How to Make ChatGPT Remember Your Preferences (So You Stop Retyping the Same Instructions Every Time)

You check into a hotel for a week-long conference. Every single morning, the concierge greets you like a complete stranger. You explain again that yes, you need extra towels. No, you can't eat gluten. The room temperature should be exactly 67 degrees. By day three, you're wondering if they've hired a new person overnight. By day five, you finally scribble everything on a card and hand it over: "Keep this at the desk. Please."

That's ChatGPT without Custom Instructions. Every conversation starts fresh, with no memory of how you like things done.

Why the concierge keeps forgetting you

ChatGPT treats each new chat window as a blank slate. Open a new chat, and everything you explained yesterday about your preferred writing style, your audience, or the format you need is gone. The concierge has no idea who you are.

So you end up copy-pasting the same instructions into every conversation. "Write this in a casual tone." "Keep it under 200 words." "Don't use corporate buzzwords." You're maintaining a document of your standard instructions just so you can paste them in faster.

This isn't a bug. The system was designed this way for privacy. Your conversations don't automatically feed into some growing profile of you. But for regular users who want consistent help, it's wildly inefficient. By the fifth email of the week, you're ready to throw your keyboard.

Writing down your preferences on the card

Custom Instructions is the card you hand to the concierge. It's ChatGPT's built-in memory system, tucked away in your account settings. Once you fill it out, those preferences stick around for every new conversation you start.

You'll find two text boxes. The first asks "What would you like ChatGPT to know about you?" Your role, your projects, recurring constraints that shape most of your requests. The second asks "How would you like ChatGPT to respond?" Your output preferences: tone, format, length, things to avoid.

Fill these in once, and they persist across all new conversations unless you deliberately turn them off.

A marketing manager writes "I run email campaigns for a sustainable fashion brand, audience is 25-40, values authenticity" in box one, and "Always write in a warm, conversational tone. Keep paragraphs short. Avoid buzzwords like synergy or empower" in box two. Now every campaign idea automatically matches that style. No more explaining it fresh each time.

What to write on the card (and what to leave off)

The concierge can't anticipate every room service order you might place. The card is for recurring needs, not one-off requests.

Include your role, industry, or constraints that affect most of what you ask for. "I'm a software engineer who writes documentation for non-technical users." "I'm a grad student in psychology, APA format required." "I manage a remote team across four time zones."

Specify how you want output formatted. Bullet points or paragraphs? Casual or formal? Length limits? Level of detail? "Keep responses under 300 words unless I ask for more." "Use numbered lists for step-by-step instructions." "Write at an 8th grade reading level."

List things to avoid. Jargon you hate, overly complex sentences, certain phrases that make you cringe. "Never use 'circle back' or 'touch base.'" "Skip the inspirational quotes." "Don't explain things I didn't ask about."

But don't stuff in details relevant to only one project. "I'm writing a blog post about sea turtles" doesn't belong here. That still goes in your individual prompt.

Start with three or four instructions. Add more as you notice patterns in what you keep re-explaining.

A teacher might write "I teach high school biology" and "Explain concepts at a 9th grade reading level, use everyday analogies, break complex ideas into numbered steps." Now every lesson plan or quiz question comes back classroom-ready instead of sounding like a textbook. No more translating output into something teenagers can actually parse.

When the concierge checks the card

Every time you start a new chat, the concierge pulls your card. Custom Instructions apply automatically.

They don't apply retroactively to old conversations already in progress. If you're halfway through a chat when you update your instructions, that conversation still runs on the old settings. Start fresh to see the changes.

You can toggle the whole system off for specific tasks where you need different behavior. The card stays filed at the desk, but you're telling the concierge to ignore it temporarily.

The instructions live in your account settings, so they follow you across devices. Set them on your laptop, they're active on your phone.

If responses start feeling off, check your instructions. Sometimes they're too restrictive or they contradict each other. "Be extremely detailed" and "keep it under 100 words" can't both happen.

A consultant who normally wants concise, bullet-pointed answers can toggle off Custom Instructions when asking for a long-form article. The card stays on file, but the concierge knows this particular guest is ordering something different tonight.

So what can YOU do with this?

Set up your writing style once. Tone, sentence length, formatting preferences, words to avoid. Done. Every response now matches your voice automatically.

Save your role context so ChatGPT tailors examples to your industry without you specifying each time. "I work in healthcare" means you get HIPAA-aware suggestions. "I'm a journalist" means you get AP style by default.

Stop re-explaining recurring constraints. "I need this accessible to non-technical readers" goes on the card, not in every single prompt.

Create a default template for emails, reports, or brainstorming sessions. Your first draft is now your tenth draft in terms of quality.

Experiment with different instruction sets by saving versions in a document and swapping them when your needs change. Summer camp counselor mode versus grant writing mode.

A freelance designer sets "I work with small business clients who don't know design terminology" and "Explain design concepts in plain language, use metaphors, keep recommendations to 3-5 options max." Every client proposal and design rationale now comes out client-friendly on the first try. No more backtracking to simplify or cutting the list of twelve font options down to something manageable.

TL;DR

• ChatGPT forgets your preferences between conversations unless you use Custom Instructions (found in Settings).
• Box one holds context about your role and audience. Box two specifies tone, format, and output style.
• Instructions apply automatically to all new chats, eliminating the need to retype standard requests.
• Keep instructions general and recurring, not project-specific. Refine them as you notice what you keep explaining.

Now the concierge greets you by name, the towels are waiting, and the thermostat is already set to 67.