How to set up ChatGPT properly (the bit nobody actually does)

Most people use ChatGPT the way you'd use a stranger you sat next to on a flight. Type a question, get an answer, never tell them anything about you, never expect them to remember the last conversation. Then complain that the answer sounds generic.

ChatGPT has three settings that most people have never touched. They take fifteen minutes between them, and they're the difference between an AI that gives you what it gives everyone and an AI that gives you something only you'd have written. Here's where they live and what to put in them.

A note on the setup. AI tools change quickly, so the specific menu paths and character limits in this post are accurate as of May 2026 but will age over time. The principle, configure your AI properly rather than starting from scratch every chat, holds regardless of which version of ChatGPT you're using.

Custom instructions: the part that does most of the work

This is two text boxes inside Settings, Personalisation, Custom Instructions. Each one takes 1,500 characters, which is roughly a paragraph. They apply to every conversation going forward, so you fill them in once and forget about them.

Box one is "what would you like ChatGPT to know about you." Tell it your job, the kind of work you do, the people you work with, and the style you write in. Be specific: "Marketing manager at a fintech, mostly writing internal updates and brand-side content. I work with developers who write in plain English and a CEO who skims everything. I'd rather sound dry than corporate." That's already more useful than what most people type, which is nothing.

Box two is "how would you like ChatGPT to respond." This is the tone box. Tell it what you don't want as much as what you do: "Match my register, which is dry and conversational, never enthusiastic. No exclamation marks. No 'as an AI' disclaimers. No bullet points unless I ask. Short paragraphs. Don't recap what I said before answering." The first time you do this and the response comes back sounding like you, the difference is genuinely jarring. (If you've ever wondered why your AI sounds nothing like you, this is the box that fixes it.)

The mistake is leaving the boxes blank, or filling them with generic professional language ("I'm a busy professional looking for clear, helpful responses"). Generic in equals generic out. The boxes are not asking you to be polite. They're asking you to be specific.

Memory: the part you can ignore until you can't

Memory is what lets ChatGPT remember things across conversations rather than starting from scratch every time. It lives in Settings, Personalisation, Memory.

Two things happen when you turn it on. ChatGPT starts saving facts you mention, your job, your team's names, the project you're working on, your style preferences, and applies them silently in future chats. You can also see what it's saved by asking "what do you remember about me," and if something's wrong or you don't want it remembered, you can delete the entry directly.

Most people don't need to do anything clever with memory. Turn it on, use ChatGPT normally for a week, audit it once, and delete anything weird. That's the whole job.

The reason it matters is that it removes the most annoying friction in working with AI, which is starting from zero every chat. Memory means you don't re-explain that you work in financial services, or that you hate the word "delve," or that your boss only reads the first two lines of any email.

Projects: the part that solves the messy chat problem

Projects are separate workspaces for separate jobs. They're available on the paid plans, and most people still haven't opened one.

Here's the problem they solve. You use ChatGPT for client emails, board reports, holiday planning, and birthday gifts. All of those conversations live in one infinite scroll of chats, and after two weeks you can't find any of them. Worse, the context bleeds across them: ChatGPT starts answering your client emails like it's still planning your holiday.

A project is a folder with its own instructions and its own files. Inside a "client work" project you can drop your tone of voice document, your usual templates, and a one-line instruction about how you want the AI to reply. Every chat inside that project starts pre-loaded with that context, and outside the project, ChatGPT behaves normally.

The rule of thumb is that anything you do more than twice deserves its own project:

• Client work
• Board reporting
• Newsletter writing
• Personal admin, and even something low-stakes like "interesting things to read" can live in its own project so you stop scrolling through old chats to find it

The fifteen-minute setup

In order, what to do this afternoon:

• Open ChatGPT, click your name, then Settings, Personalisation, Custom Instructions. Fill both boxes properly, not generically, and save.
• Stay in Personalisation and open Memory. Toggle it on, then close the menu.
• Go back to the main chat screen and look in the left sidebar for Projects. Create one for the thing you do most often at work, and add a one-line instruction about how you want ChatGPT to behave inside it.

That's the whole setup. Most people will get more out of these three settings than out of any prompt template they've ever read. The reason "AI isn't working for me" is a much more common complaint than it should be is that almost nobody does this part.