Weird Time to be Human: AI Medical Advice

TL;DR: Please don't go to your AI platform for an answer to your health problems. You don't know what you don't know. Your medical provider should be trained to find out/explain to you what you don't know. Treatment should be personalized and capture the nuance our quirky lil human lives.

---

A patient came to me at the outpatient physical therapy clinic with pages upon pages of her AI generated recommendations for back pain. "Do you not like cat-cow? Why do we not do it? I think I should be doing it," she has new confidence in what might solve her low back pain after a few weeks of stubborn symptoms.

"No, I don't hate cat-cows. We tried those during the first week of treatment. You said your shoulder hurts when you're putting weight on it."

My response is not sufficient. We spend the next ten minutes attempting some movements in quadruped, that is, on hands and knees. But the positions, on various surfaces, were unkind to this patient's shoulders and wrists.

"What about clamshells? I think I should be doing clamshells," she continues to leverage this AI generated list of solutions, which are not wrong, but also not qualified medical advice.

"I don't hate clamshells either. They'll strengthen or activate the glutes, and that's important. But you're very active. You've biked nearly 30 miles today, you go to Pilates, you walk all the time, you're far more active and I'd like to do things that are more specific to your activity level and everything you want to get back to doing," I try to meet her where she is.

My response is not satisfactory. We do clamshells next.

I finished that day of work questioning where I had failed this patient. We had discussed her injury, the movement modifications that could help with recovery, exercises to do at home, the timeline and expected milestones that may indicate progress... we collaborated on so many components of her treatment.

Objectively, she was doing better than the first day of treatment. She was back to biking for two hours a day, and becoming more active with fewer symptoms. Still, as her physical therapist, I felt like I let her down.

---

In the last few months, I have had a handful of patients fact-check my prescribed exercise and interventions, or bring in print outs of ChatGPT responses to prompts they've generated. In those cases, most of them knew firsthand that I prioritize a collaborative space and want to make myself available to them for this -- to discern useful information from unrelated noise. Those interactions had been fruitful, and from their direct feedback, helpful in making sense of their treatment plan or creating greater understanding.

Use of AI in seeking medical advice is missing is just that: discernment and direction. Artificial intelligence is savvy at collecting an enormous amount of data and pattern recognition. The description it can generate for degenerative disc disease, a general treatment plan or course of action, is not necessarily incorrect. But it is generalized, non-specific, and not weighing high quality research over more popular articles or posts.

More and more, I find educated people turning to AI to solve their wide array of problems. I see perfectly capable people turn to ChatGPT for answers in place of their medical providers. In creating an accessible powerful machine that can scour the internet for answers, so that we maybe no longer "Google it" into a rabbit hole of "I'm dying, it's cancer", we've forgotten the role of the provider. Or maybe medical professionals, myself included, are not stepping up to the plate. Is it the user and AI, or the lack of trust in the medical field? Is it C - all of the above?

No matter how you look at it: AI is not meant to do the job of doctors.

When my patient typed her MRI findings into ChatGPT, she neglected to share the pertinent pieces of information that could have further streamlined the response from her prompts. Even if she had typed her medical history, AI can only get incrementally closer to a well suited program for her to follow. At the end of the day, the nuance of each patient, each person, is missed.

On May 25, 2026, Pope Leo XIV warned against the ethical risks of AI (while not so secretly quoting Tolkien).

As a society, we have gotten ignorantly bold at the use of AI. Again, myself included. And I have sat by as I watched peers use it for their mental health solution and as a medical advisor.

When we turn to trust our own prompts to leverage this unimaginable amount of data, we are inadvertently trusting and assuming that the question asked is appropriate for the answers we seek. We are funneled towards responses. We close off nuance alongside distraction. We anchor in our own perception of what is important, and we perpetuate a series of answers that validate our thought patterns.

---

Getting coffee with friends, a conversation on AI in clinical settings surfaced. The friend next to me, a licensed (mental health) therapist, discloses that her patients and clients have gone to ChatGPT for decision making. As a therapist, she reiterates that her job is not to help patients make decisions, but to empower them to be guided towards growth and navigating their decision making.

Likewise, in the physical therapy world, I am not opposed to patients finding tools outside of our treatments to explore movement patterns, modalities, and strategies to better understand their health or body and to feel more empowered in the process of recovery. I will gladly be fact-checked and happily collaborate on a plan that only feels more accessible and feasible and personalized. I am, however, not a big fan of going to AI for the solution.

AI is not a trusted source. AI leverages a massive data bank. What answer it gives is plausible. What it answers for you is not the gold standard with respect to research or expertise.

As AI becomes more prevalent and unavoidable part of our lives, I hope that those who opt in to use it can continue to do so in increasingly more ethical ways. I hope that we don't lose the plot of human existence, in conversation and nuance. I hope that I can continue to help people feel empowered in their bodies, help them understand their medical condition, and better communicate their plan of care.

---

When I think of my patient now, and that day of treatment, giving in to her requests and allowing her to trial all of the exercises ChatGPT advised to her, then watching her get more discouraged, I am not sure what exactly I could have done differently.

The question I ask myself: did I do the best that I could that day?

And, given everything that happened that day, I think I did the best I could. Maybe I can pivot next time and prioritize sitting down for a more thorough discussion about the AI prompt and answers. Ultimately, I can only hope that she lives with less pain and finds greater joy in whatever she chooses to do -- whether that is with my help or with AI.

---

Comment, like, or leave a message. Happy to connect and share stories over the years!

---

Written by:

Judith Wang

Founder, Project Green Beard PT, DPT, CSCS and UCSC Banana Slug