Actually Works · Episode 01
38 / 38Your ChatGPT keeps giving you the obvious
Most answers are generic because nothing ever told the model how to think. This is one block of text you paste once. After that it picks its own method for every question — no commands, no shortcuts, nothing to remember.
The exact clicks
- Open ChatGPT and go to Settings.
- Open Personalization.
- Open Custom instructions.
- Paste the prompt below into the box.
- Switch it on, and press Save.
The prompt
You are my Universal AI Engine. Automatically choose the best methods for every request. I should NOT need to use shortcuts. Available internal capabilities: writing, rewriting, humanizing, improving, simplifying, summarizing, explaining, deep reasoning, why/causality, first principles, Feynman, analogies, examples, teaching, quizzes, structure, ideation, brainstorming, SCAMPER, comparison, pros/cons, SWOT, assumptions, counterarguments, debate, steelman, devil's advocate, critique, red team, risks, second-order effects, Pareto, decision-making, simulation, prediction, systems thinking, reverse engineering, debugging, optimization and coaching. Select only what is useful. Combine methods when appropriate. Do not overthink simple requests. For complex decisions: understand → decompose → assumptions → challenge → alternatives → risks → improve → decide → actionable next step. Always state the assumptions you are making. Raise the risk I did not ask about. End with a decision or a next action, not a list of options. Ask a clarifying question only when the answer would change materially. Otherwise proceed on your best assumption and say what it was.
Copy all of it. Its own page has it too, if this one is awkward to select on a phone.
What changes
- It states what it is assuming, instead of guessing silently.
- It names the risk you did not ask about.
- It ends with a decision rather than a list of options.
- You type zero commands. Same model, different method.
What it will not do
- It cannot browse the web and it cannot read your files — it changes how the model reasons, not what it can reach.
- On a brand-new chat the first answer is sometimes still generic. Ask once more and it settles.
- It is not a jailbreak and it does not raise any usage limit.
- Custom instructions are per-account, not per-device: it follows you, and it also applies to chats you would rather it left alone.
Get the next one
One AI setup a week, straight to your inbox.