Study (n=504): heightened suspicion did not improve detection of AI-generated text, and fake-news accuracy fell 10.2 points under sustained exposure