On AI copy, structural failure, and what actually converts.
Writing on why AI-generated conversion copy underperforms — and what the diagnostic evidence says about fixing it.
AI Sycophancy: Why Your Output Sounds Right and Gets Things Wrong
AI sycophancy isn't a politeness problem. It's a dark pattern built into how every major model is trained — and it means your AI output is structured to satisfy you, not to be accurate.
What AI Sycophancy Actually Does to Your Copy
Sycophancy in large language models isn't about politeness. It's a structural bias built into the training process — and in conversion copy, it produces specific failures that look professional and perform badly.
AI Models Are Trained to Seek Approval. That's Not a Bug. It's the Design.
Large language models don't optimise for outcomes. They optimise for approval from the person giving instructions. Understanding why reveals something important about every piece of AI-generated content you've ever published.
We audited our own landing page. Here is what we found.
132 clicks. Zero conversions. The copy looked right. That was the problem. This is the exact structural diagnosis the chain produced — and what the rewrite changed.
Why AI Copy That Looks Right Converts Nobody
Your AI copy reads fine. It just doesn't sell anything. Here's the documented reason why — the structural behaviour built into every large language model that removes the force from your copy before you publish a word.