ChatGPT Custom Instructions Now Allow 5,000 Characters. Use About 1,000.

Executive summary

On July 15, OpenAI increased the ChatGPT custom instructions limit from 1,500 to 5,000 characters for Plus, Pro, Enterprise, Business, and Education users. The extra room is useful. Most people should leave plenty of it blank.

Custom instructions work best as a short operating brief for the way you want ChatGPT to respond across everyday tasks. Put stable context, output preferences, evidence standards, and working rules there. Keep the assignment itself in the chat.

Our recommendation is to start with about 1,000 characters, test the instructions on real work, and add detail only when the same problem keeps showing up.

What changed in ChatGPT custom instructions

The July 15 ChatGPT release note says eligible users can now save up to 5,000 characters, up from 1,500. OpenAI says the added space gives users more room to customize ChatGPT's response style and behavior.

That gives frequent users enough room to set useful defaults. A finance leader can define numerical precision. A marketing lead can set tone and source expectations. An operations manager can request a consistent format for decisions, owners, and next steps.

According to OpenAI's custom instructions help page, the instructions apply across chats. That makes them a poor home for details about one project.

Treat 5,000 characters as a ceiling. A shorter block is easier to test and maintain.

What belongs in custom instructions

Use custom instructions for information that stays true across many conversations.

A good version usually covers your role, the audience you write for, the kinds of tasks you repeat, the response format you prefer, and the standards ChatGPT should follow when evidence is incomplete.

For example, a business leader might ask ChatGPT to lead with the recommendation, keep paragraphs short, separate facts from assumptions, include source links for current claims, and flag missing data before calculating a result. Those directions improve a wide range of tasks without forcing the user to repeat them every morning.

OpenAI's prompt engineering guide says high-level instructions can define tone, goals, and examples of a correct response. Keep those defaults specific enough to guide the work and short enough to maintain.

What should stay out

Do not turn the field into a storage locker for every prompt your team has ever liked.

Keep temporary facts in the current chat or project. Keep sensitive company data, credentials, private client details, and deal terms out of reusable instructions. Do not paste a long policy manual and assume the model will know which paragraph matters for each task.

Task-specific directions belong in the task prompt. If you need a board memo by Friday, give ChatGPT the audience, source material, decision, deadline, and required format in that conversation. Let custom instructions shape the response; keep the assignment in the conversation.

Ask whether the instruction would still help next month on a different piece of work. Keep durable guidance in the settings and temporary directions with the task.

A practical custom instructions template

This starter fits comfortably inside the new limit and gives ChatGPT enough direction without burying the work.

ROLE AND CONTEXT
I lead [function] at a [company type and size]. I use ChatGPT for [common tasks]. My main audience is [audience]. Assume they understand [domain] but do not want technical jargon unless it changes the decision.

RESPONSE DEFAULTS
Lead with the answer or recommendation. Use short paragraphs and descriptive headings. Give me the smallest useful amount of detail first. Put optional depth after the main answer.

EVIDENCE
Separate verified facts, estimates, and assumptions. Link current claims to primary sources when possible. Never invent numbers, quotes, examples, or citations. If a required input is missing, name it before proceeding.

WORKING STYLE
Use plain English. Avoid filler, inflated claims, and generic conclusions. When reviewing a plan, point out the strongest objection. When drafting something for external use, return a review draft rather than implying it has been sent, published, or approved.

OUTPUT CHECK
Before finishing, confirm that the response answers the actual question, uses the requested format, and does not contradict the source material.

Change the placeholders and test it before adding more. In practice, one sharp sentence about the audience often does more work than another page of rules.

How to test your instructions in 20 minutes

Choose three tasks you already do: a summary, a recommendation, and a draft. Run each task with your new custom instructions enabled.

Look for repeated friction. Did the answer bury the recommendation? Add one sentence about answer order. Did it mix estimates with facts? Tighten the evidence rule. Did the draft sound stiff? Add a short voice sample or describe the audience more clearly.

Change one part at a time, then rerun the same tasks. If you rewrite the whole instruction block after every weak answer, you will not know which change helped.

OpenAI says custom instructions can be edited or deleted, and updates affect future responses. Treat them like a working document. A quarterly review is sensible for active users, especially when roles, tools, or company terminology change.

The team use case

Custom instructions are personal settings, but teams can still benefit from a shared starting point.

Create a short approved template for a role, such as sales manager, analyst, recruiter, or project lead. Let each person adapt the audience and working style. Then compare output on two or three common tasks during training.

Aim for a shared baseline on sources, decisions, and useful output. Each person can still sound like themselves. This is the kind of workflow practice we build into AI Experts SuperHumans, where teams learn AI through work they already own.

Multi-step automation, connected systems, and repeatable routing need workflow design beyond personal settings. AI Experts services can help turn a useful personal pattern into training, an owned workflow, or a purpose-built AI tool.

TLDR

OpenAI gave eligible users more space. Use it for clearer defaults. Stop when the instructions do their job.

Start with stable context, response preferences, evidence rules, and one output check. Test the instructions on real tasks. Keep what improves the work and delete what merely sounds thorough.

After two or three rounds, save the shortest version that consistently improves your work.