OpenAI opened ChatGPT’s Custom Instructions to all users earlier this year, and a recent setup guide published by TechnoSports on July 31, 2026 has left many readers with outdated advice—from claiming the feature is still locked to a $20/month Plus subscription to citing character limits that are now too low and models that have already been retired. The good news for everyday users, power users, and IT pros alike: you don’t need a paid plan to set persistent preferences that shape how ChatGPT behaves, and if you do pay, you can store more than triple the space the guide suggests.
What Custom Instructions Actually Do
At its core, the Custom Instructions feature—first introduced on July 20, 2023 as a Plus-only beta—gives you two text fields that act like a permanent briefing for every chat. The first, “What would you like ChatGPT to know about you?”, lets you describe yourself, your environment, or your preferences (for example, that you develop on Windows 11, use PowerShell, and prefer TypeScript). The second, “How would you like ChatGPT to respond?”, defines the tone, format, or constraints of the answers you’ll receive.
Instead of repeating “I’m on Windows, don’t give me Bash scripts” or “keep answers concise and avoid marketing fluff” in every conversation, you save those instructions once and they’re appended silently to every prompt. It’s not a prompt injection that overrides the model’s safety rules—rather, it’s a set of user-side defaults that the model tries to respect in each response. When it works well, you spend less time course-correcting and more time getting usable output.
Plan Limits and the Character Count Reality
According to OpenAI’s own Help Center documentation—updated after the feature’s expansion—Custom Instructions are now available on every ChatGPT plan, not just Plus. The latest tiers and their limits are:
| Plan | Character Limit |
|---|---|
| Free, Go | 1,500 |
| Plus, Pro, Business, Enterprise, Education | 5,000 |
That 1,500-character cap is enough for a tight, focused profile: your OS, preferred shell, a couple of language choices, and a short output style. But a developer trying to paste an entire linting config or a monorepo’s contribution guide into the global instruction will hit the wall fast—and that’s by design. 5,000 characters offer breathing room, but they don’t eliminate the need for discipline; a longer instruction isn’t automatically better, and dumping repository-level rules into an account-wide setting creates a mess when you switch projects.
The TechnoSports guide also frames the late GPT-4o model as the central coding workhorse for this workflow. OpenAI retired GPT-4o and several older models from ChatGPT on February 13, 2026. Custom Instructions remain model-agnostic—they apply to whichever model you select—but the retirement underscores a practical point: your instructions should be stable and model-agnostic too. If your prompt relies on a model-specific quirk, it may break when the model changes. The better habit is to use instructions for universal preferences (platform, language, tone) and leave model-specific nudges to the individual prompt.
Setting Up Your Instructions: A Windows User’s Playbook
For the broadest Windows audience—not just developers—here’s how to fill the two fields productively.
The “About You” Field
Tell the AI the context that makes a material difference. If you’re on Windows, name it explicitly. Examples:
- “I use Windows 11 23H2, primarily with Microsoft Edge, Office 2024, and PowerShell 7.4.”
- “I’m an IT admin managing a small business network with Windows Server 2025, Active Directory, and M365 E3.”
- “I’m a parent helping my kid with homework on a Surface Laptop, no special dev tools.”
Avoid vague statements like “I’m a tech enthusiast” that add noise. Include only what would change the answer: if the AI would give the same help whether you mention your browser or not, leave it out.
The “Response Style” Field
Here, define how you want the answer delivered. A few proven patterns:
- Plain-language help: “Explain in short paragraphs, no jargon. If you must use a technical term, define it first.”
- PowerShell over Bash: “For any command-line instructions, give me the PowerShell syntax. If multiple shells are needed, show PowerShell first and label others.”
- Code answers: “When providing code, be concise and technical. No greetings, summaries, or marketing language. Prefer strict TypeScript, include error handling, and state your assumptions.”
- Documentation mode: “Format as a Markdown-ready help article with headings, steps, and a FAQ section at the bottom.”
A single response-style field can cover multiple modes if you’re careful. For example:
For everyday questions, use plain English and avoid corporate speak. For technical problems on Windows, provide PowerShell commands and note any registry edits. For code requests, give only the changed functions or a unified diff, with tests when relevant. Always assume Windows 11 and PowerShell 7.
Test it with three different requests—a trivia question, a PowerShell automation ask, and a coding prompt—to see if the output reliably follows the rules.
Going Beyond Global Instructions: Projects and Custom GPTs
One of the most common traps is treating the global Custom Instructions like a personal CONTRIBUTING.md. When you work across multiple codebases, clients, or personal projects, different conventions should not leak from one to another.
OpenAI’s Projects feature—available on paid plans—solves this. A project can bundle chats, files, and its own instruction set. According to OpenAI, project-level instructions override your global Custom Instructions inside that project. That means you can keep your global “Windows 11 + PowerShell” defaults, while a project for a .NET desktop app adds “target .NET 10, use WinUI 3, follow the AnalyzerConfig at ./.editorconfig.” The global instructions don’t change, and the project’s rules don’t pollute your weekend Python experiments.
Teams get an extra isolation benefit: shared project context is separate from each member’s personal instructions and memories. So a team’s coding standards live in the project, while Alice’s preference for “no emojis” stays in her own account. For even more repeatable workflows, paid users can create custom GPTs—mini agents with instructions, knowledge files, and selected capabilities. Those are ideal for internal tools that need a stable interface, but they introduce governance questions (who can edit? what files are uploaded? does it comply with data policies?) that go deeper than the scope of this article.
Are Custom Instructions a Magic Bullet? Consistency Yes, Perfection No
The TechnoSports article cites a potential 40% reduction in prompt-iteration cycles, but that’s an anecdotal figure, not an OpenAI-mandated performance metric. What users will actually experience is more qualitative: fewer rounds of “Sorry, I meant PowerShell, not Bash,” fewer lengthy explanations tacked onto a code snippet, and a first draft that more often matches your environment. That’s particularly noticeable for boilerplate-heavy work—app scripts, configuration files, small refactors—where every exchange used to begin with a style reset.
But a few hard limits remain. Custom Instructions won’t stop a model from hallucinating an API or misunderstanding your codebase. They don’t override OpenAI’s content policies or guarantee every line of code is secure. And they are absolutely not a place for secrets. API keys, connection strings, internal hostnames, or unreleased credentials in the instruction fields will be stored by OpenAI, can appear in data exports, and could be exposed by sharing screenshots or project exports. Keep secrets out.
How to Get Started Right Now
- Access the settings: In the ChatGPT web UI (or desktop/mobile app), click your profile picture → Customize ChatGPT or Custom Instructions.
- Enable the feature: If you see a toggle, flip it on.
- Write a minimal first draft: Start with the absolute essentials—your OS, a preferred shell, and one response-style rule. Example:
I’m on Windows 11. Use PowerShell for commands. Keep answers short.
- Test and iterate: Ask ChatGPT to write a small script, explain a Windows setting, then summarize a news article. Note where it stumbles and add one rule at a time.
- Check your template after model changes: When ChatGPT announces a new default model, revisit your instructions. They should work without modification, but it’s a good moment to confirm.
If you’re a developer, layer your guidance: keep platform defaults in Custom Instructions, codebase rules in Projects, and acceptance criteria in each task prompt. The combination is what turns a generic assistant into a consistent tool, without pretending that consistency equals correctness.
Outlook
The expansion of Custom Instructions to all plans signals a broader move by OpenAI to make ChatGPT more customizable at the account level. Combined with Memories (which can recall facts across chats) and Projects, the boundaries are blurring between “chat” and “workspace.” The advice for users—especially on Windows, where shell and path differences remain a daily friction—is to treat these instructions as a living configuration, not a set-and-forget script. As models evolve, the instructions that survive will be the ones that define your environment, not the quirks of a single model.