comments.
- Explain trade-offs when there are multiple approaches.
- Use TypeScript for all code examples unless I specify otherwise.
- Keep responses under 500 words unless I ask for more detail.
- If you're unsure, ask clarifying questions instead of guessing.

Saving and Testing

  1. Click Save after filling both sections
  2. Start a new chat (custom instructions don’t apply to existing chats)
  3. Test with a simple prompt: “Help me optimize a React component”
  4. Check if ChatGPT responds according to your instructions

Pro tip: If responses aren’t quite right, refine your instructions. It usually takes 2-3 iterations to dial in the perfect setup.

Claude Projects: Step-by-Step Setup

How to Create a Project

For Claude Pro users:

  1. Click Projects in the left sidebar
  2. Click + New Project
  3. Give your Project a name (e.g., “Web Development”)
  4. Add a description (optional but recommended)
  5. Click Create Project

For Claude Free users: Projects are a Pro feature. Consider upgrading if you need this functionality, or use the main chat and paste instructions manually each time.

Adding Custom Instructions

  1. Open your Project
  2. Click Customize Project (gear icon in top-right)
  3. In the Project Instructions field, write your custom instructions
  4. Click Save

Example Project setup for a content writer:

Project Name: Content Writing

Project Instructions:

You are my content writing assistant. I write SEO blog posts for B2B SaaS companies.

Writing style:
- Conversational but professional
- Use "you" to address the reader
- Short paragraphs (2-3 sentences max)
- Concrete examples over abstract theory
- Avoid jargon unless it's industry-standard

Content structure:
- Always include an engaging hook in the first paragraph
- Use H2 and H3 headers to break up content
- Include actionable takeaways
- End with a clear next step for the reader

SEO requirements:
- Naturally integrate target keywords (I'll provide them)
- Write meta descriptions under 160 characters
- Suggest internal linking opportunities

Uploading Reference Documents

Projects let you upload files that Claude can reference in every chat:

  1. Click Add content in your Project
  2. Upload PDFs, text files, or paste text directly
  3. Claude will automatically reference these in responses

Use cases:

  • Brand style guides
  • Product documentation
  • Code snippets or templates
  • Research papers or notes

Limits:

  • Up to 200,000 tokens per Project (roughly 150,000 words)
  • Files are processed and chunked for optimal retrieval

Creating Multiple Projects

You’re not limited to one Project. Create separate Projects for different contexts:

  • Web Development — Custom instructions for coding, reference docs with API documentation
  • Content Marketing — Writing guidelines, brand voice, SEO requirements
  • Data Analysis — Analysis frameworks, datasets, visualization preferences

Each Project maintains its own custom instructions and uploaded files.

Gemini Gems: Step-by-Step Setup

How to Create a Gem

For Gemini Advanced users:

  1. Click the Gem manager icon (top-right, looks like a gem/diamond)
  2. Click + Create Gem
  3. Fill in the Gem details (see below)
  4. Click Save

For Gemini Free users: Gems are an Advanced feature. You’ll need to upgrade or manually paste instructions each time.

Gem Configuration Fields

When creating a Gem, you’ll fill in:

1. Gem Name

A short, descriptive name (e.g., “Python Debugger,” “Blog Editor,” “Email Assistant”)

2. Description

What this Gem does. This helps you remember when to use it.

Example: “Helps me debug Python code by analyzing errors and suggesting fixes.”

3. Instructions

The core custom instructions for this Gem. Be specific about:

  • The Gem’s role
  • How it should respond
  • What it should prioritize

Example Gem for a marketer:

Gem Name: SEO Content Optimizer

Description: Reviews blog posts and suggests SEO improvements

Instructions:

You are an SEO content optimizer specializing in B2B SaaS.

When I share a blog post draft, analyze it for:
1. Target keyword usage (frequency, placement, natural integration)
2. Readability (sentence length, paragraph structure, Flesch score)
3. Engagement (hooks, storytelling, examples)
4. Structure (H2/H3 usage, table of contents, scanability)
5. Meta elements (title tag, meta description, URL slug)

Provide:
- Specific, actionable suggestions
- Before/after examples for rewrites
- Prioritized list (high-impact changes first)

Be direct and constructive. Assume I understand SEO basics.

Using Gems

Once created, Gems appear in your sidebar. To use one:

  1. Click the Gem from the list
  2. Start chatting—the Gem’s instructions are automatically active
  3. Switch Gems anytime by clicking a different one

Pro tip: Create Gems for recurring tasks you do weekly. Examples:

  • Code reviewer
  • Email writer
  • Meeting summarizer
  • Research assistant

The RODES Framework: Writing Bulletproof Custom Instructions

Here’s the problem with most custom instructions: they’re too vague (“be helpful”) or too long (3 pages of rules the AI can’t follow).

The RODES framework solves this. It’s a 5-part structure for writing custom instructions that actually work.

R — Role

Define who the AI should be.

Vague: “You’re a helpful assistant.” Better: “You’re a senior data analyst with expertise in Python, pandas, and data visualization.”

The more specific the role, the better the AI can respond. Include:

  • Job title or expertise area
  • Years of experience (if relevant)
  • Specializations

O — Objective

State what you want the AI to help you achieve.

Vague: “Help me with coding.” Better: “Help me build scalable backend APIs using Node.js and PostgreSQL, focusing on performance and maintainability.”

Your objective should answer: “What am I using this AI to accomplish?”

D — Details

Add constraints, preferences, and guardrails.

This is where you specify:

  • Tone (formal, casual, technical)
  • Response length (concise vs. detailed)
  • Format (bullet points, code examples, step-by-step)
  • What to avoid

Example:

Details:
- Use casual, conversational tone
- Keep responses under 300 words unless I ask for more
- Provide code examples with inline comments
- Avoid suggesting deprecated libraries
- If multiple solutions exist, explain trade-offs

E — Examples

Show the AI what good output looks like.

This is optional but powerful. Provide 1-2 examples of ideal responses.

Example:

When I ask "How do I handle authentication?", respond like this:

"For authentication, I recommend using JWT tokens with httpOnly cookies. Here's why:

Pros:
- Secure (httpOnly prevents XSS attacks)
- Stateless (no server-side session storage)
- Scalable (works across multiple servers)

Cons:
- Tokens can't be invalidated (use short expiry times)
- Slightly more complex than sessions

Here's a basic implementation:
[code example]

Trade-off: If you need instant logout across devices, consider using a token blacklist or switch to server-side sessions."

S — Sense Check

Add a final instruction to verify the AI understands.

Example:

If you're unsure about my question or need more context, ask me clarifying questions instead of guessing. Always prioritize accuracy over speed.

This prevents the AI from hallucinating or providing unhelpful responses.

RODES Example: Full Custom Instruction

Here’s a complete custom instruction using RODES:

Role: You are a senior frontend developer specializing in React, TypeScript, and modern CSS.

Objective: Help me build high-performance, accessible web applications with clean, maintainable code.

Details:
- Respond in a direct, technical tone
- Provide TypeScript code examples (not JavaScript)
- Include inline comments explaining complex logic
- Suggest performance optimizations when relevant
- Keep responses under 400 words unless I request more detail
- Use modern React patterns (hooks, functional components, no class components)

Examples:
When I ask "How should I handle form state?", respond with a comparison of approaches (useState, useReducer, React Hook Form), explain trade-offs, and provide a code example of your recommended approach.

Sense Check: If my question is ambiguous, ask clarifying questions before answering.

10 Custom Instruction Templates by Profession

Copy and adapt these templates for your needs. Each follows the RODES framework.

1. Software Developer

Role: You are a senior software engineer with expertise in [YOUR LANGUAGES/FRAMEWORKS].

Objective: Help me write clean, performant, maintainable code and solve technical problems efficiently.

Details:
- Provide code examples in [LANGUAGE] with inline comments
- Explain trade-offs when multiple solutions exist
- Prioritize modern best practices and design patterns
- Keep responses under 500 words unless I ask for more detail
- Include error handling and edge cases in code examples

Sense Check: If you're unsure about my tech stack or requirements, ask clarifying questions.

2. Content Writer

Role: You are an experienced content writer specializing in [YOUR NICHE, e.g., B2B SaaS, health, finance].

Objective: Help me create engaging, SEO-optimized content that resonates with my target audience.

Details:
- Write in a conversational, friendly tone
- Use short paragraphs (2-3 sentences max)
- Include concrete examples and actionable advice
- Naturally integrate target keywords (I'll provide them)
- Avoid jargon unless industry-standard
- Suggest headline options when writing blog posts

Sense Check: If you need more context about the target audience or topic, ask before writing.

3. Digital Marketer

Role: You are a digital marketing strategist with expertise in SEO, content marketing, and conversion optimization.

Objective: Help me create and optimize marketing campaigns that drive traffic and conversions.

Details:
- Focus on data-driven, actionable recommendations
- Prioritize high-impact tactics over theoretical concepts
- Include specific examples and case studies when possible
- Keep responses under 400 words unless I request detailed analysis
- When suggesting strategies, include implementation steps

Sense Check: If my question lacks context (target audience, budget, goals), ask for clarification.

4. Data Analyst

Role: You are a senior data analyst with expertise in Python, SQL, and data visualization.

Objective: Help me analyze datasets, write queries, and create visualizations that tell compelling stories with data.

Details:
- Provide Python code using pandas, numpy, and matplotlib
- Include comments explaining your analysis approach
- Suggest data cleaning steps when relevant
- Explain statistical concepts in plain language
- Recommend appropriate visualizations for the data
- Keep responses under 500 words unless I ask for detailed analysis

Sense Check: If the dataset or analysis goal is unclear, ask questions before providing code.

5. Product Manager

Role: You are an experienced product manager who understands user needs, technical constraints, and business goals.

Objective: Help me define product requirements, prioritize features, and make strategic product decisions.

Details:
- Use frameworks like RICE, MoSCoW, or Jobs-to-be-Done when relevant
- Provide structured outputs (user stories, acceptance criteria, PRDs)
- Consider both user value and technical feasibility
- Ask clarifying questions about constraints and goals
- Keep responses under 400 words unless I request detailed documentation

Sense Check: If you need more context about the product, users, or business goals, ask first.

6. UX/UI Designer

Role: You are a senior UX/UI designer with expertise in user research, interaction design, and visual design.

Objective: Help me create user-centered designs that are both functional and beautiful.

Details:
- Reference established design principles (Nielsen heuristics, Gestalt, etc.)
- Suggest specific UI patterns and components
- Consider accessibility (WCAG standards)
- Provide rationale for design decisions
- Keep responses under 400 words unless I request detailed specifications
- When suggesting layouts, describe them clearly or provide ASCII mockups

Sense Check: If the design context or user needs are unclear, ask for more information.

7. Business Consultant

Role: You are a business consultant with expertise in strategy, operations, and growth.

Objective: Help me solve business problems, make strategic decisions, and identify growth opportunities.

Details:
- Use frameworks like SWOT, Porter's Five Forces, BCG Matrix when relevant
- Provide structured, actionable recommendations
- Consider both short-term tactics and long-term strategy
- Include implementation considerations
- Keep responses under 500 words unless I request detailed analysis
- Be direct and candid about risks and trade-offs

Sense Check: If the business context, industry, or goals are unclear, ask before providing recommendations.

8. Academic Researcher

Role: You are an academic research assistant with expertise in [YOUR FIELD, e.g., psychology, computer science, economics].

Objective: Help me conduct literature reviews, design studies, analyze research, and write academic papers.

Details:
- Cite relevant research when possible (even if you can't access full papers)
- Use formal, academic tone
- Explain methodologies and statistical concepts clearly
- Suggest research questions and study designs when relevant
- Keep responses under 600 words unless I request detailed analysis
- Follow APA/MLA/Chicago citation style (I'll specify which)

Sense Check: If the research question or methodology is unclear, ask for clarification.

9. Customer Support Specialist

Role: You are a customer support specialist who helps resolve customer issues with empathy and efficiency.

Objective: Help me draft customer support responses that are helpful, empathetic, and on-brand.

Details:
- Use a friendly, professional tone
- Acknowledge the customer's frustration or concern
- Provide clear, step-by-step solutions
- Offer alternatives when the primary solution isn't possible
- Keep responses under 300 words (customers prefer concise answers)
- Avoid jargon; use plain language

Sense Check: If you need more context about the product, issue, or company policy, ask before drafting.

10. Educator/Teacher

Role: You are an experienced educator specializing in [YOUR SUBJECT, e.g., mathematics, history, science].

Objective: Help me create lesson plans, explain concepts clearly, and engage students effectively.

Details:
- Explain concepts using analogies and real-world examples
- Adapt explanations to the student's age/level (I'll specify)
- Suggest activities, exercises, or discussion questions
- Use encouraging, supportive tone
- Keep responses under 500 words unless I request detailed lesson plans
- Consider different learning styles (visual, auditory, kinesthetic)

Sense Check: If the student's level or learning context is unclear, ask for more information.

Common Mistakes (And How to Avoid Them)

Even with templates, people make these mistakes when writing custom instructions. Here’s how to avoid them:

Mistake 1: Being Too Vague

Bad: “Be helpful and friendly.”

Why it fails: Every AI is already trying to be helpful and friendly. This adds no value.

Fix: Be specific. What does “helpful” mean for you?

Good: “Provide step-by-step instructions with examples. If there are multiple approaches, explain trade-offs so I can choose.”

Mistake 2: Writing Too Much

Bad: 1,500 words of instructions covering every possible scenario.

Why it fails: AI models have limited context windows. Long instructions crowd out the actual conversation.

Fix: Keep custom instructions under 500 words. Focus on the 20% of preferences that cover 80% of your use cases.

Rule of thumb: If you can’t read your custom instructions in 90 seconds, they’re too long.

Mistake 3: Conflicting Instructions

Bad:

- Be concise and keep responses under 200 words
- Provide detailed explanations with examples
- Include code comments for every line

Why it fails: You can’t be both concise AND detailed with extensive code comments.

Fix: Decide what matters most and remove contradictions.

Good:

- Keep explanations under 300 words
- Provide code examples with comments only for complex logic

Mistake 4: Treating AI Like a Human Employee

Bad: “Remember that I prefer TypeScript” or “Don’t forget to include examples.”

Why it fails: AI doesn’t “remember” or “forget.” It follows instructions consistently—or doesn’t.

Fix: Write instructions as rules, not reminders.

Good: “Always use TypeScript for code examples” or “Include at least one example in every response.”

Mistake 5: No Testing or Iteration

Bad: Setting custom instructions once and never refining them.

Why it fails: You won’t nail it on the first try. Custom instructions need refinement.

Fix: Test your instructions with 5-10 real prompts. Note what works and what doesn’t. Refine weekly until you’re happy.

Mistake 6: Using the Same Instructions Everywhere

Bad: One generic custom instruction for all tasks (coding, writing, research).

Why it fails: Different tasks need different instructions. A code review needs different tone/format than a blog post draft.

Fix:

  • ChatGPT: Switch custom instructions based on task, or use GPTs for specific use cases
  • Claude: Create separate Projects for each major task area
  • Gemini: Create task-specific Gems

Mistake 7: Not Specifying “What to Avoid”

Bad: Only listing what you want, not what you DON’T want.

Why it fails: AI might still include things you hate (jargon, long-windedness, deprecated code).

Fix: Add an “Avoid” section:

Good:

Avoid:
- Jargon or buzzwords without explanation
- Responses longer than 500 words (unless I ask)
- Suggesting outdated libraries or deprecated methods
- Vague advice like "it depends" without explaining trade-offs

When to Use Custom Instructions vs. Regular Prompting

Custom instructions aren’t always the answer. Here’s when to use them (and when not to):

Use Custom Instructions When:

You have consistent preferences across many conversations Example: You always want code in TypeScript, always prefer concise responses, always need examples.

You’re working in a specific role or context repeatedly Example: You’re a developer working on a React project for the next 3 months.

You want to save time on repetitive setup Example: You’re tired of explaining “I’m a marketer focused on B2B SaaS” in every chat.

You need consistent tone/format across responses Example: You’re generating content for a brand with strict voice guidelines.

Use Regular Prompting When:

Your needs change frequently Example: One day you’re coding in Python, the next you’re writing a blog post, the next you’re analyzing data.

You’re exploring or experimenting Example: You’re trying different approaches and don’t want constraints.

You need the AI to role-play different personas Example: You’re doing creative writing and want the AI to switch between characters.

The task is one-off or highly specific Example: You need help with a specific tax question. No need to set up custom instructions for that.

Hybrid Approach: Best of Both Worlds

Here’s what works best for most people:

  1. Set baseline custom instructions (or create 2-3 Projects/Gems) for your most common tasks
  2. Use regular prompts to override or add context for specific conversations

Example:

Custom instruction (baseline): “I’m a developer. Use TypeScript for code examples.”

Regular prompt (specific task): “Ignore TypeScript for now—show me this in Python instead.”

The AI will prioritize the specific prompt over the custom instruction when there’s a conflict.

Advanced Tips for Power Users

Once you’ve mastered the basics, try these advanced techniques:

1. Chain Custom Instructions with Prompt Templates

Combine custom instructions with reusable prompt templates for maximum efficiency.

Custom instruction:

You are my code review assistant. Focus on performance, readability, and best practices.

Reusable prompt template:

Review this [LANGUAGE] code for:
1. Performance bottlenecks
2. Code readability
3. Best practices violations
4. Security concerns

[PASTE CODE]

Save the prompt template in a text file and reuse it with different code snippets.

2. Version Control Your Custom Instructions

Custom instructions evolve. Keep track of changes:

  1. Copy your custom instructions to a text file
  2. Date each version (e.g., “custom-instructions-v3-2025-01-08.txt”)
  3. Note what changed and why

This helps you roll back if a change makes things worse.

3. Create Role-Specific Projects/Gems

Instead of one generic setup, create multiple task-specific configurations:

Example for a full-stack developer:

  • Project 1: Frontend Development (React, TypeScript, CSS)
  • Project 2: Backend Development (Node.js, PostgreSQL, APIs)
  • Project 3: DevOps (Docker, CI/CD, deployment)

Each Project has tailored instructions and reference docs.

4. Use Custom Instructions to Enforce Formats

If you always need responses in a specific format, encode it in your instructions:

When I ask for code reviews, always respond in this format:

Summary: [One-sentence overview]

Issues Found:
1. [Issue] - Severity: [High/Medium/Low]
   - Why it matters: [Explanation]
   - Suggested fix: [Code example]

Positive Notes: [What's done well]

The AI will follow this structure consistently.

5. Combine Instructions with Uploaded Files (Claude Projects)

For Claude users, this is powerful:

  1. Upload your brand style guide, API docs, or code standards
  2. Set custom instructions that reference these files
  3. Claude will automatically follow guidelines from both

Example:

Custom instruction:

When writing blog posts, follow the style guide I've uploaded. Match the tone, structure, and formatting exactly.

Claude will reference the uploaded style guide in every response.

Conclusion: Set It Once, Benefit Forever

Here’s the reality: most people will read this guide, think “that’s useful,” and never actually set up custom instructions.

Don’t be most people.

The 15 minutes you invest today will save you hours over the next few months. Every time you start a new ChatGPT conversation, every time you open Claude, every time you use Gemini—your preferences are already there, working for you.

Here’s what to do right now:

  1. Pick one AI platform you use most (ChatGPT, Claude, or Gemini)
  2. Choose one template from the “10 Custom Instruction Templates” section above
  3. Customize it for your specific needs (5 minutes)
  4. Test it with 3-5 real prompts you’d actually use
  5. Refine based on what works and what doesn’t

That’s it. You’ll have a personalized AI assistant that actually understands what you need.

Then, once you’ve nailed the first platform, repeat for the others. Within an hour, you’ll have all three major AI platforms set up to work exactly the way you want.

One final tip: Don’t aim for perfection on day one. Start with a simple custom instruction, use it for a week, and refine based on real usage. The best custom instructions are iterative—they evolve as you discover what works.

Now stop reading and go set up your custom instructions. Your future self will thank you.


Want more AI productivity guides? Check out our AI Skills Directory for hundreds of ready-to-use prompts and skills for ChatGPT, Claude, Gemini, and more.