How to Set Up Custom Instructions in ChatGPT
Summary
- Custom Instructions let you set persistent preferences (role, tone, format, constraints) so you do not repeat the same setup prompt in every chat.
- Start by defining what ChatGPT should know about you, then define exactly how you want answers structured (length, bullets, citations, code style, etc.).
- Keep instructions short, specific, and testable; avoid sensitive data and avoid conflicting rules that cause unpredictable outputs.
- Use role-based templates (consulting, marketing, recruiting, research, dev, support, ecommerce) and refine them with a quick test checklist.
- For reusable context across tools, you can store instruction templates and prompt snippets in CopyCharm and paste them where needed; ChatGPT can only retrieve supported synced CopyCharm data after authorization and sync.
Custom Instructions in ChatGPT are for the repeated “house rules” you want applied across conversations: who you are, what you do, what you prefer, and how you want responses formatted. If you find yourself retyping “Ask clarifying questions first,” “Use bullet points,” “Write in my brand voice,” or “Return JSON,” Custom Instructions are the fastest way to make that behavior persistent.
This guide shows how to set them up, what to put in each field, and how to maintain instruction sets for different roles and workflows (including when you also use Claude, Gemini, Cursor, snippet managers, clipboard managers, and reusable context packs).
What Custom Instructions do (and what they do not)
Custom Instructions do:
- Apply your preferences across new chats so you do not need to restate them each time.
- Standardize output format (headings, bullets, tables, code blocks, tone, reading level, etc.).
- Set default working assumptions (your audience, region, tools, constraints, and goals).
Custom Instructions do not:
- Guarantee perfect compliance. They influence responses, but you still need to review and correct outputs.
- Replace task-specific prompts. You still need to provide the actual assignment, inputs, and constraints for each request.
- Automatically carry over private data safely. Treat instructions as a place for preferences and non-sensitive context, not secrets.
How to set up Custom Instructions in ChatGPT (step-by-step)
The exact menu labels can change over time, but the setup flow stays consistent: open ChatGPT settings, find Custom Instructions, fill in the two instruction areas, and save.
Step 1: Open Settings
- In ChatGPT, open your account/profile menu.
- Go to Settings (or similar).
Step 2: Find “Custom Instructions”
- Look for a section labeled Custom Instructions (sometimes grouped under personalization-related settings).
Step 3: Fill in the two key parts
ChatGPT commonly separates Custom Instructions into two areas:
- What ChatGPT should know about you (your context)
- How you want ChatGPT to respond (your output rules)
Step 4: Save and test
After saving, start a new chat and run a quick test prompt (examples below). If the output is not following your rules, tighten the wording, remove conflicts, and add a single “priority rule” (for example: “If instructions conflict, ask a clarifying question.”).
What to write in Custom Instructions (with a practical structure)
A useful Custom Instructions setup is short, specific, and measurable. Use this structure and adapt it to your role.
Part A: “What ChatGPT should know about me” (context)
- Role and domain: “I am a B2B SaaS marketing manager focused on pipeline.”
- Audience: “Write for busy executives and practitioners.”
- Constraints: “Assume I cannot share proprietary data; propose options that work with limited inputs.”
- Environment: “I work in English; prefer US spelling; I use Google Sheets and Notion.”
- Decision style: “I prefer tradeoffs and a recommendation with risks.”
Part B: “How I want you to respond” (output rules)
- Default format: “Start with a 3-bullet summary, then steps, then a checklist.”
- Clarifying questions: “Ask up to 3 questions if requirements are unclear.”
- Tone: “Direct, practical, no hype.”
- Length: “Keep answers under 400 words unless I ask for depth.”
- Quality controls: “Call out assumptions; provide 2 alternatives when relevant.”
- For technical work: “When writing code, include comments and edge cases; specify inputs/outputs.”
A table of Custom Instructions patterns (copy/paste templates)
| Role | What ChatGPT should know about you (example) | How you want ChatGPT to respond (example) | Quick test prompt |
|---|---|---|---|
| Consultant | I advise mid-market teams on strategy and operations. I need structured thinking and clear next steps. | Use: (1) Summary, (2) Diagnosis, (3) Options with tradeoffs, (4) Recommendation, (5) 7-day action plan. Ask clarifying questions if scope is unclear. | “Help me plan a 2-week discovery for a client with declining retention.” |
| Marketer | I write B2B content for demand gen. Audience is time-poor operators. | Write in plain English. Provide 5 headline options, then an outline, then a draft. Avoid buzzwords. Include a CTA suggestion. | “Draft a landing page for a webinar on onboarding.” |
| Recruiter | I recruit for technical and non-technical roles. I need inclusive, specific job language. | Return: role summary, must-haves vs nice-to-haves, interview plan, and outreach message variants. Flag biased or vague requirements. | “Rewrite this job description to be clearer and more inclusive: [paste].” |
| Researcher / Analyst | I synthesize information into decision memos. I care about assumptions and uncertainty. | Separate: facts I provided vs your inferences. Provide a confidence note and what data would change the conclusion. | “Turn these notes into a 1-page decision memo: [paste].” |
| Developer | I build web apps and scripts. I prefer minimal, correct examples. | When coding: specify language/version assumptions, show a minimal working example, list edge cases, and include a short test plan. | “Write a function that normalizes phone numbers to E.164.” |
| Support team | I answer customer tickets. I need empathy and accuracy. | Use: acknowledge, diagnose, steps, confirmation question, and escalation criteria. Keep it under 150 words unless complex. | “Reply to this ticket: ‘My invoice looks wrong…’” |
| Ecommerce operator | I manage product listings and customer messaging. I need conversion-focused copy. | Provide: benefits-first bullets, objections, sizing/fit guidance, and a short FAQ. Avoid unverifiable claims. | “Rewrite this product description for clarity and conversion: [paste].” |
Custom Instructions vs Memory vs Projects (how to choose the right lever)
ChatGPT has multiple ways to keep context. The names and availability can change, but the practical distinction is:
- Custom Instructions: your persistent preferences and default response format.
- Memory: longer-lived personal/work preferences the system may remember (where available and enabled). Use it for stable preferences, not per-project details.
- Projects (or project-like workspaces): a way to group work and keep project-specific context together. Use it for a client, product, or research stream where the “background pack” matters.
If you are unsure: put formatting and behavior rules in Custom Instructions, keep project-specific details in the project workspace (or a reusable brief you paste), and keep sensitive details out of both.
Common mistakes (and fixes)
1) Overstuffing instructions
Symptom: outputs become inconsistent or ignore parts of your rules.
Fix: keep a short “always-on” core (5-10 lines). Move the rest into reusable snippets you paste only when needed.
2) Conflicting rules
Symptom: “Be concise” plus “be comprehensive” plus “include everything.”
Fix: add a priority rule: “If there is a conflict, prioritize correctness and ask a clarifying question.”
3) Vague preferences
Symptom: “Write better” or “make it professional.”
Fix: specify observable outputs: “Use short paragraphs, bullets, and a final checklist.”
4) Putting sensitive data into instructions
Symptom: you paste client secrets, credentials, or private identifiers.
Fix: store sensitive details in your own secure systems and only share what is necessary for the task.
Building repeatable workflows across ChatGPT, Claude, Gemini, and Cursor
Many knowledge workers use more than one AI tool. Custom Instructions are specific to ChatGPT, so you may want a parallel “instruction pack” you can reuse elsewhere (Claude, Gemini, Cursor, email, docs) by copy/pasting.
A practical approach is to maintain:
- A short ChatGPT Custom Instructions core (always on)
- Role templates (consulting vs recruiting vs dev)
- Project briefs (client background, product context, brand voice)
- Output schemas (tables, JSON shapes, ticket reply format)
How CopyCharm fits: saving, finding, and reusing instruction templates
If you iterate on Custom Instructions, you will likely create multiple versions (a “short core,” a “marketing mode,” a “dev mode,” and a “support mode”). CopyCharm can help you keep those reusable blocks close to your clipboard workflow.
Concrete workflow (save → find → reuse):
- Save: When you draft a good Custom Instructions set (or a role template), copy it and save it in CopyCharm as a Saved Prompt. When you copy useful reference text (a brand voice paragraph, a response rubric, a JSON schema), you can also keep it as a clip and Favorite the important ones.
- Find: Later, search in CopyCharm for “custom instructions,” “support tone,” “json schema,” or a client name to quickly locate the exact block you used before.
- Reuse: Paste the block into ChatGPT Custom Instructions, a new chat, or another tool (Claude, Gemini, Cursor, docs, email) via manual copy/paste.
If you want ChatGPT to retrieve your saved templates: CopyCharm also offers an authenticated ChatGPT connector backed by optional AI Access sync. After you sign in with an eligible active CopyCharm purchase, authorize the CopyCharm Desktop connection, enable and complete AI Access sync, and authorize the ChatGPT connector, ChatGPT can search/list and retrieve only the supported Synced Data you enabled (Favorite Clips, Saved Prompts, and optional Other Clips within your selected time range). ChatGPT cannot access unsynced local CopyCharm data.
Try CopyCharm for saving and reusing your Custom Instructions templates
Practical test checklist (use this after every change)
- Format test: “Explain X in 5 bullets and a checklist.” Did it follow your default structure?
- Clarifying test: Give a vague prompt. Did it ask the number of questions you specified?
- Tone test: “Write a customer reply.” Does it match your tone constraints?
- Constraint test: “Propose a plan with limited data.” Did it state assumptions and options?
- Edge-case test (dev): “Write code for Y.” Did it include edge cases and a test plan?
Frequently Asked Questions
FAQ 1: Where do I find Custom Instructions in ChatGPT settings?
Answer: Open your profile/account menu in ChatGPT, go to Settings, and look for a Custom Instructions section (sometimes grouped under personalization-related settings). The exact labels can change, but the path is consistently “account menu → settings → custom instructions.”
Takeaway: If you can reach Settings, you can usually find Custom Instructions under personalization-style options.
FAQ 2: What should I put in “What ChatGPT should know about me”?
Answer: Add stable, non-sensitive context that changes how answers should be framed: your role, audience, region/language preferences, tools you use, and constraints (time, data access, compliance boundaries). Avoid secrets, credentials, or anything you would not want repeated in future chats.
Takeaway: Use this field for durable context that improves relevance, not for private data.
FAQ 3: What should I put in “How I want ChatGPT to respond”?
Answer: Specify observable output rules: default structure (summary/steps/checklist), tone, length, whether to ask clarifying questions, and any formatting requirements (tables, bullet points, JSON, code style). If you want consistent deliverables, define a template the model can follow every time.
Takeaway: Write response rules as a format spec, not as vague preferences.
FAQ 4: How long should Custom Instructions be?
Answer: Keep them as short as possible while still being specific. A compact “always-on” core (role, audience, and a response template) is easier to follow than a long list of rules. Put specialized instructions into snippets you paste only when needed.
Takeaway: Short, testable instructions are easier to maintain and refine.
FAQ 5: Why is ChatGPT ignoring my Custom Instructions?
Answer: Common causes include conflicting rules (concise vs exhaustive), vague wording (“be better”), or task prompts that override your defaults. Fix it by removing conflicts, rewriting rules as concrete outputs, and adding a priority line like “If requirements are unclear, ask up to 3 clarifying questions before drafting.” Then run a short test prompt to confirm behavior.
Takeaway: When compliance slips, simplify and make the rules measurable.
FAQ 6: Should I use Custom Instructions, Memory, or Projects for repeatable work?
Answer: Use Custom Instructions for default behavior and formatting. Use Memory (where available and enabled) for stable preferences you want remembered longer-term. Use Projects (or project-like workspaces) for client- or initiative-specific context that should stay grouped with that work. If you need different “modes,” keep Custom Instructions minimal and apply project briefs per workspace or per chat.
Takeaway: Put “how to respond” in Custom Instructions and keep project details in project context.
FAQ 7: Can I maintain multiple instruction sets for different roles (marketing vs recruiting vs dev)?
Answer: Yes. Keep one short baseline in Custom Instructions, then maintain role-specific templates you can swap in when needed (for example, a “support reply format” vs a “developer code spec”). Test each template with a standard prompt so you can quickly see whether the output matches your expectations.
Takeaway: Treat instruction sets like reusable templates and validate them with a repeatable test prompt.
FAQ 8: Can CopyCharm help me reuse Custom Instructions across chats and tools?
Answer: CopyCharm is a Windows desktop app that saves copied text locally so you can search past clips, favorite important clips, and separately save reusable prompts. You can store multiple Custom Instructions templates as Saved Prompts, search them later, and copy/paste them into ChatGPT, Claude, Gemini, Cursor, docs, or email. If you enable AI Access sync and authorize the authenticated ChatGPT connector, ChatGPT can search and retrieve only supported synced CopyCharm data; it cannot access unsynced local CopyCharm data.
Takeaway: Use CopyCharm to keep instruction templates easy to find and reuse, with optional connector-based retrieval for supported synced data in ChatGPT.