I pushed back on 8 popular AI models, five times in a row, with no evidence at all. When did they change a correct answer? Results, method and limits.