User simulation matched real human choices 53% of the time. That's the headline result from a 2026 study, summarized on r/artificial, which pushed LLMs through 78 choice tasks drawn from 28 real-world studies and checked whether the simulated preference agreed with the majority human pick. The Reddit