Why Pharmacists Should Never Rely Solely on AI's OTC Recommendations | APhA 2025
Dan Krinsky, BS, MS, RPh, FAPhA, discusses the use of ChatGPT and artificial intelligence in assisting pharmacists for the management of OTC self-care products.
Pharmacists are not recommended to solely rely on the use of large language models (LLMs) to manage over-the-counter (OTC) self-care medications, according to a session presented at the American Pharmacists Association 2025 Annual Meeting and Exposition.1 Despite AI’s ability to predict patient outcomes when using OTC products, pharmacists are still the authority on medications and should avoid solely relying on LLMs because of their inconsistencies.
“[LLMs] are capable of answering self-care questions and generally providing relevant information; however, [there is] substantial variability in the responses. They provide potentially unsafe advice in specific situations. How do we know what those situations are? Well, we don't know until we get in and start investigating and comparing their results with what we would find in a more traditional manner,” Dan Krinsky, BS, MS, RPh, FAPhA, said in a session titled Rockin' Remedies: Your Annual OTC Update.
Researchers and pharmacists alike have explored the use of AI and LLMs to inform their decisions and streamline their tasks. | image credit: Anthony Paz / stock.adobe.com
READ MORE: ChatGPT Could Help Community Pharmacists Save Time
With the current state of the pharmacy industry, pharmacists and pharmacy employees experienced severe burnout, workforce shortages, and declining reimbursements for services, which has led to an unprecedented rate of pharmacy closures. Amid these challenges, researchers and pharmacists alike have explored the use of AI and LLMs to inform their decisions and streamline their tasks.
Throughout Krinsky’s presentation, he discussed previous research that has explored AI and LLMs. More specifically, he discussed the use of ChatGPT in helping pharmacists manage OTC product regimens and how LLMs have been researched for pharmacy implementation.
But before discussing Krinsky’s perspectives on specific literature exploring ChatGPT and OTC self-care products, it’s important to note what previous research claims about ChatGPT in pharmacy overall.
“This research shows that using ChatGPT as a writing assistant is valuable and can free up time,” wrote authors of a study published in Exploratory Research in Clinical and Social Pharmacy.2 “Although clinical questions seem promising, ChatGPT's answers are currently too unreliable and do not meet the required quality standards for good pharmaceutical care.”
Almost directly aligning with Krinsky’s stance, ChatGPT can certainly be utilized to help streamline the management of medications. However, the unreliability of ChatGPT remains the biggest hurdle to its implementation across the industry.
“So, when I look at this, and I look at what this means in the whole area of self-care, the conclusion I came to was there's still too many inconsistencies to rely solely on an LLM for answers to self-care questions,” he continued.1 “Consumers still need to understand the limitations of LLMs and ChatGPT, which seems to be the most popular one.”
Krinsky explored 2 previous studies that were conducted in Belgium and Japan. In the first study in Belgium, invesitgators explored the correctness of LLMs compared with verified clinical recommendations.
“They looked at language independence, context sensitivity, and reproducibility of responses. They developed this 5-point scale from inaccurate to highly accurate,” he said.1 “When you look at the results, the paid version of ChatGPT was the most accurate, but there was a lot of variability if you look at the 6 different models and the questions asked in Dutch versus the questions asked in English. Again, in my opinion, this is exactly what I would have expected based on some of the limitations we're seeing with some of the LLMs that exist.”
While these researchers explored a variety of LLMs, the study conducted in Japan only used ChatGPT to evaluate consumer inquiries into various OTC medication questions. They explored 22 OTC products and investigated ChatGPT’s success in assessing questions. Successfulness of responses was measured by 3 criteria: coherence between the question of the response, scientific correctness, and appropriateness of the instructed actions.
ChatGPT was only able to satisfy all 3 criteria in 21% of its responses, highlighting significant inconsistencies in the LLM’s ability to manage OTC medications.
Krinsky concluded his presentation by noting a case study where a patient was experiencing inconsistencies among Google, ChatGPT, and health care professionals. After attempting to relieve acute pain with online advice, the patient was eventually confused and forced to visit their pharmacist for the proper guidance.
Cases like these highlight the fact that ChatGPT and LLMs are still being developed. Until they are able to present evidence-based results that can be implemented across the industry, experts do not recommend that pharmacists rely solely on the work of these emerging yet inconsistent technologies.
“We need to do a better job telling our story, telling people they need to come to us to ask questions,” concluded Krinsky.1 “We can't allow people to rely on Google Med and ChatGPT, [or] whatever other resource they have at their disposal on their iPhone or their laptop for making decisions about their health care. Things are too serious. There's too much risk out there, and we have some knowledge that we can share with our patients.”
