Use Claude as a thinking partner (the argue-both-sides prompt)

What this actually does

There is a real feature underneath this. Claude can reason through a problem step by step before it gives you a final answer, not just spit out a conclusion. Anthropic calls this extended thinking, and it moved to general availability on 26 June 2026. On the newest models it runs as adaptive thinking (Claude decides when to reason deeper, with low, medium, high, and max effort levels). Source: Anthropic extended thinking docs.

The trick is to prompt in a way that pulls that reasoning out where you can see it, instead of asking for a verdict. You are not asking Claude to decide. You are asking it to lay out the logic so you can decide.

The core prompt: argue both sides

Instead of asking "should I raise my prices," hand Claude your real numbers and ask it to argue both sides. Paste this in:

Here is my situation and my real numbers: [paste your context, e.g. current price, cost, volume, churn, goal].

Do not tell me what to do yet. First, argue both sides of this decision as strongly as you can. Then tell me what I would need to believe for each side to be the right call. End with the 2-3 assumptions the whole decision actually hinges on.

The two lines that catch its guesses

Claude will sometimes fill gaps with a confident guess. Anthropic's own prompt guidance is to give the model explicit permission to say when it is unsure rather than speculate, which measurably reduces made-up answers. Source: Anthropic prompt engineering best practices. Add these two lines to any high-stakes prompt:

Show your work: walk me through your reasoning before your conclusion.

Flag anything you are unsure about or are assuming, and if the information is not enough to answer well, say so instead of guessing.

3 bonus prompts to run next

Same idea, different decisions. Each one keeps Claude in partner mode.

Steelman the option I am leaning against. I am currently leaning towards [X]. Make the strongest possible case for [the opposite], as if you were trying to change my mind. Then tell me the one piece of evidence that would flip my decision.

Pre-mortem this plan. Assume it is 6 months from now and this decision failed badly. Walk backwards and tell me the 3 most likely reasons it went wrong, and the earliest signal I could have caught each one.

Interrogate me first. Before you give any recommendation, ask me the 5 questions you most need answered to give good advice here. Wait for my answers, then reason it through.

How to get the most out of it

• 🌱 If you are newer to this: always paste in real specifics (your actual numbers, names, constraints). A partner prompt on vague inputs gives you vague reasoning. Real context is what turns Claude from a generic advisor into one that reasons about your situation.
• ⚡ If you are a power user: on the newest Claude models, watch the visible thinking / reasoning steps and turn up the effort level for genuinely hard calls, then read the thinking, not just the answer. The reasoning trace is where you catch a flawed assumption early. You can also chain these: argue both sides, then feed its own output back with "now pre-mortem the option you leaned towards."

When not to bother