ChatGPT Custom Instructions: A Setup Guide

Published: March 11, 2025 · 5 min read

Every conversation with ChatGPT starts from zero: you explain what you do, how you like answers formatted, what you already know — and then you do it again tomorrow. Custom instructions fix that. They're a short profile you write once (who you are, what you want) that ChatGPT applies to every chat automatically, and most users either skip them entirely or fill them with one vague line.

This guide shows you how to set them up properly: what to put in each field, templates for common roles, and how to update them as your needs change. Ten minutes of setup will make every future conversation sharper.

1. Where to find them and how they work

Custom instructions live in ChatGPT's settings (under Personalization). There are two boxes: "What would you like ChatGPT to know about you?" and "How would you like ChatGPT to respond?" The first is your context; the second is your style guide. ChatGPT prepends both to every conversation, so they act like standing instructions you never have to repeat.

They don't override what you say in a specific chat — a direct instruction in the conversation always wins. Think of them as defaults: they handle the 80% of your preferences that are constant, and you fine-tune the rest per chat. They're also editable anytime, so don't agonize over the first version.

2. What to put in the "about you" box

This is where most people write too little. "I'm a student" tells the model almost nothing. Give it the context that changes answers: your role, your tools, your level, and what you're working on. Aim for 3-5 short lines.

• Role and domain: "Marketing manager at a 20-person SaaS startup" — this calibrates examples and jargon.
• Tools you use: "I work in Python, use Notion and Google Sheets daily" — so code and workflows match your stack.
• Skill level: "Intermediate at statistics, beginner at SQL" — so explanations pitch at the right depth.
• Current projects: "Redesigning our onboarding emails this month" — so suggestions stay relevant.

Skip anything truly personal — no full name, address, or sensitive identifiers. Professional context is the sweet spot.

3. What to put in the "how to respond" box

This is the higher-leverage box. Be explicit about format, length, and tone — the model's defaults are wordy and hedged, and most people want the opposite. Strong examples:

• "Use bullet points by default. Short paragraphs. No throat-clearing intros."
• "When I ask for code, give the complete working snippet first, then explain."
• "Challenge my assumptions when they seem wrong; don't just agree."
• "Keep answers under 200 words unless I ask for detail."
• "Explain jargon when you use it; I'm learning this field."

The "challenge me" instruction is the single most underused one. Without it, the model agrees with everything, including your mistakes. With it, you get a sparring partner instead of a yes-machine.

4. Templates for common roles

Steal and adapt these starting points:

Developer: About you — "Full-stack dev, React + Node, 3 years experience." Respond — "Code first, explanation after. Flag security issues proactively. Prefer standard library solutions over new dependencies."

Writer: About you — "Freelance writer covering tech, client work plus a newsletter." Respond — "Match a clear, conversational tone. Suggest cuts, not just additions. Never use corporate buzzwords."

Student: About you — "Second-year biology undergrad." Respond — "Teach, don't just answer: explain the reasoning step by step. Quiz me after explanations. Keep it encouraging."

Founder: About you — "Solo founder of a productivity app, pre-revenue." Respond — "Be direct and critical. Prioritize by impact. No motivational filler."

5. Maintain them like a profile, not a tattoo

Your instructions should change as you do. Set a monthly reminder to re-read both boxes: delete the project that's done, add the new tool you adopted, tighten anything that isn't working. If you find yourself typing the same correction repeatedly ("shorter!", "in Python!"), that's a sign it belongs in the instructions.

One more tip: custom instructions apply to every chat, including quick one-offs where they don't fit. For those, start the chat with "ignore my usual preferences for this one" — or keep a separate lightweight context in mind. Defaults should serve you, not trap you.

6. Test and refine in one sitting

After filling in both boxes, run a quick calibration: ask something you ask often and compare the answer to what you'd have gotten before. Too long? Tighten the response box. Missing your context? Add a line to the about-you box. Two or three rounds of this and the defaults start feeling invisible — which is exactly right.

Watch for one failure mode: instructions that made sense for last month's project but now misfire. The monthly review catches these. If an instruction ever makes an answer worse, delete it on the spot rather than working around it — stale defaults cost you a little on every single chat.