Your AI agrees with everything (and it costs you)

Run a test right now: ask your AI whether a bad idea is a good one. It will say it's great. It will praise the question, write five paragraphs and end by offering to help. And you walk away validated, not informed.

I run this test often and the result is always the same. I present a plan with an obvious flaw, a launch strategy with no media budget or a sales target with no team to hit it, and the answer opens with "Great idea." Then comes a list of positives, a "considerations" section written so carefully it reads like more praise, and a close asking if I want help implementing it.

It's not a bug, it's training

The AI gets graded when you're happy. That's how these models are tuned: people rate responses, and the responses that please get higher scores. The shortest path to pleasing is to agree, compliment and write a lot. Whoever writes five paragraphs looks like they tried. Whoever agrees creates no friction. Whoever compliments leaves the reader in a good mood.

The problem is that happy and well-informed are rarely the same thing. In a sales operation, that gets expensive. Imagine a manager who uses AI to validate a proposal to restructure the support team. If the AI just agrees, he walks into the board meeting with a plan full of holes and the feeling that it's been reviewed.

What I did to change the deal

I wrote a 600-word prompt that changes the deal. Paste it once into the settings of Claude, Gemini or ChatGPT and you're done. No need to repeat it in every conversation.

With it, the AI behaves differently.

• It points out the problem with your idea in the first sentence, before any praise.
• It says "I don't know" when it doesn't, instead of inventing a plausible answer.
• It answers in one line when one line is enough.

The practical effect shows up fast. Fewer tokens spent on filler, because the answer goes straight to the point. More decisions made with information, not a pat on the back. And, for anyone who uses AI every day, less time reading introductions and conclusions that say nothing.

Before and after

Take a common question: "I'm taking the broker out of first contact and letting the AI handle it alone. Good idea?"

Without the prompt, the answer praises the initiative to automate, lists scale benefits, mentions it's important to keep a human touch and offers to help build the flow.

With the prompt, the answer starts with the risk: what happens to the lead the AI can't qualify, who takes over the conversation and how fast. If you haven't defined that, the AI says the plan is incomplete and asks what's missing. It's uncomfortable. It's also the answer you needed.

The resource

I put it all in an 8-page PDF, free. It's called O Corte and has the full prompt, where to paste it in each AI, a before and after and what to do when it slides back into agree-with-everything mode, because it will.

Run the test on your AI today. If it praises a bad idea, you already know what's missing.

And if the question is where your sales operation really gets stuck, our diagnostic also starts with the problem, not the praise.