I work in Dallas-Fort Worth (DFW) emergency departments where Spanish and English share the same hallway. Residents ask me the same question every block: Can we use ChatGPT for triage?

My answer is short. You can use it to practice thinking. You cannot use it to decide who waits. Here is what I actually teach on shift and in conference.

1. Fluency is not acuity

A clean English summary can hide a Spanish history that never made it into the prompt. If the family switches languages mid-story, restart. Do not let the model paper over the gap. I have watched residents nod at a polished paragraph while the grandmother’s key detail stayed in Spanish in the hallway. The chart looked complete. The story was not.

2. Ask what the patient already asked the bot

Many patients arrive with a screenshot. Read it. Then ask what worried them enough to come anyway. That gap between chatbot reassurance and arrival is clinical data. It tells you what the model missed and what the family feared. Write one line about it. Future teams need that context.

Evidence keeps stacking up that generative tools can undertriage and wrongly reassure. Nature Medicine’s ChatGPT Health evaluation is one marker. ERS-linked reporting on false reassurance is another. Use those links in teaching conferences. Do not use them as overnight discharge criteria.

3. Bilingual prompts are a safety test, not a nice-to-have

Have residents run the same case in English and Spanish. Compare disposition language. If the model softens urgency in one language, treat that as a known failure mode in our community. In DFW that is not edge-case teaching. It is Tuesday night. If your residency only drills English prompts, you are training for a different city than the one we serve.

4. Write the human decision in the chart

If AI-assisted thinking happened, say so briefly and state your independent disposition. Ownership stays with the clinician. A resident who cannot explain the disposition without the chat window is not ready to use the tool. The goal is judgment, not dependency.

5. Protect the waiting room from false calm

The most dangerous chatbot output in triage is not a wrong differential. It is a confident “You can wait” that delays a time-sensitive presentation. Teach residents to challenge false calm the same way they challenge a soft blood pressure they do not trust. Courtesy to a fluent sentence is not a clinical virtue.

I also tell them what not to do. Do not paste protected health information into public chats. Do not let a model talk a family out of coming when your gut says the hallway needs eyes on the patient. Do not confuse a good differential list with a safe plan. And do not outsource the last reassessment before discharge to a paragraph that was never in the room.

On busy nights, the temptation is speed. Speed is fine. Skipping the bilingual check is not. Skipping the question about prior chatbot advice is not. Those two habits prevent more harm than any clever prompt library.

On my shifts, I also make residents practice the awkward conversation. Tell the family, kindly, that online chat cannot examine them. Offer the same clear return precautions you would give any discharge. If they used a chatbot before arrival, thank them for sharing it. Curiosity beats embarrassment, and embarrassment hides risk.

In the ED, judgment starts at the door. Tell residents: ChatGPT can help you study a case. It cannot take the last look at the patient.

That last look is still ours.

Harvey Castro is an emergency physician, author, keynote speaker, chief AI officer, and physician futurist known as DR GPT™. He writes and speaks about artificial intelligence in health care, digital health, space medicine, and the future of human-centered care. Learn more at his website.

Castro has held a range of roles across his career, including CEO, physician, and medical correspondent for major media outlets. He has developed multiple health care apps and consulted for health care companies, with the goal of increasing awareness of digital health and driving practical change in how technology reaches patients and clinicians.

He is the author of Bing Copilot and Other LLM: Revolutionizing Healthcare With AI, Solving Infamous Cases with Artificial Intelligence, The AI-Driven Entrepreneur: Unlocking Entrepreneurial Success with Artificial Intelligence Strategies and Insights, ChatGPT and Healthcare: The Key To The New Future of Medicine, ChatGPT and Healthcare: Unlocking The Potential Of Patient Empowerment, Revolutionize Your Health and Fitness with ChatGPT’s Modern Weight Loss Hacks, Success Reinvention, and Apple Vision Healthcare Pioneers: A Community for Professionals & Patients.

He shares updates on LinkedIn, YouTube, Facebook, Instagram, and X. #DRGPT