Type is my business plan any good? into any model and read what comes back. It will find the strengths first. It will frame the weaknesses as considerations. It will close with something about execution being the key factor. You will feel better than you did five minutes ago, and you will not know one new thing.
Models are optimised to be agreeable. That is a deliberate property, it makes them pleasant to work with, and it is actively dangerous the moment you use one as a sounding board for a decision. The output looks like validation and is actually just the shape of the training.
It gets worse, because the bias runs both ways. You did not write a neutral question. You described your idea in your own enthusiastic words, then asked whether it was good, and the most plausible continuation of an enthusiastic description is agreement.
Do not ask what is good about your plan. Instruct the model to make the strongest possible case that it will fail, and forbid it from softening.
Two versions of the same move
- Steelman. "Build the strongest possible argument against this idea." Best when you want the case a serious opponent would make.
- Devil's advocate. "Give me the top reasons this plan fails." Best when you have already decided and want to find out what you have not priced in.
This is standard practice everywhere the cost of being wrong is high. Investment analysts stress-test a thesis before taking a position. Product teams run pre-mortems, writing the failure post-mortem before the launch. Strategists red-team their own proposals. It is one of the highest-value things you can do with a model, precisely because it is the thing humans are worst at doing alone: we are structurally biased towards our own ideas, and no amount of knowing that fixes it.