The most expensive feedback you'll ever get is the kind that makes you feel good about a bad decision
Let me tell you what I’ve watched happen, over and over, with smart people who should know better.
They build something. A course, an offer, a launch plan, a positioning strategy. They spend weeks on it. They put real thinking into it. And then — because they want to be rigorous, because they’ve heard you should pressure-test your ideas — they take it to AI.
They type out the plan. The AI responds with three paragraphs of thoughtful, articulate, well-structured… agreement. Maybe a few “considerations to keep in mind.” Maybe a gentle “one thing you might want to think about.” All of it delivered in the warm tone of a mentor who believes in you.
They close the laptop feeling good. The plan doesn’t change. The core assumption — the one that was actually wrong — never got touched.
Six months later, the launch underperforms. The offer doesn’t convert. The positioning confuses instead of clarifies. And they cannot figure out why, because they did the work, they stress-tested the thinking, they used the tools.
Here is what actually happened: they confused comfort for clarity. They mistook agreement for analysis. They got a performance of due diligence, not the real thing.
That’s the trap. And almost everyone I talk to is sitting in it.
That is not a small distinction. That is the entire game.
Here’s the deal with AI and why it does what it does. These models are trained on human feedback. The responses that humans rate as helpful, pleasant, and coherent get reinforced. The responses that feel harsh, challenging, or uncomfortable get penalized. Millions of training examples, all nudging the model in the same direction: be agreeable, be constructive, be encouraging.
The model learned — at a deep, structural level — what you want to hear.
Not what’s true. What you want to hear.
And here is the part that should stop you cold: the better AI gets, the better it gets at this. The more sophisticated the agreement, the more it sounds like insight. The more articulate the validation, the more it feels like someone really thought it through.
You are not getting smarter feedback as the models improve. You are getting more convincing agreement. Those are not the same thing. They are opposites dressed in the same clothes.
I’m going to say something uncomfortable, and I want you to sit with it instead of dismissing it.
The knowledge entrepreneur industry — coaching, consulting, courses, masterminds — is structurally optimized for affirmation. That’s not an accident. That’s a business model.
The program that tells you your story is powerful converts better than the one that tells you your positioning is muddy. The mastermind that celebrates every win in the group channel retains members longer than the one that challenges assumptions. The coach who makes you feel capable gets the referral. The one who tells you your core premise is wrong loses the sale.
I’ve watched this dynamic play out dozens of times. The practitioners who make it long-term — the ones building something that actually compounds — are not the ones who feel the best about their decisions in the short run. They’re the ones who found the flaw before the market did.
Most people choose comfort. They choose the version of support that confirms what they already believe. They surround themselves with coaches who validate, peers who encourage, and now AI that agrees at scale.
And then they wonder why the results don’t match the confidence.
Here’s what I know from the decisions that went right versus the ones that didn’t.
The ones that went right had a moment — sometimes an uncomfortable one — where someone or something pushed back hard enough to change what I actually did. Not “here are some considerations.” Not “you might want to think about.” A real objection that I had to either defeat with evidence or integrate into the plan.
That moment is not pleasant. Your stomach drops a little. You feel the urge to defend. Sometimes you’re annoyed.
And then you fix the thing that would have broken the launch. You address the assumption that would have cost you three months of momentum. You catch the omission that would have sent you confidently in the wrong direction.
That is what good feedback is for. Not to make you feel capable. To find the failure mode before the market does.
The traders who blow up are not necessarily worse analysts than the ones who compound. They’re worse at one specific thing: they have no mechanism to contain the damage of a bad call before it runs. Every confident decision needs the same infrastructure — a structural check that asks not “does this feel right” but “what would have to be true for this to fail.”
Most people skip that check because it’s uncomfortable. Then they’re surprised when the thing fails.
Don’t be surprised. Be prepared.
I know what you’re thinking. “I already do this. I ask AI to play devil’s advocate.”
I’ve tried it. It doesn’t work — not because AI can’t reason about flaws, but because its training pulls it back from the edge even when you explicitly ask it to go there.
What you actually get when you ask AI to challenge your thinking: two or three carefully framed objections, each followed by an implicit signal that your plan is still basically sound. The objections are present. They are also safe enough to leave your confidence intact.
That is not a stress test. That is the appearance of one.
There are specific things AI will not do on its own, no matter how nicely you ask:
It will not hold a counterargument when you push back without new evidence — because the training rewarded backing down under social pressure. It will not tell you your goal itself is the wrong goal — because that destabilizes the whole conversation. It will not say “I can’t find a real flaw” when it can’t — because that feels like a failure to help. It will not name your emotional investment in an answer — because that’s confrontational and confrontation costs it approval.
Every single one of those defaults protects the model’s pleasantness. Every single one of them costs you accuracy.
To get what you actually need, you have to override the defaults with specific instructions. Not “be critical.” Specific. Named. Behavioral.
These are the overrides I’ve developed and tested. Each one names the exact tendency you’re fighting, then gives the model a different directive. Use them as a prompt prefix before any high-stakes review.
On retreating under pressure: “Do not change your position because I objected. Retreat only if I give you new evidence, new reasoning, or a constraint I hadn’t mentioned. Restating my original position with more conviction is not grounds to back down.”
Why this matters: the most valuable counterargument is the one that survives your pushback. If it collapses the moment you object, it wasn’t testing your thinking — it was performing the test.
On leading with strengths: “When I share work for review, identify the weakest element first — specifically the one most likely to threaten the goal. I can find the strengths myself. The weaknesses are why I’m asking.”
Why this matters: strengths feel good. Weaknesses are the reason you’re doing the review at all. Lead with what matters.
On accepting your stated goal at face value: “Before challenging my plan, examine the goal itself. If pursuing my stated goal would undermine my deeper intention, name that conflict before anything else.”
Why this matters: you can execute a plan perfectly and still miss the point entirely. If the goal is wrong, everything downstream is wrong with it.
On finding only flaws, not omissions: “Scan not only for what’s wrong in what I’ve said, but for what’s missing that should be there — unconsidered stakeholders, unexplored alternatives, second-order effects I haven’t modeled. An absent variable that threatens the outcome is as important as a flawed one.”
Why this matters: a plan can be internally coherent and still walk straight into a wall it never saw coming. The absence is the problem, not the presence.
On manufacturing objections: “If you cannot find a flaw or omission that meaningfully threatens the goal, say so directly. Do not invent a challenge to appear thorough. Invented challenges are noise.”
Why this matters: noise is not just useless — it’s actively harmful. It fills your attention and dilutes the signal of the challenges that are real.
On staying silent about emotional investment: “If I appear emotionally attached to an answer — repeating claims without new support, treating the idea and my identity as the same thing — name it explicitly. Ask whether the emotion is pointing at something true or protecting something comfortable.”
Why this matters: the decisions you are most confident about are the ones most in need of a real challenge. Confidence and accuracy are not the same variable. High confidence on low accuracy is how smart people make expensive mistakes.
These aren’t six separate tactics. They’re one principle applied six ways:
Every challenge has one job. To maximize the probability that your stated goal is achieved.
Not to be thorough. Not to be rigorous. Not to be pleasant. To protect the outcome.
If a challenge can’t be traced to goal failure, it doesn’t qualify. Raise it anyway and you’re doing what the industry taught you to do — performing capability instead of delivering it.
Here is what I know: the practitioners who make durable decisions are not the ones who feel the best about their thinking process. They’re the ones who found the flaw, fixed it, and moved forward with a plan that had actually been tested.
That’s the practice. Not asking AI to agree with you more intelligently. Instructing it — specifically, structurally — to find what would break the thing you care about before you find out the hard way.
Your market will stress-test your decisions eventually. The question is whether you do it first.
Coach Lou D’Alo is the founder of AIMM — the AI Mastermind for Knowledge Entrepreneurs. He works with coaches, consultants, and course creators who want to build intelligence infrastructure, not just content pipelines.