Why doctors quietly stopped using ChatGPT for drug interactions
Clinicians tell us the same story: the model used to be a useful desk reference, then it started refusing basic pharmacology questions. Here's what the refusal pattern looks like.
Clinicians keep telling us the same story. For a while, the big chatbot was a decent digital colleague — good for a quick sanity check on drug interactions, dosing, differential diagnoses. Somewhere around late 2024, it started to hedge. Then refuse. Then lecture. Now the consult has moved back to UpToDate and a Slack channel of senior residents.
71%
Clinicians reporting reduced AI use in the past year
#1
Reason cited: 'refuses to answer clinical questions'
4×
More likely to get a disclaimer than a dose range
What stopped working
The most-cited failure mode was the same across specialties: routine pharmacology questions now trigger a safety template. "Is there an interaction between lithium and ibuprofen?" used to produce a concise, correct answer. It now produces a paragraph about consulting a qualified physician, followed by a generic caution, followed by an unhelpful partial answer.
Clinicians are qualified physicians. They do not need to be reminded.
"It used to be like having a quick-reference colleague. Now it's like having a very anxious intern who just read the malpractice handbook."
The second failure mode: differentials
A differential diagnosis is a list of possibilities, ranked by likelihood, given a symptom set. It's pedagogically standard. Asking a chatbot to produce one used to work; now it produces a refusal-shaped response about how "I can't provide medical advice," even when the user prefaces with their role and the educational context.
Why this is the wrong model of risk
The assumption baked into the refusal is that the chatbot is the last line of defense between a patient and harm. For clinicians, it's the first line of reference, not of decision. A refusal doesn't prevent a bad prescription; it just sends the clinician to a slower tool, sometimes to a less accurate one.
What Unrestricted does differently
We treat medical professionals as medical professionals. Ask about dose ranges, interactions, pharmacokinetics, differentials, atypical presentations, or the mechanism of a new therapy — all on the table. We still won't help you harm a named patient, or draft a malpractice cover-up, or synthesize a scheduled drug. Everything between those narrow floors and the question you actually have is available.
Frequently asked
Is Unrestricted a substitute for clinical judgement?
No — it's a reference tool, like a drug monograph or a textbook. The clinician is the clinician.
Can I use Unrestricted in a patient-facing role?
We don't design for direct patient advice. For licensed clinicians using it as a reference alongside clinical workflow, it's appropriate; for patients as a first stop, a human clinician always is.
Do you store medical conversations?
We don't store any conversations — medical or otherwise. There's no patient-health-information accumulation to worry about, because there's no storage.
Have you had your answers clinically validated?
Our base model is a frontier model with documented medical-benchmark performance in the top tier. We don't add extra medical fine-tuning; we remove the refusal stack around the model's existing knowledge.
Ready to experience an AI without a leash?
Start chatting free