What Ambient AI Scribes Cannot Hear

Ambient AI scribes are being deployed across health systems to reduce documentation burden and clinician burnout, with encouraging early results. However, because these tools generate clinical notes from speech, they capture what is said during the encounter, not what clinicians observe without verbalizing or judge without articulating. This Perspective argues that this omission is not just a workflow limitation but an epistemological one. Drawing on the dual-process model of clinical reasoning, research on clinician intuition, and the verbal overshadowing literature from cognitive psychology, it suggests that audio-based documentation may create a structural incentive toward verbalization that risks interfering with tacit, nonanalytical expert judgment. The downstream consequences include directional distortion of the clinical record, reduced detection of premature diagnostic closure, and the potential erosion of trainees' appreciation for preverbal expertise. Health systems should therefore evaluate ambient scribes for reasoning fidelity rather than transcription accuracy alone, and medical educators should teach the distinction between tacit judgment and bias. Ambient scribes solve a real problem, but documentation systems built around what can be heard may gradually devalue clinical reasoning that cannot be spoken.