Direct answer
The limit is 1,500 characters on Free and Go, and 5,000 on eligible paid plans
OpenAI currently allows up to 1,500 characters in Custom Instructions for Free and Go users, and up to 5,000 for Plus, Pro, Business, Enterprise, and Education users. Use the global field for stable, non-sensitive rules that should follow you across ordinary chats. Keep client details, project files, and one-off task context closer to the work. More room can support a clearer operating brief, but it does not prove better answers or expand the model’s context window.
What to do now
Choose the lane that matches how you work
You repeat the same setup in ordinary chats
Move stable role, audience, format, and quality rules into a concise global brief. Keep a 300-character edit buffer.
Your work lives inside Projects
Keep project-specific instructions and files with that project. Do not duplicate every rule globally until you verify the behavior you want.
You use Free or Go, or rarely repeat setup
Free and Go currently provide 1,500 characters. Do not upgrade for extra text alone unless repeated setup is a measurable cost.
Implementation path
Turn a longer instruction field into a governed team workflow
TPTK can map durable rules, project context, approval checks, and the automations around the work without putting sensitive business data into a global brief.
Current Custom Instructions Limits by Plan
OpenAI’s current Custom Instructions documentation confirms two limits: 1,500 characters for Free and Go, and 5,000 for Plus, Pro, Enterprise, Business, and Education. The July 15 release raised the eligible paid-plan limit from 1,500 to 5,000; it did not announce a larger context window or guarantee that longer instructions improve answers.
Method: 5,000 divided by 1,500 equals 3.33. The difference is 3,500 characters, or about 233% of the Free/Go limit. These are TPTK calculations from OpenAI’s published values.
How to Use the 5,000-Character Limit
A longer field is useful only when each line has a stable job. TPTK’s allocation below totals 4,700 characters and preserves 300 characters, or 6%, for future edits. It is an operating model, not an OpenAI requirement.
Role and audience
Define your recurring role, the people you serve, and the expertise ChatGPT may assume.
Goals and decisions
State durable priorities, decision principles, and the tradeoffs you usually prefer.
Output defaults
Set structure, tone, units, length defaults, and how finished work should be delivered.
Quality and risk checks
Require source checks, uncertainty labels, privacy boundaries, calculations, and professional review where needed.
Escalation and exceptions
Explain when to ask a question, stop, surface a conflict, or present alternatives.
Evidence boundary: This allocation proves that a more complete brief fits inside the announced capacity. TPTK did not run a controlled model-quality test, so this report does not claim that longer instructions automatically improve output.
Global, Project, or Prompt: What Goes Where
More global space does not mean every instruction should become global. Put a rule as close as practical to the work it governs.
Use Custom Instructions for stable defaults
- Your durable role and audience
- Preferred output structure and units
- Cross-chat quality and privacy rules
Use project instructions for bounded work
- Client-neutral project goals
- Team terminology and reusable files
- Instructions that should not affect unrelated chats
Use the prompt for today’s task
- The immediate objective and source material
- Deadlines, constraints, and requested deliverable
- Temporary exceptions to your defaults
OpenAI’s Projects guidance says project instructions apply only inside that project and override global Custom Instructions. Shared projects are bounded context spaces and do not use members’ outside Custom Instructions or memories. That makes duplication a governance choice, not an automatic benefit.
What Not to Put in a Global Instruction Field
OpenAI says Custom Instructions can affect chats broadly, and Temporary Chats still follow them when they are enabled. Temporary Chat is therefore not a way to bypass an unsafe global brief.
Data handling also depends on the workspace. OpenAI says content from ChatGPT Business, Enterprise, and Edu is not used to train its models by default. On consumer plans, users can manage the Improve the model for everyone setting in Data Controls. Neither policy turns a global instruction field into the right place for confidential records.
Should This Change Which ChatGPT Plan You Buy?
The extra capacity matters when
- You already use an eligible plan.
- You repeat a long setup across ordinary chats.
- Consistency checks save real review time.
- The global brief contains no sensitive project data.
Do not upgrade for this alone when
- Your work is mostly project-specific.
- A short brief already covers your stable defaults.
- You need shared governance more than personal space.
- You have not measured repeated prompting as a bottleneck.
For the durable product decision, see TPTK’s ChatGPT guide for real estate professionals and AI automation tools for small business. This news page owns the July 15 limit change, not the broader subscription comparison.
What Is Confirmed—and What Is Not
OpenAI’s current help page states a 1,500-character limit for Free and Go and a 5,000-character limit for Plus, Pro, Enterprise, Business, and Education.
The 4,700-character allocation and placement map are editorial recommendations derived from the published capacity and current product documentation.
A larger saved field does not by itself prove better output quality. TPTK has not run a controlled model-quality test comparing shorter and longer instruction sets.
TPTK will update this URL if OpenAI changes plan eligibility, the limits, data guidance, or the documented interaction with Projects.
Build the system, not just the prompt
Need a reusable AI operating brief and workflow?
TPTK can turn your recurring rules into a practical brief, route project context safely, and build the forms, approvals, integrations, and internal tools around the process.
ChatGPT Custom Instructions Limit FAQ
What is the current ChatGPT Custom Instructions limit?
OpenAI currently allows up to 1,500 characters for Free and Go users and up to 5,000 characters for Plus, Pro, Business, Enterprise, and Education users.
Does the 5,000-character limit apply to ChatGPT Free or Go?
No. OpenAI’s current documentation lists a 1,500-character limit for Free and Go. The 5,000-character limit applies to Plus, Pro, Enterprise, Business, and Education.
Should confidential client information go in Custom Instructions?
No. Use Custom Instructions for stable, non-sensitive working preferences. Keep private client facts, credentials, records, and project-specific source material out of a field that can influence unrelated chats.
Are Custom Instructions the same as memory or project instructions?
No. Custom Instructions are explicit global guidance, memory can draw from information shared in conversations, and project instructions belong to a bounded project context. Project instructions override global Custom Instructions inside that project.
Update Log
August 26, 2026: Reverified OpenAI’s current plan limits; replaced the resolved documentation-mismatch warning with the confirmed 1,500 Free/Go versus 5,000 eligible-paid split; updated Projects guidance, FAQs, sources, metadata, and schema.
July 15, 2026 at 6:05 PM PDT: Initial publication based on OpenAI’s release note and then-current documentation.
Primary Sources
- OpenAI Help: ChatGPT Custom Instructions. Current plan limits; accessed August 26, 2026.
- OpenAI: ChatGPT Release Notes. July 15, 2026 limit-change entry; accessed August 26, 2026.
- OpenAI Help: Projects in ChatGPT. Project-instruction scope; accessed August 26, 2026.
- OpenAI Help: Temporary Chat FAQ. Custom Instructions behavior in Temporary Chat; accessed August 26, 2026.
- OpenAI: Business data privacy, security, and compliance. Accessed August 26, 2026.
- OpenAI Help: Data Controls FAQ. Accessed August 26, 2026.