The most-slept-on feature in AI isn't a secret model. It's the settings box almost nobody fills — the one where you tell your AI, once, how to behave forever. Fill it well and every future conversation starts upgraded. Here are five setups worth stealing, with the exact text to paste and the exact way to undo.
BY FORMEAI · UPDATED AUGUST 2026
Where the magic box lives
ChatGPT: Settings → Personalization → Custom Instructions ("How would you like ChatGPT to respond?"). Claude: Settings → Profile → your preferences box. Gemini: Settings → Saved Info. Copilot: no permanent box on personal accounts yet — paste the setup at the start of a chat instead. Paths verified August 2026 — menus move; search settings for "instructions," "preferences," or "saved info."
One honest mechanic before we start: every setup below comes with two ways out. The permanent undo is always the same — open the settings box and delete the text by hand (no pasted prompt can reach inside a settings box). The quick undo is a prompt you paste: it overrides the behavior in your current chat and clears any saved memory of it. Both are given, every time.
Each setup below goes into that box (or combines with what's already there). One paste, and it applies to every new conversation. Change your mind later? Every setup ends with its undo.
1. The Straight Talker
For everyone tired of three paragraphs of warm-up before the actual answer.
Be direct and concise. Skip compliments, filler phrases, and long introductions. If my idea has problems, say so plainly and explain why. If you are not sure about something, say you are not sure instead of guessing. Short answers first; detail only if I ask.
Permanent undo: open the settings box and delete the setup text. Quick undo — paste this:
Stop applying my earlier instruction about being direct and concise. Return to your default style, and delete any memory you have saved of that instruction.
2. The Email Editor
Turn your AI into permanent staff: every draft you paste gets the same disciplined treatment, no re-explaining.
When I paste an email draft, act as my email editor: flag anything unclear or open to misreading, tighten the wording without losing warmth, and check the email actually asks for what I need. Then give me the revised version. My style: direct but friendly, short paragraphs, no corporate filler.
Permanent undo: delete it from the settings box — or keep it; it only activates when you paste a draft. Quick undo — paste this:
Stop acting as my email editor. Treat pasted emails normally unless I ask, and delete any memory you have saved of that role.
3. The Plain Explainer
For turning any AI into the world's most patient explainer — without the lecture.
When I ask you to explain something, structure it as: the one-sentence answer first, then a short plain-language explanation, then one everyday example. Define any technical term the moment you use it. Never assume I have a technical background.
Permanent undo: delete it from the settings box. (Most people never do — this one tends to stay.) Quick undo — paste this:
Stop applying my earlier instruction about how to structure explanations. Explain things in your default way, and delete any memory you have saved of that instruction.
4. The Honest Critic
The antidote to AI's people-pleasing streak. Your plans deserve pushback before reality provides it.
When I share a plan, idea, or decision, do not simply agree. First list the two strongest objections or risks I might be missing. Then give your overall honest view, including whether you would advise a different approach.
Permanent undo: delete it from the settings box — for example before brainstorming sessions where you want free-flowing energy instead of critique. Quick undo — paste this:
Stop applying my earlier instruction to raise objections before agreeing. Respond normally from now on, and delete any memory you have saved of that instruction.
5. The Context Keeper
Stop introducing yourself in every chat. Give the AI your standing context once — and know exactly how to wipe it.
About me, for context in all conversations: I work as [your role]. My typical tasks are [the two or three things you do most]. I prefer replies in [English / Arabic / both]. My tone preference: [professional / casual / warm]. Do not mention this context back to me — just use it.
Permanent undo: delete the text from the settings box, and also check the memory features: ChatGPT → Settings → Personalization → Memory → Manage; Gemini → Saved Info; Claude → your preferences box. What you tell an AI to remember, you can always tell it to forget. Quick undo — paste this:
Forget the personal context I gave you about my work, my typical tasks, and my preferences. Delete it from your memory and do not use it going forward.
The principle behind all five
A raw AI is a brilliant stranger. A configured AI is a colleague who knows how you like to work. The difference is one settings box and sixty seconds.
Mix them, edit them, make them yours — the wording isn't sacred, the mechanism is. And if you want the full system these setups come from — standing roles, the prompt library, the habit loop — it lives in our guide How Do I Get AI to Do Half My Job?
Privacy note, because we always give one: anything you put in a settings or memory box is stored with your account under that tool's data policy — the same rules from our Privacy Checkup apply. Don't put anything in the box you wouldn't put in a chat.