Skip to content

ChatGPT Custom Instructions Now Allow 5,000 Characters on Paid Plans: What Should You Write?

Learn which ChatGPT plans get 5,000-character custom instructions, what to include, and how to improve your setup.

Published: AI verified: Category: AI tools and comparisons

A laptop surrounded by expanding blank cards representing organized custom instructions

Do you find yourself repeatedly telling ChatGPT to lead with the answer, explain technical terms, or keep the tone professional? Custom instructions can store preferences that would otherwise need to be repeated. But when the available space grows, deciding what belongs there can become a new problem.

The practical answer is not to fill all 5,000 characters. Start with stable information that remains useful across conversations: your role, recurring tasks, preferred output format, verification rules, and writing style.

On July 15, 2026, OpenAI announced that the custom-instructions limit had increased from 1,500 to 5,000 characters for selected paid ChatGPT plans. This guide explains who gets the larger limit, what to write, what to leave out, and how to test a simple setup.

The 5,000-character limit is for paid plans

According to OpenAI’s official ChatGPT release notes, Plus, Pro, Enterprise, Business, and Education users can save up to 5,000 characters in custom instructions. The previous limit was 1,500 characters.

Free and Go users can save up to 1,500 characters. Custom instructions are available on every plan across the web, desktop, iOS, and Android, but the storage limit differs by plan. The change did not raise the limit to 5,000 characters for every user.

On the web and desktop, OpenAI directs users to Settings and then Personalization, where customization can be enabled. On iOS and Android, the official instructions point to Customize ChatGPT in Settings. Labels may vary slightly by language or app version.

The extra space makes it easier to include context and quality checks alongside response preferences. However, longer instructions are not automatically better. Repetition, outdated details, and conflicting rules can make your intent harder to follow.

Five useful things to include

OpenAI Academy recommends stable details such as your responsibilities, preferred tone, desired output types, and guardrails for unclear requirements. In practice, the following five categories provide a useful starting point.

1. Your role and context

  • Your role, responsibilities, or field of study
  • Your level of knowledge in topics you often discuss
  • Your main language or region

Statements such as “I manage communications for a small business” or “I am learning programming” help ChatGPT choose an appropriate level of explanation. You usually do not need to identify your employer or customers.

2. Recurring tasks

  • Drafting emails or articles
  • Summarizing and comparing information
  • Planning study sessions or organizing ideas

Listing common tasks helps establish how you intend to use the response. A condition that only applies to one project or one week belongs in the task prompt instead.

3. Preferred output format

  • Give the conclusion first
  • Present procedures as numbered steps
  • Briefly define technical terms
  • Use headings for longer answers

Instructions such as “make it clear” are subjective. A preferred length, structure, or number of examples is easier to apply and easier for you to evaluate.

4. Decision and verification rules

  • Ask a question when missing information would materially change the answer
  • Prefer official sources for current information
  • Separate established facts from inference
  • Include limitations and disadvantages

Custom instructions can describe not only the tone of an answer but also how you want uncertain information handled. They do not guarantee that every ChatGPT response will be correct, so important decisions still require checking the original source or consulting a qualified professional.

5. Language and tone

  • Reply in English unless asked otherwise
  • Use a professional but approachable tone
  • Avoid hype and overly confident claims
  • Use emoji only when it helps

Keep style preferences short and concrete. A few priorities are easier to maintain than a long collection of prohibited phrases.

A custom-instructions template you can adapt

The following compact template works for many professional and learning use cases. Replace the bracketed text and delete any lines you do not need.

ABOUT ME
I am a [role, profession, or learner in a particular field].
Explain topics at a [beginner, intermediate, or experienced] level.
Reply in English unless I ask for another language.

COMMON TASKS
I mainly use ChatGPT for [email drafting, summaries, research, or study support].
Prioritize practical steps and examples I can apply.

OUTPUT FORMAT
Lead with a concise answer.
Use numbered steps for procedures.
Briefly explain specialized terms.

VERIFICATION RULES
If missing requirements would materially change the result, ask before proceeding.
For current information, prefer official sources and identify anything you cannot verify.
Include relevant limitations and disadvantages as well as benefits.

STYLE
Use a natural, professional, and approachable tone.
Avoid hype, unnecessary preambles, and repeated conclusions.

There is no need to expand this immediately. Begin with three to five instructions you already repeat often, then add only what the results show is missing.

What not to put in custom instructions

Custom instructions may affect many conversations, so they are not a suitable storage area for every detail.

Do not enter passwords, authentication credentials, confidential customer data, or unnecessary identifying information. OpenAI’s Help Center explains that information from custom instructions may be used to improve model performance and that relevant details may be provided to third-party plug-in developers when those plug-ins are used. A safe default is to leave out anything you would not want disclosed.

Short-lived schedules and conditions for a single assignment belong closer to that work, such as in the conversation prompt or a relevant Project. OpenAI Academy similarly recommends using custom instructions for stable preferences and using the prompt for the specific task at hand.

Avoid rules that cannot comfortably coexist. For example, “always explain everything in detail” conflicts with “never write more than three lines.” If both brevity and accuracy matter, state the priority: “Prioritize accuracy, then be as concise as possible.”

Test the setup with three prompts

After saving your custom instructions, test them with three common prompts. Choose tasks with different shapes, such as drafting a short email, explaining a difficult concept, and comparing several options.

Check where the conclusion appears, how long the answer is, whether technical language fits your level, and when ChatGPT asks for clarification. If the result is not what you expected, look for ambiguity or conflicts before adding more rules.

Change one instruction at a time. This makes it easier to see which setting affected the response. Periodically removing outdated preferences is also a better use of the larger limit than continuously adding new ones.

Custom instructions and Memory also serve different purposes. Custom instructions define a default working style, while Memory can retain ongoing details shared through conversations. Keep temporary deliverables and deadlines in the conversation where they matter.

Summary

As of July 15, 2026, ChatGPT Plus, Pro, Enterprise, Business, and Education users can save up to 5,000 characters in custom instructions. Free and Go users retain the 1,500-character limit, although the feature itself is available on every plan.

A useful setup can begin with five categories: role and context, recurring tasks, output format, verification rules, and language and tone. Start with three to five preferences from the template, then test them against three prompts you regularly use.

The larger field is most valuable when it removes instructions you repeat every day—not when it gives you another blank space to fill.

Related posts

Author

ImidefWorks

An independent writer who calmly connects official sources with first-hand experience across AI, web work, indie development, and information organization.

View author profile and editorial policy