Any model, your call.
Write one prompt that works anywhere, and know when to skip AI.
All five parts together/5 min
The same request, twice.
Based on everything you know about me, which phone plan should I get?
I’m choosing a phone plan and want a short comparison I can act on this week. About us: [who is on the plan and how we use our phones]. We live in [city and country] and pay [current monthly cost] now. Give me a table of three plans with the monthly cost, what’s included and the catch, then the one question to ask before switching. If you can search the web, use current prices and link the source for each. If you can’t, say so and mark every number I need to look up myself.
Swap the brackets for your own details.
A prompt that worked beautifully in one app falls flat in another, or works on Monday and shrugs on Friday. Meanwhile someone online is very sure you are using the wrong AI.
You can relax about that last part. In one experiment by Jahani and colleagues, nearly 1,900 people wrote prompts for an image generator, trying to recreate a target picture. A newer model gave better results, and about half of that improvement came from people changing how they prompted. (On an open-ended creative task in the same study, the newer model did most of the work.) The model matters. So does the person typing, and that part comes with you to every app.
Assistants do differ, in ways that are easy to miss. Some remember you between chats and some don’t. Some can search the web, read a file or switch to a slower, deeper thinking mode, depending on the app and the plan. Their default style differs too. So a prompt that travels assumes nothing. It introduces you as if to a stranger, puts anything personal in square brackets you can refill, uses plain sentences with no app commands, and says what shape the answer should take. When it wants a feature, it says so and carries a plan B.
That is the gap between the two prompts on this page. The weak version only works where the assistant already knows you, and even there you can’t see what it thinks it knows. Anywhere else it guesses, with a straight face. The strong version carries its own context in three brackets, asks for a table so the shape holds in every app, and handles web search both ways: link the sources if you can, mark what I need to look up if you can’t. Send it to a friend on a different app and it should still work.
Choose the gear
The prompt is half of the setup. The other half is the switches around the chat box. For a quick question, the fast default is fine, and so is a free plan. For something you will act on, turn on the deeper thinking mode if your app has one. For anything that depends on current facts, such as prices, rules or opening hours, turn on web search and ask for links. If the question is about a document, attach the document. And for a topic you would not want remembered, ChatGPT, Claude and Gemini each offer a temporary or incognito chat that stays out of memory.
Your call
The last choice is whether to ask at all. Some things are worth doing with your own head: the apology, the eulogy, the first attempt at a problem you are trying to learn. When the goal is to learn, the effort is the product, and several apps now have a learning mode built for that. Use the assistant to think better, not to think less.
That is the course. The whole anatomy lives on one page for when you need a reminder, and the library is full of prompts built this way. Pick one, fill in the brackets, and go ask a better question.
Exercise
Your turn.
Write on the blank, then copy the prompt into any assistant. What you type stays on this device.
Here is a prompt I use: . Rewrite it so it works in any AI chat app, including one that has never met me: plain sentences, square brackets around anything personal for me to fill in, no commands that belong to one app, and the length and format of the answer spelled out. Then tell me which features would improve the result (web search, a file upload, a deeper thinking mode) and what to do if my app doesn’t have them.
Check
One question.
You’re sharing a prompt with a friend who uses a different AI app. What makes it most likely to work for them?
Why
Apps differ in memory, features and default style. A prompt that assumes nothing, marks what is personal and says what should come back gives every assistant the same clear job.
Try these.
House prompts that put this lesson to work.
-
AI basics 5 min Make this prompt work in any AI, not just the one I use.
Your prompt rewritten to run in any assistant, with what it still needs and what was cut.
-
AI basics 5 min Should I even be using AI for this?
An honest verdict on one task: hand it over, do it together, or keep it for yourself.
-
Everyday life 5 min I’ve had two tabs open for a week. Which one do I buy?
A side-by-side on the few differences that matter for your use, and a pick with reasons.