ChatGPT Custom Instructions: How to Write, Save, and Reuse Them

Summary

• ChatGPT Custom Instructions are reusable preferences set in Personalization that apply across future chats and can be edited or deleted for future conversations.
• Use Custom Instructions for stable defaults (tone, formatting, how you want answers structured), not for long, changing project briefs.
• Edits to Custom Instructions affect future conversations; they do not retroactively change earlier chats.
• Custom Instructions differ from Memory (what ChatGPT may remember when enabled) and Projects (a workspace that groups chats, files, and project instructions and includes built-in project memory).
• A simple writing pattern (clear rules, conditions, and conflict handling) makes instructions easier to follow and easier to maintain.

If you find yourself typing the same preferences at the start of every ChatGPT conversation (for example: "be concise," "use bullet points," "ask clarifying questions," "write in my brand voice"), Custom Instructions are the built-in way to make those preferences reusable.

This guide explains how Custom Instructions work today, where to enable them, how "save and reuse" actually behaves, and how to write instructions that stay useful over time. It also clarifies how Custom Instructions differ from ChatGPT Memory and ChatGPT Projects so you can choose the right tool for the job.

For OpenAI's official overview of Custom Instructions, see: https://help.openai.com/en/articles/8096356-custom-instructions-for-chatgpt.

What ChatGPT Custom Instructions are (and what they are not)

Custom Instructions are reusable instructions you configure in ChatGPT's Personalization settings. Once set, they apply across chats, and you can edit or delete them for future conversations.

What they are good for

• Stable defaults you want in nearly every chat (tone, formatting, level of detail).
• Process preferences (ask clarifying questions first, show assumptions, include a checklist).
• Output constraints that are consistent (word limits, structure, "include examples").

What they are not good for

• Long, evolving project briefs that change frequently.
• One-off task context (the specifics of today's email, today's dataset, today's requirements).
• A multi-variant system where you constantly switch between different "modes" (for example, "legal reviewer" vs "marketing writer" vs "coding tutor"). In that case, you may prefer to keep a baseline in Custom Instructions and put the mode-switching instruction in the first message of a chat.

Where to enable and edit Custom Instructions

Custom Instructions are configured in ChatGPT's Personalization settings. From there, you can add, edit, or delete them, and those changes apply to future conversations.

Because menu labels can change over time, the most reliable step-by-step reference is OpenAI's help page: https://help.openai.com/en/articles/8096356-custom-instructions-for-chatgpt.

How "write, save, and reuse" works in practice

1) Write: treat them like defaults, not a prompt

Custom Instructions work best when they read like standing preferences rather than a full task prompt. Your actual task prompt still matters, and you'll still provide task-specific context in each conversation.

2) Save: they persist until you change them

Once you save Custom Instructions, ChatGPT uses them as part of your personalization going forward. If you later decide they are too strict, too long, or no longer match your workflow, you can revise them.

3) Reuse: they apply across future chats

The key reuse behavior is scope: Custom Instructions apply across chats. That makes them useful for preferences you want almost everywhere.

4) Edits: changes affect future conversations

When you edit Custom Instructions, the updated version applies to future conversations. Existing chats are not retroactively rewritten. If you want a different behavior inside an existing chat, add a message in that chat that overrides or refines the instruction for that thread.

How to write Custom Instructions that are easy to follow

The goal is clarity and maintainability: instructions that are specific enough to be followed, but not so long that you constantly tweak them.

Use a simple structure: Output, Process, Constraints, Conflict rule

• Output: What the answer should look like.
• Process: How ChatGPT should proceed (questions first, show assumptions, provide options).
• Constraints: What to avoid or limit (hype, jargon, length).
• Conflict rule: What happens when your message conflicts with the defaults.

Prefer "When X, do Y" to reduce conflicts

Conditional rules reduce the chance that your defaults fight your task prompt.

• When I ask for a rewrite, provide two versions: (1) concise, (2) detailed.
• When requirements are unclear, ask up to 3 clarifying questions before answering.
• When I ask for code, include a minimal usage example.

Make instructions testable

Vague instructions are hard to follow consistently. Replace "be helpful" with something you can verify:

• Start with a 1-sentence answer, then bullets.
• Include a "Watch-outs" section when there are risks or edge cases.
• If you make assumptions, list them explicitly.

Keep them lean to leave room for task context

Every model has a finite context window measured in tokens, and practical limits vary by model version and usage tier. If your Custom Instructions are very long, they can compete with the task-specific context you paste into a chat. OpenAI's background on tokens and limits is here: https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them.

Practical Custom Instructions examples you can copy and adapt

Use these as starting points, then adjust based on what you actually do in ChatGPT. After you set them, run a few normal requests and see what you want to tighten or remove.

Example A: Clear business writing defaults

• Write in clear international English. Prefer short paragraphs.
• Avoid hype and exaggerated claims.
• When you propose changes, explain the reason briefly.
• If you make assumptions, list them under "Assumptions."
• If my message conflicts with these instructions, follow my message.

Example B: Consistent formatting for copy/paste into docs

• Start with a 1-sentence direct answer.
• Then provide steps as bullet points.
• Use headings when the answer has multiple parts.
• Include a short checklist at the end when the task is procedural.

Example C: Coding help with guardrails

• When I ask for code, ask up to 3 clarifying questions if requirements are missing.
• Prefer small functions and descriptive names.
• Include a minimal example showing how to run or call the code.
• If there are trade-offs, list them briefly before the code.

Example D: Decision support (options + recommendation rule)

• When I ask for a recommendation, give 2-3 options with pros/cons.
• End with a clear rule: "Choose X if …; choose Y if …".
• If you are uncertain, say what information would change your answer.

Custom Instructions vs Memory vs Projects (what to use when)

These features can feel similar because they all influence what ChatGPT does next, but they solve different problems.

Custom Instructions: your explicit, global defaults

• Best for: Stable preferences you want across future chats.
• Control: You write them deliberately and can edit or delete them for future conversations.

Memory: continuity that can draw from your usage (when enabled)

ChatGPT Memory is a separate feature that, when enabled, can use context from chats, files, and connected apps. OpenAI notes that the memory summary does not include everything ChatGPT may remember. OpenAI also notes that saved memories are stored separately from chat history, so deleting a chat does not by itself delete a saved memory from that chat. Details: https://help.openai.com/en/articles/8590148-memory-in-chatgpt-faq.

Projects: a workspace for ongoing work

ChatGPT Projects group chats, uploaded files, and project instructions for ongoing work, and projects have built-in memory. Project-only memory can reference chats inside the project but not conversations outside it; default-memory behavior depends on plan and settings. Official overview: https://help.openai.com/en/articles/10169521-projects-in-chatgpt.

Decision table: where to put different kinds of information

Common pitfalls (and how to fix them)

Pitfall 1: Instructions are too vague

Symptom: Results feel inconsistent.

Fix: Replace vague goals with observable behaviors (structure, length, number of options, when to ask questions).

Pitfall 2: Conflicting defaults

Symptom: You ask for "be concise" and also "be exhaustive," and you get unpredictable depth.

Fix: Add conditions: "Be concise by default; be detailed only when I say 'deep dive'."

Pitfall 3: Putting changing project requirements into global defaults

Symptom: You keep editing Custom Instructions and it affects unrelated chats.

Fix: Keep Custom Instructions stable. Put project-specific requirements in a Project's instructions (for that project) or in the first message of a dedicated chat.

Pitfall 4: Expecting Custom Instructions to replace task context

Symptom: You want ChatGPT to "already know" a long brief without you providing it.

Fix: Treat Custom Instructions as defaults, then paste the relevant context for the specific task (or keep it inside a Project where appropriate).

Frequently Asked Questions

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Related Guides

• What Is an AI Context Pack?
• How to Prepare Better Context for ChatGPT
• How to Organize Copied Text for AI Work
• Source-Labeled Context for AI: Why It Matters
• ChatGPT Context Pack for Consultants