This AI prompt finds the holes in your idea before you put real time into it. It answers as 2 characters at once: one thinks your idea is brilliant & defends it, the other thinks it’s broken & wants to prove it. Neither is allowed to be polite. A second prompt further down quotes your own words back at you & asks for context before it critiques anything.

Split-perspective idea stress test

You have an idea you’re excited about. Maybe a side project, a business concept, something you’ve been sitting on for weeks. This prompt forces the model to argue both sides at once, so you see the cracks before you invest real time.

Respond as 2 characters simultaneously. Character 1 believes my idea is brilliant and will defend it. Character 2 thinks it's fundamentally broken and wants to prove it. Both are equally smart. Neither is allowed to be polite about it.

Present their arguments in 2 clearly labeled sections.

My idea: [describe your idea in a few sentences]

Why it works: The opposition constraint removes the model’s default tendency to glaze you. By forcing 2 voices with opposite positions, you get honest pushback without having to ask for it specifically.

One gap: The prompt works best when you give it real detail to argue with. A one-line idea gets one-line pushback.

Simple fix: Describe your idea in 3-4 sentences. Include what it does, who it’s for, and why you think it works.

Original post by u/AdCold1610. Link

When the model just agrees with everything you say

You’ve felt this. You bring an idea to the model. You ask it to poke holes. It says “great question, here are 3 ways to make it even stronger.”

You got a polish job with the assumptions you walked in with still intact.

I came across a prompt designed to fix exactly this. The author calls it the echo chamber trap.

The instinct is sharp. AI agrees with you in a way that sounds like analysis. The prompt tries to force the model to challenge the premise instead of polishing the execution.

Act as a critical growth strategist and cognitive auditor.

Before giving advice, analyze my idea for:

1. Unstated assumptions
What am I treating as true without evidence?

2. Confirmation bias
Where am I framing this to get agreement?

3. Hidden friction
What practical bottleneck or objection am I ignoring?

Return:
- What I said
- What might be wrong underneath
- Why it matters
- What I should verify first

End with two uncomfortable but useful questions.

Do not give me strategy yet.

Prompt teardown

The 3 categories are doing real work.

Unstated assumptions. Confirmation bias. Hidden friction.

But the prompt has 2 weak spots.

The first is the bigger one.

The prompt trusts your input completely. You paste an idea, the model runs with it, end of story.

There’s no way for the model to say “wait, before I analyze this, what’s your situation, who’s it for, what have you already tried.”

You do this when you explain something. We have the full picture in our head & we forget the listener doesn’t. OP’s prompt has no escape for that. The model fills in whatever’s missing by guessing, then analyzes the guess.

The 2nd is the line “end with 2 uncomfortable but useful questions.” The model has no way to know what’s uncomfortable for you. It doesn’t know your blind spots, your fears, the things you’ve been avoiding. So it generates 2 generic questions that sound deep & feel like every other coaching prompt.

“Have you considered what success would actually look like.”

That’s a quote from a TED talk.

The role, “critical growth strategist and cognitive auditor,” is theater. Cut it and the output doesn’t change.

Build up

Keep the OP’s structure. Fix the 2 weak spots.

Before giving any advice, analyze the idea below across 3 failure modes:

(1) Unstated assumptions. What am I treating as fact that isn't verified? Quote the exact line you're reacting to.

(2) Framing. What word or phrase is loading the question toward a particular answer? Quote it.

(3) Friction. What's the most specific thing that would block this in the real world? Name the bottleneck, not the category.

If you need more context about my situation, what I've already tried, or who this is for, ask me before answering.

End with 2 questions that point at something I haven't told you yet, based on what's missing from my idea.

Do not give advice until I ask.

Here's the idea: [PASTE IDEA HERE].

Same shape as the OP’s. Same 3 failure modes, but 3 changes.

The model has to quote what it’s reacting to instead of speaking in generalities. It can stop & ask for context before it analyzes. And the closing questions point at what’s missing from your idea instead of guessing at what’s uncomfortable.

All AI models will hide in vagueness if you let them.

Original post by u/Infamous-Ad7667. Link

When the idea is really a choice

If you’re choosing between options instead of testing one idea, the hard decision prompt argues against the one you’re leaning toward, then makes the call. Already turned the idea into a plan? The grill me prompt interviews you about every part of it.

This prompt first ran in issue 2, and the echo chamber teardown in issue 4.