ChatGPT Custom Instructions: The 10-Minute Setup That Makes Every Answer Better

Why most people get worse answers than they should

Here is the thing almost nobody tells you about ChatGPT: the reason it keeps giving you bland, generic, slightly-off answers often has less to do with the model than with one simple fact: it knows nothing about you. Every new chat starts from zero. It does not know your job, your industry, your audience, or that you hate corporate filler as much as the next person.

So you end up re-explaining yourself at the start of every conversation. “I work in HR.” “Keep it short.” “Write for a UK audience.” “No jargon.” That repetition is exhausting, and most people quietly stop doing it. Then they wonder why the output feels like it was written for nobody in particular.

Custom instructions fix this in about ten minutes, once, for good. You tell ChatGPT who you are and how you want it to respond, and it remembers that for every future chat. With ChatGPT now serving more than 800 million people every week [4], this is one of the highest-leverage settings most people have never touched.

What custom instructions actually are

In plain English: custom instructions are two text boxes where you save background about yourself and your preferences, and ChatGPT applies them to every conversation automatically. [1] You set them once. You do not paste them again.

Think of it like briefing a new assistant on day one instead of re-briefing them every single morning. The first box answers “what should ChatGPT know about you?” The second answers “how should ChatGPT respond?” That is the whole idea.

They are available on every plan, free included, on web, desktop, and mobile. [1] Each box holds up to 1,500 characters, which is more room than it sounds, so you can be specific without writing an essay. [1]

How to set them up (step by step)

This takes under ten minutes. Here is the exact path on the web app, and it is nearly identical on mobile.

Step 1. Click your name or profile icon in the bottom-left (or top-right on some layouts).

Step 2. Open Settings, then Personalization, then Custom instructions. [1]

Step 3. You will see the two boxes. Fill in the first with who you are. Fill in the second with how you want answers delivered.

Step 4. Save. That is it. Open a brand new chat and the instructions are already working in the background.

You can edit or delete them any time, and changes only affect future chats, not old ones. [1] So there is no risk in experimenting. If a setting backfires, change one line and move on.

What to put in box one: about you

The first box is context. The goal is to give ChatGPT the facts it would need to tailor an answer to you specifically. Be concrete. “I work in marketing” is weak. “I am a marketing manager at a B2B software company, my audience is IT directors, and I write mostly LinkedIn posts and email campaigns” is strong, because every word changes the output.

Good things to include: your role and industry, who your audience is, the tools or platforms you use, your location (for spelling and tone), and anything ChatGPT keeps getting wrong. If it always assumes you are American and you are not, say so once here.

You do not need a technical background to do any of this, and you do not need a technical background to get real value from AI. This is just describing your own job in plain language.

What to put in box two: how to respond

The second box is where the magic actually happens, because it controls tone, length, and format on every answer. This is where you kill the generic feel for good.

Be specific about what you want and what you do not. My own version of this box tells ChatGPT to be direct, skip the throat-clearing introductions, use British spelling, avoid corporate buzzwords, and give me the answer first and the explanation second. Yours will be different, and that is the point.

Useful instructions to consider: preferred answer length (“keep responses concise unless I ask for depth”), format (“use short paragraphs, not walls of text”), tone (“conversational, not formal”), and honesty (“if you are not sure, say so instead of guessing”). That last one tends to reduce confident, wrong answers.

Copy-and-paste examples by role

Here are starting points for three common roles. Paste the relevant one, then tweak it to fit you. These pair well with the 4-part prompt formula, which handles the per-task instructions on top of this permanent context.

For an HR or people ops manager. Box one: “I am an HR manager at a mid-sized company. My audience is employees and senior leadership. I write policies, job descriptions, and internal comms.” Box two: “Write in clear, warm, professional language. Keep it concise. Flag anything that could be a legal or compliance risk so I can check it with our advisor.”

For a marketer. Box one: “I am a marketing manager in B2B SaaS. My audience is mid-level decision-makers. I produce email, social, and landing-page copy.” Box two: “Be punchy and specific. No clichés. Give me three options when I ask for copy. Match a confident, friendly brand voice.” This stacks neatly with our guide to using ChatGPT for marketing.

For a finance or operations lead. Box one: “I work in finance operations. I analyse data, write reports for non-finance stakeholders, and build processes.” Box two: “Translate financial detail into plain English for non-experts. Always show your reasoning for any calculation. Be precise with numbers and flag assumptions.”

The mistakes that make custom instructions backfire

Custom instructions are powerful, which means a bad setting affects everything, not just one chat. A few honest warnings.

First, do not overstuff them. If you cram in twenty rules, ChatGPT starts ignoring some of them or applies them awkwardly to tasks where they do not fit. Keep each box tight and high-value.

Second, watch for instructions that fight your actual requests. If box two says “always keep it under 100 words” and you then ask for a detailed 2,000-word brief, you have created a tug-of-war. Phrase rules as defaults you can override: “keep responses concise unless I ask for depth.”

Third, remember they apply everywhere. A playful, emoji-heavy tone might be great for social drafts and terrible for the board report you write in the same account. If your work spans very different modes, that is exactly what Projects are for, which we will get to next.

Custom instructions vs Projects vs memory

People mix these up, so here is the clean version. Custom instructions are your account-wide defaults: they apply to every chat. Projects are folders with their own separate instructions and files, ideal when you want one set of rules for client work and another for internal work. Memory is ChatGPT quietly remembering details across chats over time.

Use custom instructions for the things that are always true about you. Use Projects when a specific stream of work needs its own context, and if you also use Claude, the same logic applies to Claude Projects. Let memory build up naturally and prune it occasionally.

The honest take: most people only need custom instructions to get the bulk of the benefit. Set those first. Add Projects later if and when you feel your single set of defaults pulling in two directions.

What to do this week

Block ten minutes today. Open Settings, then Personalization, then Custom instructions. Write three or four honest sentences in box one about your role and audience. Write four or five preference lines in box two using the examples above as a starting point.

Then run your next five real ChatGPT tasks without re-explaining yourself, and notice how much less you have to correct. That gap, the stuff you are no longer repeating, is the time custom instructions just gave back to you.

One small habit: when you catch yourself typing the same instruction for the third time in different chats, that is a signal. Move it into your custom instructions and never type it again.

Related reading

ChatGPT Prompts for Managers: 11 That Save Real Time Every WeekChatGPT vs Gemini for Marketing: Which Should Your Team Use in 2026?How to Use ChatGPT for Financial Analysis (Without Being a Finance Expert)

Sana Mian — Co-Founder, Future Factors AI

Sana is an AI educator and learning designer specialising in making complex ideas stick for non-technical professionals. She has trained 2,000+ learners across corporate teams, bootcamps, and keynote stages. Future Factors offers AI Bootcamps, Corporate Workshops, and Speaking & Consulting for teams ready to move from AI-curious to AI-confident.

More about Sana →