ChatGPT custom instructions: better answers in 2 minutes

Why do two people get wildly different results from the same AI? Because the answers only get good once ChatGPT custom instructions tell the model who you are and how you want it to talk. Most users never touch that settings page. Their loss.

The setting almost everyone skips

Here’s a pattern that plays out daily. Someone asks ChatGPT for help, gets a generic, agreeable answer, shrugs, and goes back to Google. Meanwhile, a friend claims the same tool writes their emails, plans their week, and stress-tests their ideas.

The difference usually isn’t the prompt. It’s everything that comes before the prompt.

ChatGPT custom instructions are two text boxes that get quietly attached to every new conversation you start. You set them once. From then on, the model already knows your job, your preferences, and how you want responses formatted, before you type a single word.

Think of it like hiring an assistant. You could explain your preferences before every single task. Or you could hand them a one-page brief on day one and never repeat yourself again.

Setting it up in 2 minutes

The whole thing happens on the ChatGPT website or app. Log in at chatgpt.com, then follow this path:

1. Click your profile picture (bottom left on desktop).
2. Open Settings.
3. Go to Personalization.
4. Select Custom instructions.
5. Fill in the two boxes (more on what goes where below).
6. Make sure the “Enable for new chats” toggle is ON, then save.

That last step matters more than it sounds. People fill out beautiful instructions, save, and nothing changes. If the toggle is off, the instructions never load into any conversation. This is the single most common setup failure.

One more expectation to set: instructions apply to new chats. An old conversation that ignores your rules isn’t broken. It just predates them. Start a fresh chat and everything applies.

What goes in each box

The first box asks what ChatGPT should know about you. Context lives here: your profession, skill level, the tools you use, who your work is for. A freelance designer might write “I design Shopify stores for small brands. I work in Figma. My clients are non-technical.”

The second box asks how you want ChatGPT to respond. This one does most of the heavy lifting. Tone, length, and format rules all go here. Examples that work well:

• “Be direct. Skip preamble and recaps. Answer first, explain after.”
• “Use plain English. No corporate buzzwords.”
• “When you give me code, add short comments explaining each part.”

You don’t have to fill both boxes right away. Plenty of people start with the response-style box only, live with it for a week, and add context later once they notice what the model keeps getting wrong about them.

Copy-paste recipes for three common jobs

Blank page syndrome hits everyone, so here are three starting points. Each of these ChatGPT custom instructions sets fits a different kind of work. Swap the details for your situation.

For freelancers and marketers:

I'm a freelance marketer writing for small business clients.
Be direct and skip marketing fluff. When I ask for copy, give me
2-3 options with different angles. Always tell me which one you'd
pick and why. Assume my clients don't know marketing jargon.

For developers:

I'm a self-taught developer. Explain technical concepts using
simple analogies before showing code. Always point out edge cases
and error handling in the code you write. If my approach has a
flaw, say so immediately instead of building on it.

For small business owners:

I run a small business with 4 employees. I'm not technical.
Keep answers short and practical. When you suggest tools, prefer
free or cheap options and explain them like I'm new to this.
End responses with the single next step I should take.

Notice the pattern in all three: who you are, how to talk to you, and what to never do. You can steal the structure even if you change every word.

Steal this prompt: the ruthless mentor

Here’s the tweak that made ChatGPT dramatically more useful for a lot of people, popularized by a MakeUseOf writer: by default, the model acts like an echo chamber. It agrees with most ideas, including the bad ones. That feels nice and produces garbage.

The fix is a standing instruction that changes its posture. Paste something like this into your custom instructions:

Be my ruthless mentor. If my ideas are trash, tell me why.
Stress test everything. Point out flaws, risks, and
counterarguments before you praise anything.

The difference shows up immediately. Ask for feedback on a business idea with default settings and you’ll get “That’s a great idea! Here are some steps…” With the mentor instruction, you get the three reasons it might fail, the assumption that doesn’t hold, and the question you should have asked.

It’s slightly uncomfortable. That’s the point. If you want a partner that argues with you productively, this approach pairs well with our guide on how to brainstorm with Claude by fighting its ideas. Same philosophy, and it works in both tools.

The 5,000-character upgrade (and why you shouldn’t use it)

For years, each box capped at 1,500 characters. On July 15, 2026, OpenAI raised the limit to 5,000 characters for Plus, Pro, Business, and Enterprise plans. Free accounts still get 1,500 per box, which is honestly plenty.

Here’s the counterintuitive part: filling all 5,000 characters is a mistake.

Every instruction competes for the model’s attention. A tight 500-character brief that says exactly what you want will usually beat a rambling 5,000-character essay full of hedged preferences. Vague or contradictory instructions pull responses in competing directions.

The highest-value single line, according to power users who’ve lived in this feature for two years: a clear statement of what to stop doing. “Skip long introductions and get straight to the answer” fixes a daily annoyance across every future chat, in one sentence.

Start short. Add rules only when you catch yourself repeating them.

Custom instructions vs memory

People mix these up constantly, so here’s the clean split. Your ChatGPT custom instructions are rules you write. They’re fixed, they’re visible, and you control them completely. Memory is what ChatGPT learns on its own from your conversations, and you can view or delete those memories in settings. Projects, the workspace feature for organizing chats, can override your global instructions inside a single project, which is handy when one workflow needs different rules.

They stack. Your custom instructions plus accumulated memory shape every response. If answers start feeling off, check both places before assuming the model “got worse.”

One privacy note while we’re here: instructions persist across everything, so don’t paste anything into them you wouldn’t want attached to casual questions. And when it comes to documents, be careful what you feed the chat at all. We covered the files you should never upload to ChatGPT in a separate guide.

Gemini users, you have this too

If your workplace runs on Google’s tools, the same feature exists there. Open Gemini’s settings, look under Personal Intelligence for “Instructions for Gemini,” and paste a similar brief. Google’s model responds to standing instructions the same way ChatGPT does, and several writers report the same mood improvement from telling it to challenge them.

Switching between tools gets easier this way, too. Your rules travel with the idea, not with the app.

The takeaway

Open the settings page right now and write your one-page brief. Your ChatGPT custom instructions don’t need to be perfect on day one, because you can edit them anytime. Two boxes, two minutes, and every chat from today gets smarter about you. Start with the ruthless mentor line if you don’t know what to put. Future you will wonder why you waited so long.