Custom Instructions let you save stable preferences for how ChatGPT should respond, so you do not have to repeat the same guidance in every conversation. OpenAI documents the feature as available on all plans through the web, macOS, Windows, iOS and Android versions of ChatGPT.
How to enable Custom Instructions in ChatGPT
The navigation differs slightly between desktop and mobile. Menu labels and the number of instruction fields can also change as the interface is updated.
On the web, macOS or Windows
- Open ChatGPT and go to Settings.
- Select Personalization.
- Open Custom Instructions.
- Turn on Enable customization.
- Enter the preferences you want ChatGPT to follow, then save them.
On iOS or Android
- Open the ChatGPT app and go to Settings.
- Select Customize ChatGPT.
- Turn on Enable customization.
- Enter your preferences and save them.
Depending on the interface available to your account, ChatGPT may present the preferences in one field or several structured fields. Follow the prompts shown without entering information that is not necessary for your preferred response style. These paths and availability are documented in OpenAI’s Custom Instructions documentation.
What to write in your Custom Instructions
Custom Instructions work best for preferences that should remain stable across conversations. Useful categories include tone, answer length, formatting, working style and recurring guardrails.
Write each preference as a direct, observable request. For example:
- Use plain language and explain unfamiliar technical terms.
- Start with a concise answer, followed by bullet points when more detail is useful.
- When giving instructions, use numbered steps and state any requirements first.
Avoid elaborate personas or conflicting rules. Short instructions make it easier to determine whether ChatGPT is following the requested style, although they do not guarantee that every response will use it perfectly.
OpenAI currently documents a limit of 1,500 characters for Free and Go accounts and 5,000 characters for paid plans. Use that space for recurring preferences rather than details needed for only one task.
Do not include passwords, financial details, confidential work information or other secrets. OpenAI also notes that relevant instructions may be shared with third-party plugins when you choose to use those plugins.
How to test, edit, disable or remove your instructions
After saving your instructions, start a new chat and use a harmless prompt that makes the requested format easy to see. For example:
Explain why leaves change color in autumn.
If you requested a concise opening followed by bullet points, check whether the answer uses that structure. A successful test shows that the preference influenced that response; it does not guarantee identical behavior in every ChatGPT feature or future answer.
OpenAI says Custom Instructions apply across chats, while additions or updates are reflected in future conversations. Previously generated responses are not rewritten.
- To edit: Return to the same Custom Instructions or Customize ChatGPT setting, change the text and save it. Test the revision in a new conversation.
- To disable: Turn off Enable customization.
- To remove: Delete the instruction text from the setting and save the change.
Disabling and deleting are separate actions. Because the supplied official guidance does not confirm whether every current interface retains entered text after the toggle is turned off, check the field before assuming disabled instructions will remain available for later reuse.
Custom Instructions vs Memory, project instructions and one-off prompts
Choose the option according to how broadly and how long the context should apply. OpenAI’s personalization guide treats Custom Instructions and Memory as separate ways to personalize ChatGPT.
| Option | Best use | Scope |
|---|---|---|
| Custom Instructions | Preferences you explicitly define, such as tone, format or a stable working style | Broadly across ChatGPT conversations |
| Memory | Information ChatGPT remembers separately from your written instructions | Conversations where Memory is available and active |
| Project instructions | Context or rules associated with one body of work | A specific project |
| One-off prompt | A temporary request, format or role needed for one task | The current request or conversation |
If you need to review information stored separately through Memory, see how to manage or clear ChatGPT memory. If instructions and context should be limited to a particular body of work, it is more appropriate to use a ChatGPT Project for project-specific context.
Do changes to Custom Instructions affect existing ChatGPT conversations?
OpenAI says Custom Instructions apply across chats, while updates affect future conversations. Earlier generated responses are not rewritten. To test a change clearly, start a new conversation and use a harmless prompt that reveals the requested tone or format.
Should I use Custom Instructions or ChatGPT Memory?
Use Custom Instructions for preferences you explicitly write, such as answer length, tone or formatting. Memory is a separate feature for information ChatGPT remembers. Use the dedicated Memory controls when you need to review or remove remembered information.