ChatGPT custom instructions let you tell ChatGPT who you are and how you want it to respond — once, for every conversation. Instead of repeating "I'm a senior developer, keep responses concise with code examples" in every chat, custom instructions do it automatically. This guide covers everything: setup, real-world examples, advanced tips, and how to combine custom instructions with ChatGPT memory for maximum personalization.
What Are ChatGPT Custom Instructions?
Custom instructions are two text fields that tell ChatGPT about you and your preferences. They were introduced in July 2023 and have become one of the most powerful — yet underused — ChatGPT features.
The Two Fields
Field 1: "What would you like ChatGPT to know about you?"
Your role, background, expertise level, projects, tech stack, goals. This gives ChatGPT context about who you are.
Field 2: "How would you like ChatGPT to respond?"
Tone, length, format, language, detail level, code style. This controls how ChatGPT communicates with you.
These instructions are injected into every new conversation as a system-level context. ChatGPT treats them as persistent preferences — not commands — so it will follow them while still being responsive to specific requests that may override them.
How to Set Up Custom Instructions (Step by Step)
Open Settings
Click your profile icon in the bottom-left corner of ChatGPT → Settings
Navigate to Personalization
In the left sidebar, click "Personalization" → "Custom Instructions"
Enable Custom Instructions
Toggle the switch to "On" to enable the feature
Fill in Field 1 — About You
Describe your role, expertise, current projects, and what context ChatGPT should know
Fill in Field 2 — Response Style
Specify tone, format, length, language preferences, and any rules for how ChatGPT should respond
Save and Test
Click Save, then start a new conversation to verify your instructions are working
Custom Instructions Examples by Role
Here are battle-tested custom instruction templates for different roles. Copy, modify, and use them as starting points.
For Software Engineers
Field 1 — About You:
I'm a senior full-stack developer with 8 years of experience. My primary stack is TypeScript, React, Next.js, Node.js, and PostgreSQL. I'm currently building a SaaS product for AI conversation management. I prefer production-ready code over toy examples. I'm familiar with design patterns, clean architecture, and distributed systems.
Field 2 — Response Style:
Always provide code examples with TypeScript types. Use modern syntax (ES2022+). Explain trade-offs when there are multiple approaches. Keep explanations concise — I don't need basics explained. Include error handling in code. When suggesting architectural decisions, consider scalability and maintainability. Format code with proper indentation and comments for non-obvious logic.
For Product Managers
Field 1 — About You:
I'm a product manager at a B2B SaaS startup. I manage a team of 5 engineers and 2 designers. My focus areas are user growth, retention, and monetization. I use data-driven decision making and need help with PRDs, user research synthesis, competitive analysis, and stakeholder communication.
Field 2 — Response Style:
Be structured and actionable. Use frameworks like RICE, ICE, or Jobs-to-be-Done when relevant. When I ask for analysis, include both qualitative insights and quantitative backing. Format outputs as ready-to-use documents (PRDs, briefs, strategy docs). Challenge my assumptions when you see blind spots. Keep executive summaries at the top.
For Data Scientists
Field 1 — About You:
I'm a data scientist with a PhD in statistics. I work with Python, pandas, scikit-learn, PyTorch, and SQL daily. My current projects involve NLP, recommendation systems, and time-series forecasting. I'm comfortable with advanced math and prefer rigorous statistical reasoning over heuristics.
Field 2 — Response Style:
Use Python for all code examples. Include type hints. When discussing statistical methods, cite relevant papers or established references. Prefer scikit-learn and PyTorch over TensorFlow. Always consider edge cases, data leakage, and evaluation methodology. When suggesting models, discuss assumptions and limitations. Include visualization code with matplotlib/seaborn when relevant.
For Content Writers & Marketers
Field 1 — About You:
I'm a content marketer at a B2B tech company. I write blog posts, case studies, email campaigns, and social media content. My audience is technical decision-makers (CTOs, VPs of Engineering). I focus on SEO-driven content strategy and need to balance technical depth with readability.
Field 2 — Response Style:
Write in a professional but conversational tone. Use short paragraphs (2-3 sentences max). Include data points and statistics when available. Suggest SEO-optimized headlines with target keywords. When writing blog posts, include a compelling hook in the first paragraph. Use bullet points and subheadings for scannability. Always include a clear CTA.
For Students & Researchers
Field 1 — About You:
I'm a graduate student in computer science, researching large language models and AI safety. I'm familiar with transformer architectures, RLHF, and alignment research. I'm writing my thesis and need help with literature reviews, paper writing, and experimental design.
Field 2 — Response Style:
Be precise and academic in tone. Cite specific papers with authors and years when referencing research. Use proper scientific notation for mathematical expressions. When discussing concepts, distinguish between established findings and open questions. Help me think through experimental design by pointing out potential confounds and controls. Format references in APA style.
Custom Instructions vs Memory: What's the Difference?
Many users confuse custom instructions with ChatGPT memory. They serve different purposes and work best together.
| Aspect | Custom Instructions | Memory |
|---|---|---|
| What it is | Explicit preferences you set manually | Facts ChatGPT learns automatically from conversations |
| Control | 100% user-controlled | Auto-generated (you can review and delete) |
| Content | Who you are + how to respond | Facts, preferences, project details learned over time |
| Updates | Manual — you edit it | Automatic — evolves with each conversation |
| Storage limit | ~1,500 characters per field | ~1,500 words total |
| Cross-platform | ChatGPT only | ChatGPT only (but AI Memory extends this) |
| Best for | Setting baseline behavior and tone | Accumulating project context and preferences |
💡 Pro Tip: Use Both Together
Custom instructions set your baseline — who you are and how you want responses. Memory builds on top of that with learned context. For maximum personalization, use custom instructions for stable preferences (role, tone, format) and let memory handle evolving context (current projects, recent decisions, specific preferences you've mentioned in conversations).
Custom Instructions Limitations
While powerful, custom instructions have important limitations you should know about:
- Character limit: Each field has a ~1,500 character limit. You can't write a novel — be concise and prioritize the most important context.
- ChatGPT only: Custom instructions don't transfer to Claude, Gemini, DeepSeek, or any other AI platform. If you use multiple AI tools, you need to set up instructions separately on each one.
- No versioning: There's no way to save multiple instruction profiles or revert to previous versions. If you change your instructions, the old version is gone.
- Not always followed: ChatGPT may sometimes ignore or partially follow custom instructions, especially in long conversations where the system prompt's influence diminishes.
- No programmatic access: You can only edit custom instructions through the ChatGPT UI. There's no API to read or update them automatically.
- Lost with new chats: While custom instructions apply to every new conversation, they don't carry over the specific context from previous conversations. That's what memory (and tools like AI Memory) are for.
Advanced Custom Instructions Tips
1. Use Structured Formatting
Instead of writing a paragraph, use structured bullet points. ChatGPT follows structured instructions more reliably:
Context:
- Role: Senior DevOps Engineer
- Stack: AWS, Terraform, Kubernetes, Python
- Team size: 12 engineers
Response rules:
- Always include IaC examples (Terraform preferred)
- Use production-grade patterns (not toy examples)
- Include cost estimates when suggesting AWS services
- Flag security implications in red
- Max response length: 500 words unless I ask for more
2. Define Persona Shortcuts
Tell ChatGPT to adopt specific personas when you use trigger words:
When I say "code review" — act as a strict senior engineer reviewing a PR. Focus on bugs, performance, security, and code style.
When I say "brainstorm" — be creative and suggest 5-10 ideas without judgment. Wild ideas welcome.
When I say "ELI5" — explain like I'm 5. Use analogies, no jargon.
3. Set Output Format Preferences
Specify default formats for common output types:
Output format preferences:
- Code: Always in TypeScript with types, include imports
- Lists: Use numbered lists for steps, bullets for features
- Comparisons: Use tables with clear headers
- Explanations: TL;DR at top, then details
- Emails: Professional tone, max 200 words, bullet points for action items
4. Include Negative Instructions
Tell ChatGPT what NOT to do — this is often more effective than positive instructions:
Don't:
- Start responses with "Sure!" or "Absolutely!"
- Apologize unnecessarily
- Add disclaimers about being an AI
- Use filler phrases like "It's worth noting that"
- Repeat my question back to me
5. Maintain a Custom Instructions Version History
Since ChatGPT doesn't version your instructions, keep a backup. You can use AI Memory to store your custom instruction templates and track how they evolve over time. This way, you can always revert to a previous version or adapt instructions for different contexts.
Combining Custom Instructions with AI Memory
Custom instructions and AI Memory create a powerful personalization stack:
🎯 Custom Instructions = Your Profile
Set once: who you are, your expertise, preferred response style, formatting rules. This is your static identity layer.
🧠 ChatGPT Memory = Learned Context
Auto-accumulated: project details, preferences you've mentioned, patterns in your requests. This is your dynamic context layer within ChatGPT.
🌐 AI Memory = Cross-Platform Knowledge
Unified search across ChatGPT, Claude, DeepSeek, Gemini. Never lose context when switching AI tools. Inject relevant memories into any conversation. This is your universal memory layer.
The key insight: custom instructions only work within ChatGPT. If you also use Claude for coding, DeepSeek for research, or Gemini for creative work, your personalization is fragmented across platforms. AI Memory bridges this gap by letting you search and inject context from any AI conversation into any other.
Try AI Memory — Free Chrome Extension
Upload your ChatGPT, Claude, or DeepSeek conversations. Search across all your AI chats instantly. Your data stays private — processed in your browser session.
Get the Extension →Frequently Asked Questions
Can I have different custom instructions for different conversations?
No. Custom instructions are global — the same instructions apply to every conversation. You can't set per-conversation or per-GPT custom instructions from your profile. However, GPT creators can define system instructions for their specific GPTs, which may override your personal custom instructions.
Do custom instructions use up my context window?
Yes. Custom instructions are injected into the system prompt, which consumes part of your context window. A typical custom instruction uses about 200-500 tokens. In most conversations, this is negligible. But for very long conversations approaching the context limit, it's worth keeping instructions concise.
Can I share custom instructions with my team?
Not natively in ChatGPT. Each team member sets their own custom instructions individually. For teams using ChatGPT Team or Enterprise, admins can create shared GPTs with predefined system instructions, but personal custom instructions remain per-user. Consider maintaining a shared document with recommended custom instruction templates for your team.
Why isn't ChatGPT following my custom instructions?
Common reasons: (1) Instructions are too vague — be specific. (2) Instructions conflict with the user's current request — explicit prompts take priority. (3) The conversation is very long — system prompt influence diminishes in long chats. (4) The model update may have changed instruction-following behavior. Try starting a new conversation to test.
Are custom instructions available on ChatGPT free plan?
Yes, custom instructions are available on all ChatGPT plans including Free, Plus, Team, and Enterprise. This is different from memory, which has more limited availability on the free tier. Check your Settings → Personalization menu to confirm access.