Davis Austria, an emergency and critical care nurse, describes being handed the National Comprehensive Cancer Network's clinician-facing treatment guidelines after her own head and neck cancer diagnosis. The document was accurate and evidence-based, but written for clinicians, not for her. Her real questions—about what swallowing should feel like, whether her hearing would recover, how to eat after a feeding tube—went unanswered. She eventually turned to a Facebook group for advice from strangers.
Austria notes that AI is now being used to rewrite patient information, and she sees both promise and danger. AI can quickly turn scattered medical knowledge into plain language for a patient awake at 2 a.m., but it can also be wrong or confidently misleading. She cites a small German trial where patients who received an AI-generated, patient-oriented discharge summary scored higher on patient activation than those given the standard physician-written version.
The author argues that health systems must ask whether patient materials work for the people who actually receive them, across language, age, and circumstance. Medical accuracy alone is not enough; real patients should test whether they can understand and act on the information. As she puts it, the people best positioned to judge a discharge instruction are the patients holding it at 2 a.m. with no one to call.