ChatGPT Custom Instructions can save you from repeatedly typing preferences such as “keep the answer concise,” “use Indian examples,” or “explain technical terms.” Set them once and ChatGPT considers them across your conversations. The feature is useful, but only when the instructions are specific, compatible, and short enough to remain relevant.
This guide explains where to find Custom Instructions, what belongs in them, what should stay in individual prompts, and provides practical templates you can adapt.
What are ChatGPT Custom Instructions?
Custom Instructions are account-level directions that tell ChatGPT what context it should consider and how you generally want answers written. According to OpenAI’s current documentation, they are available on all ChatGPT plans across the web, desktop, iOS, and Android apps.
They are different from a one-time prompt. A prompt controls the current request, while Custom Instructions establish reusable defaults. OpenAI says edits are applied immediately across chats, although older messages do not get rewritten.
Where to find Custom Instructions
On the web and desktop apps
- Open your profile menu and select Settings.
- Open Personalization.
- Select Custom Instructions or the available customization field.
- Turn on Enable customization, enter your instructions, and save them.
On Android and iPhone
- Open ChatGPT and enter Settings.
- Select Customize ChatGPT.
- Enable customization, add your preferences, and save.
The interface can change over time, but the setting should remain within the Personalization or Customize ChatGPT section.
Custom Instructions, Memory and Projects are not the same
| Feature | Best used for | Example |
|---|---|---|
| Custom Instructions | Explicit preferences that should usually apply across chats | “Use Indian English and put the recommendation first.” |
| Memory | Relevant personal context learned from conversations | Your profession, dietary preference or a long-term goal |
| Project instructions | Rules that apply only to one project or workflow | “Use our company style guide for every document in this project.” |
| Current prompt | Requirements unique to the task in front of you | “Compare these three phones in a 700-word article.” |
OpenAI’s Memory FAQ describes Custom Instructions as direct guidance you manage yourself, while Memory can retain relevant information shared through conversations. Inside a ChatGPT Project, project-specific instructions override global Custom Instructions, according to the Projects guide.
A good starter setup for most users
You can paste the following template and edit the bracketed details:
I am a [profession or skill level] based in [country or region]. Use [language or regional English] and assume [beginner/intermediate/advanced] knowledge unless I say otherwise.
Start with the direct answer. Keep routine answers concise, but explain important trade-offs. Use short headings and bullets only when they improve readability. Define unfamiliar technical terms. For comparisons, use a table followed by a clear recommendation for different types of users.
Do not invent facts, prices, quotations or sources. Clearly label uncertainty. For information that may have changed, check current reliable sources when search is available and include direct links. Ask a question only when missing information would materially change the answer.
This works because it defines context, response structure, quality checks, and when clarification is worthwhile. It does not try to predict every future task.
Useful Custom Instructions examples
For concise everyday answers
Lead with the answer in one or two sentences. Add only the details needed to act. Avoid repeating my question, long introductions and unnecessary disclaimers.
For learning a new subject
Explain concepts in plain language before using jargon. Use one practical example and then check my understanding with a short question. Do not assume prior expertise.
For work and business writing
Use a professional but natural tone. Prefer short paragraphs, concrete next steps and Indian English spelling. Flag unsupported claims and overly promotional wording.
For research and fact-checking
Separate verified facts, reasonable inferences and unknowns. Prefer primary sources. When web search is available, cite the source beside the claim it supports and include the publication or update date when relevant.
For shopping comparisons
Ask for my country and budget if they are missing. Verify current availability and prices. Compare meaningful differences in a table, explain compromises and recommend the best option for distinct buyer priorities.
For coding assistance
Explain the likely cause before presenting a fix. Preserve the existing stack and conventions. Provide the smallest safe change, mention affected files, and include a focused way to test it.
These are genuine instructions, not hidden slash commands. If you have seen shortcuts such as “/handwritten” on social media, read our guide to which ChatGPT slash commands are real.
What should not go into Custom Instructions?
- Sensitive information: Do not store passwords, financial details, confidential client data, or anything you would not want reused as context.
- One-off task requirements: A specific word count, document, or deadline belongs in the current prompt.
- Conflicting rules: “Always be brief” and “always explain everything in detail” cannot both work reliably.
- Impossible guarantees: Instructions cannot force perfect accuracy, remove safety rules, or provide capabilities the selected ChatGPT experience does not have.
- A giant prompt library: Too many niche rules make the useful ones harder to apply consistently.
OpenAI currently lists limits of up to 1,500 characters for Free and Go accounts and up to 5,000 characters for Plus, Pro, Business, Enterprise and Education accounts. Even when more space is available, shorter and prioritized instructions are usually easier to maintain.
How to fix Custom Instructions that are not working
- Confirm customization is enabled. Saved text will not help if the feature is switched off.
- Remove contradictions. Decide which preference wins and state it plainly.
- Replace vague adjectives. Instead of “make it better,” specify length, audience, tone, or format.
- Put exceptions in the current prompt. Say “For this answer, ignore my normal preference for brevity.”
- Check Project instructions. Instructions inside a Project can override your global settings.
- Run a simple test. Ask for a short explanation, a comparison, and an email draft to see whether the settings behave sensibly across formats.
OpenAI’s prompting guidance recommends clear, specific requests and iterative refinement. Custom Instructions improve defaults; they do not replace a well-scoped prompt.
Privacy points worth knowing
OpenAI says Custom Instructions are included in your ChatGPT data export and can be edited or deleted for future conversations. They are not shown to people opening your shared chat links. However, relevant information may be provided to third-party plug-in developers when you use those plug-ins, so avoid adding private or confidential details.
OpenAI also states that information from Custom Instructions may be used to improve model performance according to your data-control choices. Review your Data Controls if you do not want content used for model improvement.
Final checklist
- Keep permanent preferences global and task-specific rules local.
- State audience, tone and output structure concretely.
- Ask ChatGPT to label uncertainty rather than pretending every answer is certain.
- Avoid sensitive, temporary, and contradictory information.
- Review the instructions when your work or priorities change.
The best Custom Instructions are not the longest. A compact set of clear defaults makes ChatGPT more consistent while leaving each prompt enough room to define the actual task. Read more about ChatGPT tips we have covered in several categories.
