Workbranch › Guides

How to set custom instructions in ChatGPT

Custom instructions let you tell ChatGPT once how you want it to answer, instead of repeating yourself in every chat. Here's where the setting lives, what's worth putting in it, and how it works alongside projects, memory and personality.

Updated 15 October 2026

These steps were checked against OpenAI's help pages in October 2026. Menu labels in ChatGPT change from time to time, so look for the closest match if a name differs slightly.

Set custom instructions on the web or desktop app

1. Select your profile icon and open Settings.
2. Choose Personalization.
3. Check that Enable customization is switched on. While it's off, your saved instructions aren't used.
4. Click into the Custom instructions field and write what you want ChatGPT to keep in mind (examples below).
5. OpenAI says updates take effect right away. Its FAQ also says edits show up in future conversations, so start a new chat to be sure you're seeing the new version.

Set them in the iPhone or Android app

1. Tap your profile icon to open Settings.
2. Open Personalization. OpenAI's custom instructions page calls this entry Customize ChatGPT, so your app may show that name instead.
3. Turn on Enable customization.
4. Enter your text in the Custom instructions field.

What to write: examples you can adapt

Good custom instructions describe things that are true in almost every chat: who you are, what you usually use ChatGPT for, and the shape of answer you find most useful. Aim for short, specific lines rather than a long wish list. A few starting points:

• Context about you: "I run a two-person bakery in Leeds. Most of my questions are about pricing, suppliers and social media posts. Use British spelling and pounds sterling."
• Answer format: "Start with the answer in one or two sentences, then the details. When I ask how to do something, give numbered steps."
• Level of detail: "I'm comfortable with spreadsheets but not with code. If a solution needs code, explain where to paste it and what it does."
• Working style: "If my request is ambiguous, ask one clarifying question before answering. Tell me plainly when you're unsure or when a fact may be out of date."
• Things to stop: "Skip the recap at the end of an answer." Where you can, pair a ban with what you want instead. OpenAI's advice for writing GPT instructions also prefers clear, positive "do this" rules to a long string of don'ts.

Keep anything task-specific out of the global field. A brief for one client or a style guide for one report belongs in a project (see below), not in instructions that will shape your recipe questions too. For prompts that make answers easier to act on, see our prompt patterns for actionable answers.

Personality and characteristics: the other half of the page

The same Personalization page has two controls that change tone without any writing:

• Base style and tone is a dropdown of preset personalities. As of October 2026 the options are Default, Professional, Friendly, Candid, Quirky, Efficient and Cynical.
• Characteristics let you nudge four things up or down: warmth, enthusiasm, the use of headers and lists, and emojis.

If you only want shorter, plainer answers, these may do the job faster than writing instructions. They all work together: OpenAI notes that your custom instructions and saved memories add to the chosen personality, and that a conflicting preference, or something you say mid-chat, can tone its traits down. A personality also doesn't decide the style of things you ask ChatGPT to produce, such as an email or code; your request and instructions do.

How custom instructions work with projects

• Project instructions win. Each project can have its own instructions (open the project's ••• menu and choose Project settings). Those instructions only affect chats in that project, and there they take priority over your general custom instructions.
• Enterprise and Edu are stricter. In Enterprise and Edu workspaces, chats inside a project use only the project's instructions, not your personal ones.
• Shared projects ignore personal settings. A shared project can't see any member's custom instructions or memories from outside it.
• Moved chats follow the project. When you move an existing chat into a project, it picks up that project's instructions.

For more on how ChatGPT's projects compare with Claude's, see ChatGPT Projects vs Claude Projects.

How they work with memory and temporary chats

• Memory is separate. Custom instructions are guidance you write yourself. Memory is what ChatGPT picks up from past chats, saved memories and other sources. You manage it under Settings → Personalization → Memory.
• Memory can draw on your instructions. OpenAI lists custom instructions among the sources memory may use, and a Sources note under a reply can show that one of your instructions shaped it.
• Temporary chats ask first. Before you start a temporary chat you pick Personalized or Unpersonalized. Only the personalized kind uses your custom instructions and memories, and you can't switch once the conversation has begun.

Good to know

• Every plan has it, on web, desktop, iOS and Android. The main difference between plans is the character limit: 1,500 on Free and Go, 5,000 on Plus, Pro and workspace plans.
• Model training: OpenAI says what you write in custom instructions may be used to improve its models. On personal accounts you can switch off Improve the model for everyone in Data controls; Business, Enterprise and Edu content isn't used for training by default.
• Plugins: if you use third-party plugins, relevant details from your instructions may be passed to their developers, so leave out anything you wouldn't want to share.
• Old chats keep old versions. Editing or deleting your instructions doesn't erase earlier versions from past conversations. To remove them, delete those chats.

If it doesn't work

• ChatGPT ignores your instructions. Check that Enable customization is on, and test your edits in a new chat. If you're in a project, its own instructions take priority, and in a temporary chat you need to have chosen Personalized.
• You can't save everything. You've probably hit the character limit. Cut repetition, or move task-specific material into project instructions.
• The tone still feels off. Look at Base style and tone and Characteristics, and at your memory summary: a remembered preference can pull answers in another direction.
• You want them off for now. Switch Enable customization off. Your text can stay saved, or you can clear the fields.

Turning good answers into finished work

Custom instructions can get ChatGPT to answer in clear, numbered steps, but those steps still scroll away with the chat. Workbranch is a separate workspace where answers from Google Gemini become a point-by-point checklist with progress, and you can open a side thread on any point without derailing the main conversation. There's a 3-day free trial.

Sources

• ChatGPT Custom Instructions (OpenAI Help Center, checked 15 October 2026)
• Customizing Your ChatGPT Personality (OpenAI Help Center, checked 15 October 2026)
• Characteristics in ChatGPT (OpenAI Help Center, checked 15 October 2026)
• Projects in ChatGPT (OpenAI Help Center, checked 15 October 2026)
• Memory in ChatGPT (OpenAI Help Center, checked 15 October 2026)
• Creating and editing GPTs (OpenAI Help Center, checked 15 October 2026)
• ChatGPT release notes (OpenAI Help Center, checked 15 October 2026)

Frequently asked questions