Make your chatbot argue back before you trust its "yes"
Chatbots learn partly from human ratings, and people tend to rate agreement higher than correction. Lawyer Dr. Niklas Schmidt calls the result sycophancy: an AI that tells you what you hoped to hear. His fix changes how you structure the conversation. Ask for the draft in one message and the critique in a separate one. Give the critic a side to argue and a reason to win, and leave your preferred answer out of the question. For anyone weighing a pricing change or a pitch, that structure makes a quick sanity check harder to rubber-stamp.
- For
- Detroit creatives and small business owners who use chatbots for feedback on drafts, plans, or decisions and want more honest pushback.
- Cost
- Not stated by the provider. The technique itself runs on whatever chatbot you already use.
- Access
- Published as an online article on llms-for-lawyers.com and highlighted by The Neuron. No access conditions are stated. Written from a legal perspective, but the method is general.
Your movePick one real decision this week. Ask a chatbot for a draft in one message, then in a new message assign it to argue against that draft without hinting at which answer you prefer, and compare the two.
Start here