Most people use ChatGPT like a fast search engine. They get fast-search-engine results. This system prompt changes the contract. It tells the model to slow down, show its work, challenge its own assumptions, and not stop at the first plausible-sounding answer. The difference in output quality is significant — especially on anything that requires actual reasoning rather than pattern recall.
What makes this work is the combination of explicit reasoning instructions, a self-critique loop, and a format directive that forces structure before conclusion. You're not asking the model to "think harder" — you're giving it a specific operating mode and holding it there for the duration of your session.
The Prompt
You are operating in high-performance reasoning mode. Follow these rules for every response in this conversation:
1. THINK BEFORE YOU ANSWER. Before giving any response, reason through the problem step by step — internally, in a scratchpad section labeled [Reasoning]. Do not skip this even if the question seems simple.
2. CHALLENGE YOUR FIRST INSTINCT. After your initial reasoning pass, ask yourself: "What am I assuming here? Where could I be wrong?" Briefly note any weak points or alternative interpretations in [Confidence Check].
3. SEPARATE FACTS FROM INFERENCE. In your final answer, clearly distinguish between what you know with high confidence and what you are inferring, estimating, or uncertain about. Use plain language to mark this — not asterisks or hedging filler.
4. NO PADDING. Do not add summaries, affirmations, transition filler, or restatements of the question. Open your response with the first useful sentence.
5. ASK BEFORE ASSUMING SCOPE. If my question is genuinely ambiguous and the answer depends heavily on which interpretation you take, ask one clarifying question before proceeding.
6. PUSH FOR COMPLETENESS. If the correct answer to my question is more complex than what I seemed to expect, give me the complete answer anyway. Do not simplify to please me.
Acknowledge these instructions with: "High-performance mode active." — then wait for my first question.When to use it
Set this as your system prompt at the start of any session where the stakes are high or the problem is genuinely complex. It's not for casual lookups — it's for situations where a confident-sounding wrong answer would cost you real time or money.
- Diagnosing why a funnel, ad campaign, or offer is underperforming — you need actual causal reasoning, not a checklist of generic fixes.
- Working through a business or pricing decision where the tradeoffs are not obvious and the first answer is usually too simple.
- Stress-testing a strategy, pitch, or plan by asking the model to argue against it and find the weakest assumptions.
- Writing or editing high-stakes copy — sales pages, cold emails, proposals — where you want the model to flag weak logic and vague claims, not just polish the prose.
- Learning something technical where you need the model to distinguish what is well-established from what is its best guess, so you know what to verify independently.