Adapted from BMJ 7-14 February 2026 Professor Charlotte Blease, Uppsala University, Sweden.
Patients are increasingly disclosing information to artificial intelligence tools (AI) and seeking health advice from them. Large language models such as ChatGTP, Claude and Deep Seek, are being used by millions of people to describe symptoms, seek second opinions, or explore stigmatised topics. Yet many clinicians continue to believe that they bring special empathy to the consultation and that sensitive disclosures to them are irreplaceable by technology.
Even in the 1960s, patients spoke more candidly to non-human interfaces than to clinicians. When it comes to alcohol and tobacco use, suicidal ideation, intimate partner violence, sexual behaviour, and workplace stress, paper based or online questionnaires elicit more information than in-person consultations.
In Denmark, 20% of primary care consultations are conducted by digital messaging. Patients say that they prefer to broach embarrassing topics in writing. As the perceived risk of judgment reduces, disclosure increases.
Unlike humans, digital systems can’t signal disapproval through tone, posture or facial expression. They can’t raise an eyebrow, sigh or signal boredom. They don’t hold any social power over the person confiding in them. For patients who fear being dismissed, blamed or embarrassed, such as adolescents exploring their sexuality, adults hiding alcohol dependence, or older patients reluctant to discuss continence or cognitive decline, digital interfaces feel safer than clinicians.
AI tools can even respond conversationally, offering reassurance, suggestions, or next steps. A recent comparative study across 149 simulated primary care cases revealed that an AI clinical interviewer was felt to be more polite, more attentive, clearer in explanation and better at shared decision making than doctors.
The most valuable aspect of AI consultations for patients is the feeling of freedom for being judged. This fear results in the withholding and distorting of information from doctors by patients. Yet this tendency is rarely acknowledged in medicine.
Clinicians often believe that their empathy can’t be matched by a machine. But machines never get compassion fatigue or even just plain fatigue. They offer a non-judgemental space, at all times.
Yet machines have their risks too. There are worries about data safety and the accuracy of advice given. The advice may not be direct enough and may pander too much to the patient’s presumptions. So, since AI is notably better trusted, more skilled, and usually better informed than the individual doctor, how can such technology be adopted into care systems safely and ethically?
AI disclosure could improve early detection, reduce missed diagnoses, and strengthen shared decision making. If AI took a preliminary history from the patient, this would save the clinician time, give a wider symptom and concerns picture for the doctor to act from, and would save patients discussing distressing details under pressure.
Diabetes Diet 