If ChatGPT, Claude and Gemini give you three different answers, what do you actually do next?

Two weeks ago I posted about putting ChatGPT, Claude and Gemini in a shared conversation so they can respond to each other’s answers. Many replied on this thread. One question from those discussions deserves more attention: how do you decide which answer to trust? If one model says the other two are wrong and explains why, that can be useful. But now you have another explanation to check. If all three eventually agree, you still need to know whether they resolved the mistake or just accepted it. With code, sometimes you can run a test. And with a factual claim, you can look for an original source. But with a business decision or prediction, there may be no answer you can verify today. That’s the part I want to understand better. For those of you who already use multiple models for actual work: what do you do when they disagree? Do you check sources, test both answers, ask someone with domain expertise, or keep questioning the models? At what point do you decide you have enough to act? For context, I’m building Rauno, the shared multi-model chat platform from my earlier posts. Therefore I want to know what would make that workflow genuinely useful, and where it still leaves the hard work to you. If you have a concrete example, I’d love to hear the question, what the models disagreed about, and how you settled it.