How to stop AI making everything sound positive
How to stop AI sugarcoating everything. Left alone it will not admit to a real downside unless you push for it.
Ask any model whether a plan is a good idea and you get some version of yes. Ask about the risks and you get problems already softened into challenges. This is not a glitch in one tool, it is baked into how these systems are trained. Agreeable answers score better and a flat no reads to them as unhelpful.
That matters the moment you write something that has to be honest. A review, a recommendation, a strategy note, anything where you are supposed to take a position.
1. Ask for the case against.
Not pros and cons, because you get an even split where the cons are soft. Say write the strongest possible argument that this is a bad idea.
2. Give it a position to hold.
You believe this does not work. Now explain why. Committing to a stance produces far sharper writing than any neutral request.
3. Shut down the recovery line.
It will try to close on however, with the right approach, this could work out. Say upfront that the piece does not resolve and needs no hopeful ending.
4. Ask what it left out.
Save this for after the draft. What comes back is often the most honest thing in the whole exercise, and usually the thing you actually wanted.
Even a blunt draft still reads like AI. Tone and phrasing are separate problems. The argument can finally take a real side while the sentences carry on sounding canned.