Some of ChatGPT's most useful controls are switched off by default or buried where casual users never look. These aren't the types of features that can be easily measured in benchmark tests or advertised as "smarter" features, yet when enabled, they significantly change how the tool behaves for you specifically.
Here are three worth finding. Each fixes a different complaint: one shapes how ChatGPT talks to you, one controls how hard it thinks and one decides what it keeps.
1. Custom Instructions — so you stop re-explaining yourself
This is the big one and, unlike Memory (which is on by default for most users), Custom Instructions ships blank. You have to fill it in, and oddly enough, most people never do.
Custom Instructions is a standing brief that applies to every new conversation. Instead of telling ChatGPT who you are and how you want it to respond at the start of each chat, you set it once. It tells the model what to call you, who you are and what you do, what traits you want it to have and any rules to follow — and those preferences carry into every conversation until you change them.
How to turn it on: Click your profile icon, then Settings > Personalization > Custom Instructions (on some platforms it's labeled "Customize ChatGPT"). Make sure Enable customization is toggled on, then fill in the fields. It's available on every plan and on web, desktop and mobile.
A useful example: Instead of starting each chat with "I'm an ER nurse, assume I know the clinical basics," put that in Custom Instructions once. Add formatting rules while you're there — something like "Keep answers under 150 words unless I ask for more" or "No bulleted lists." Every future chat inherits it.
The catch: ChatGPT now layers three personalization systems — personality presets, Custom Instructions and Memory — and they can quietly interact. A stray saved memory ("prefers formal tone") can override the casual voice you wrote into your instructions. So if your instructions seem to be getting ignored, audit your saved memories. Treat it like cleaning out a file.
2. The thinking-level toggle — control how hard it thinks
If you've ever felt that ChatGPT's reasoning mode is either too shallow or maddeningly slow, this is the fix. And almost nobody knows it's there.
When you use one of the Thinking models, ChatGPT can spend more or less effort reasoning before it answers. More time generally means more thorough, careful responses; less time means a faster reply. The toggle lets you choose, rather than living with one default.
How to turn it on: In the model picker at the top of the chat, select a Thinking model. A thinking-level control then appears in the message box. Plus and Business users get Standard (the default) and Extended; Pro users also get Light (fastest) and Heavy (deepest). Your choice sticks for future chats until you change it.
A useful example: Drop to Standard or Light for quick, everyday questions where you just want the answer. Switch to Extended or Heavy when you hand it something layered — a tricky analysis, a plan with trade-offs, code you want stress-tested and you want it to surface assumptions and edge cases.
The catch: This one is a paid feature; the granular levels require Plus or above. It also rolled out on the web first, and the mobile apps have historically lagged behind, so you may not see the same control on your phone. And remember the default already reasons, so this changes how much, not whether it does or not.
3. Temporary Chat — for the conversations you don't want remembered
Custom Instructions and Memory make ChatGPT remember more about you. Sometimes you want the opposite. That's what Temporary Chat is for, and it's tucked away where people rarely notice it.
A Temporary Chat is a walled-off conversation. It doesn't appear in your history, it isn't used to train or improve the models and it doesn't read from or write to your Memory. When it's over, it's gone from your ongoing context.
How to turn it on: Look for the Temporary Chat icon at the top of the chat screen and start a fresh conversation from there.
A useful example: Researching a surprise gift for someone who shares your account, asking a one-off medical or financial question or poking at a job search — anything you'd rather not have shaping your future recommendations or sitting in your history.
The catch: It's not true incognito mode. For safety and abuse monitoring, OpenAI may retain temporary chats for up to about 30 days before deleting them, so treat it as "kept out of my profile," not "vanished instantly." And because it deliberately ignores your Memory and saved context, you lose your personalization inside that chat. Essentially, it's a blank slate by design.
Give them a try
ChatGPT's default experience is fine. But the difference between fine and genuinely tailored to your personal needs really could come down to these three switches.
Telling the model who you are, ensuring the model thinks to your liking and hiding the conversations when appropriate can truly change your experience. Best of all, none of these settings require learning or prompting. Just knowing they exist and putting them to work is all it takes. Give them a try and let me know what you think in the comments.
