How innovators are using tech to get drug dosing right

A few months ago, Rhodes Hambrick, a pediatric neurologist, was at a party in Chicago when a man with a slight cold asked how much Sudafed he should take. He told him he could take two, but he’d probably feel jittery and should probably just take one pill. Dissatisfied with this answer, the man put the question to ChatGPT, which told him to take two pills. He took two and twenty minutes later was surprised to find he did indeed feel jittery. 

“You're not gonna have a stroke from taking two Sudafed,” Hambrick told me. “But I happen to have some clinical experience here.”

Ironically, Hambrick has tested AI’s ability to dose patients. He found that while large language models get dosing right most of the time, when they get it wrong, the results can prove fatal. In 2025, he challenged six LLMs (ChatGPT, Claude, Google Gemini, Microsoft Copilot, OpenEvidence, and Perplexity) to accurately dose antibiotics for complex patients either on dialysis or with impaired kidney function. The questions were intentionally difficult and though many answers were right, LLMs varied in their recommendations widely. Hambrick said what was really interesting is that even in cases where LLMs offered up a wrong dose, they referenced the right information.

“It would say all the right things, but then do the wrong math,” he said.   

Still, doctors are keen to use AI to help them fix a common cause of patient harm: medication errors. Just as patients are turning to AI with their medical questions, so too are doctors. Treatment and disease management accounted for a third of physicians’ OpenEvidence queries, according to a recent investigation published in JAMA. This included selecting the therapy, medication dosing, and managing patient safety. 

Why this matters: AI has the potential to help doctors, who are frequently burned out or under strain, to reduce medication errors. But it is not as simple as asking ChatGPT to suggest a dose. It potentially requires building fit-for-purpose tech to handle dosing or taking a different approach to AI prompting. 

In this article we discuss: 

• The research on AI’s ability to dose medication
• How one company built a fit for purpose tool
• Tips on how doctors can better prompt AI to get more accurate doses
• What history tells us about how we should be building for health care