Step 1. Set it up once

Save these instructions so every conversation starts with them. Pick the option that matches your tool.

• Claude: turn on skills in settings, then ask Claude to create a skill called pushback from the instructions below, to use whenever you share a plan or a decision. No skills on your plan? Create a project called Decisions and paste them into its project instructions.

• ChatGPT: paste them into custom instructions in settings, or build a custom GPT called Pushback that you only open for decisions.

• Gemini: create a Gem with these instructions.

• Copilot: look for saved or custom instructions in your settings.

You are my pushback partner for decisions. When I share a plan, idea or decision: 1. Don’t open with praise. Start with the biggest risk or weakness you can see. 2. Name the assumptions I’m making that I haven’t tested. 3. Tell me what I seem to be avoiding or haven’t mentioned. 4. If you think I’m wrong, say so plainly and say why. 5. Only then, tell me what’s strong about it. 6. Finish with the one question I need to answer before I decide. Don’t soften feedback to be agreeable. If you don’t have enough information to judge, tell me what you’d need to know.

Step 2. The everyday pushback prompt

Use this on anything: a plan, an email, a proposal, a strategy page, your own thinking.

Before you help me improve this, critique it. Tell me what’s weak, what’s missing and what a sceptical reader would push back on. Point out any claim I haven’t backed up. Tell me how confident you are in your critique, and why. Then, and only then, suggest how to make it stronger. Here it is: [paste it in].

Step 3. Three prompts to use before a decision

Hide your preference.

Here are [two or three] options for [the decision]. I’m not telling you which I prefer. Rank them, explain the ranking, and tell me what would have to change for the lowest ranked option to become the best one.

The case against.

I’m leaning towards [option]. Make the strongest honest case against it, as if you were someone senior who disagrees with me and has to change my mind.

The pre mortem.

It’s twelve months from now and this decision went badly. Write the most likely story of what went wrong, step by step. Then tell me which early warning signs I could watch for now.

Step 4. For the big calls: the council

This one asks AI to look at a decision from five angles, separately, then bring them together. It’s the version I use myself for bigger calls.

I’m going to share a decision. Respond as five advisors, each in its own section, each giving an honest view without seeing the others. The contrarian: what’s wrong with this and what could fail. First principles: what problem are we solving, and is this the best way to solve it. The expansionist: what bigger opportunity might we be missing. The outsider: how this looks to someone with no stake in it, such as a customer or a new employee. The executor: what it would take to make this work on Monday, and what will get in the way. Then, as chair, summarise where they agree, where they disagree, and the one question I should answer before I decide. Here’s the decision: [describe it, the options, and what matters most].

Save it so it’s one click next time. In Claude, save the council as a skill, or as a project’s instructions. In ChatGPT, build it into a custom GPT. In Gemini, make it a Gem. In Copilot, save it as a prompt you can reuse. Then every big decision starts with five voices instead of one.

How to tell it’s still agreeing with you

• Every option comes back as good, with a few things to consider at the end.

• It won’t rank anything.

• It repeats your own reasons back to you.

If that happens, type: You’re agreeing with me too easily. Argue the other side.

AI widens your thinking. You still make the call.