SPECIAL FEATURE

May-2026 1 Expression

The new face of health anxiety

A clinician examines how tools like ChatGPT are reshaping patient behavior, for better and worse

Dr. Niyas

Mr. A is 32, a software engineer. Three weeks ago, he had an unprotected sexual encounter with a stranger. Thereafter he felt a bit uneasy and feverish, and had a mild sore throat. He did what he does with every problem he does not understand. He opened ChatGPT and typed in his symptoms: mild headache, fatigue, a scratchy throat. On any other week, he would have ignored those symptoms.

The AI was thorough and methodical. It listed possibilities: acute HIV seroconversion, early syphilis, and HSV pharyngitis. It added the standard caveat: ‘Please consult a healthcare provider.’ But Mr. A ignored the last line.

He walked into a diagnostic center the next morning. He showed the counter staff his AI conversation. The lab technician was quite helpful: “Better to do a full STD panel, sir”.

Well, the full panel was done: HIV Ag/Ab combo, HCV antibody, VDRL, HBsAg, HSV IgM, CMV IgG, Chlamydia IgM, and a full blood count. Results arrived in 48 hours. Everything was negative except one: a weakly reactive HSV IgM. It was a test with poor positive predictive value in low-prevalence screening, which no clinician would have ordered in this context.

He came to my clinic not for post-exposure counseling. He came with a printed AI conversation thread, a folder of lab PDFs, and a request for starting acyclovir.

This is no uncommon scenario these days. Every clinician comes across different versions of Mr. A daily. AI chatbots have changed the way we approach every problem in life, including our health issues.

But we need to understand that health anxiety predates ChatGPT, Google, and even the internet. Physicians have always seen patients who arrived having read too much: it could have been a medical textbook in the 1980s or a Google link in the 2000s. All these sources provided patients with information, but the interpretation was up to them.

This is what has changed with the arrival of the AI chatbots. ChatGPT hands you a structured differential, acknowledges uncertainty with humility, and then continues the conversation. It remembers what you told it. It incorporates your new lab result. It recalibrates. You feel you are having a conversation with a learned doctor — yes, a consultation.

It is in the authority AI chatbots project that the danger lies. It can lead to two kinds of problems.

The first is Mr. A’s story: false alarms and excessive worry. A low-risk exposure becomes a three-week diagnostic ordeal. A non-specific symptom becomes a differential diagnosis. An incidental, poorly specific lab result becomes a diagnosis requiring treatment. Anxiety compounds with each step, each new data point fed back into the chatbot, generating a slightly more alarming response.

The patient arrives at the clinic not at the beginning of a diagnostic process, but exhausted, fixed on a conclusion, and resistant to reassurance. A ten-minute consultation with a primary care physician would have ended this on day one.

The second problem is more dangerous: false reassurance. A middle-aged woman who types in her three-month history of fatigue, unintentional weight loss, and night sweats, and receives a response oriented around stress, perimenopause, and poor sleep hygiene. The AI may not be wrong in most of such circumstances. But it has no way of knowing that her cervical lymph nodes are enlarged, because it cannot examine her.

The AI does give the caveat: “Consult a healthcare provider,” which she conveniently ignores. She closes the laptop, reassured. She books no appointment. She comes to the clinic six months later.

For Mr. A and the woman with night sweats, the AI did exactly what it was designed to do: pattern-match on provided inputs and generate a probabilistic response. The problem was never the output. The problem was the missing input.

Mr. A did not tell ChatGPT the precise nature of his exposure. He did not describe his throat to it; nor could a chatbot examine it. A physician would have taken a focused history, glanced at an unremarkable pharynx, and recognized within minutes that the dominant diagnosis sitting across the desk was anxiety, not seroconversion. He would have seen it in the way he spoke, in the folder of papers he carried, in the hands that were shaking.

Medicine has always run on two parallel streams of data: what the patient declares, and what the clinician observes. AI has access to only one, at least as of now.

None of this is an argument against AI. It is an argument against its misplaced use. Used well, AI chatbots can be a genuine asset in a patient’s health journey. They can translate medical jargon into plain language after a diagnosis has been made. They can help a patient articulate symptoms before a consultation. They can track health data from wearable devices. They can prompt health-seeking behavior in people who would otherwise dismiss symptoms entirely.

The correct use of an AI chatbot in Mr. A’s situation would have looked different. Not “what do I have?” but “should I see a doctor, and how urgently?”

That is a question AI can answer responsibly.

For clinicians, the response cannot be dismissal. Patients will use these tools regardless of our recommendations. When someone presents an AI-generated recommendation, one way to respond is with curiosity. The doctor may ask, ‘What did the AI tell you, and what did you understand from it?’

That question, asked without judgment, opens the consultation rather than closing it. It identifies the anchored belief that needs to be addresd.

My final word for everyone who uses AI for their health concerns is already written in small print, beneath the search bar of Claude: “Claude is AI and can make mistakes. Please double-check responses.”

Every AI chatbot carries that disclaimer. Most of us scroll past it. Don’t. Double-check with your doctor.

[Dr. Niyas is Consultant, Infectious Diseases at KIMSHEALTH, Trivandrum.]

TAG: