The Reality Check Prompt

Ask AI for advice on a hard situation and it hands you agreement dressed as advice. Telling it to "be more direct" does not fix this. It needs a structure that forces the model to build both sides of the case before it hands you a verdict.

We built this exact prompt to pressure-test real decisions: a hard conversation with a partner, a co-founder, or a client. It works because it separates what actually happened from the story you are telling yourself about it. Steal the structure below.

Part 1: Separate Fact From Story

Every hard situation is really two things stacked together: what happened, and what you decided it means. Most advice, from AI or from a friend, answers the story. This prompt makes the model answer the fact first.

• State the objective summary: ask for what happened with zero interpretation layered in, before anything else gets discussed.
• Name your story separately: your read on why it happened gets its own labeled section, never blended into the facts.
• Give the other side a fair read: whoever else is involved gets their most plausible interpretation stated, with a confidence level attached, not an excuse.
• No guessing when data is missing: anything you have not said gets marked unknown. A good prompt refuses to fill gaps with assumptions.

Part 2: Force It To Argue Both Sides

A model that only validates your first read is a mirror, not an advisor. This part makes it build the strongest case against your conclusion, then the strongest case for it, before you get a verdict.

• Counterargument first: ask it to argue against your current read, as if it were debating you. This alone helps catch catastrophizing before it runs the show.
• Then the strongest case for you: have it flip and argue why your original read could be right. If it cannot make that case well, your read was probably weak to start with.
• No default side-taking: tell it directly, never default to agreeing with me. State this once and it holds for the whole conversation.
• Update, do not defend: if you add new information mid-conversation, the model should change its read, not defend what it said five minutes ago.

Part 3: Score The Evidence Like A Court, Not A Confidant

Feelings are real. They are not evidence. This part makes the model sort what you actually know from what you are assuming, then puts a number on how sure anyone should be.

• Three columns, not one list: evidence supporting your read, evidence against it, and what remains unknown. Unknowns never get filled with a guess.
• Red flags versus a bad day: one incident is not a pattern. Make it separate one-time events from things that keep repeating before it calls anything a red flag.
• Green flags get equal airtime: ask for the signs of good faith too, like ownership, repair attempts, and honesty. A one-sided report is not a reality check.
• Put a percentage on it: ask for a probability your read is correct, a probability the other explanation is correct, and a probability nobody has enough information yet. Numbers force honesty that words let you dodge.

Part 4: Get The Exact Move, Not A Menu Of Options

Ten maybes is not a decision, it is homework. Close every reality check with one move and the literal words to say.

• Name the move, not the options: reacting, over-functioning, and a boundary are different things. Make it name which one this calls for.
• Ask for the exact words: not a summary of what to say, the actual sentences. Calm and direct, never a script built to manipulate.
• Demand the blind spot: the last question, every time, is what am I not seeing here? This catches what your emotions are hiding from you.
• Save the prompt, reuse it: works on a hard call with a partner, co-founder, or client. Save it once, reuse it on every decision that has you spinning.

The 60-Second Reality Check

Run this before you trust any AI read on a hard situation. Two misses means run the prompt again.

• I got the objective summary before any interpretation.
• My read and the other side's read are in separate sections.
• It argued against my conclusion, not just for it.
• Evidence is split into for, against, and unknown.
• No unknown got filled in with a guess.
• It gave me a percentage, not just a feeling.
• The recommended move has a name, not a menu.
• I have the exact words, not a summary.
• It told me my blind spot.
• I would use this exact structure on my next hard call.

Ready to Go Deeper?

This prompt gets you a straight answer. The certification gets you a system that runs your week.

In the 12-Week AI Certification we sit with you live and build a working team of AI helpers around your actual business, your real decisions, your real week. You leave with helpers that keep working after the call ends, and a credential to show for it. No tech skills needed.

The model argues both sides. You still make the call.