FAQ: Where AI Use Is Now, Where It’s Headed in Pharmacy Practice

Although automation and data-driven applications have found relatively mature footing in pharmacy workflows, clinical uses remain far murkier, complicated by questions of accuracy, regulation, liability, and the irreplaceable judgment of a trained pharmacist.

Understanding where AI genuinely helps, and where it still falls short, has become essential knowledge for an industry balancing innovation with patient safety.

For pharmacists and pharmacy organizations, the stakes extend well beyond convenience. Emerging research shows mixed results on whether AI tools can match human expertise in drug information services, underscoring the continued need for pharmacist verification and oversight. Meanwhile, a fragmented regulatory landscape leaves liability questions largely unresolved for business leaders adopting these technologies.

In this exploration of AI’s use in pharmacy practice, both on the clinical and operational sides of the industry, we met with Timothy Aungst, PharmD, founder of the Digital Apothecary and professor of pharmacy practice at the Massachusetts College of Pharmacy and Health Sciences. Tapping into his research and expertise regarding AI’s use in pharmacy practice, he provided insights into where the technology is at now, and where it is headed in the near future.

READ MORE: FAQ: Navigating the Use of Diabetes Technology

At this point in AI’s evolution, what are its main universal uses in pharmacy practice?

I currently have 2 buckets I put AI pharmacy applications into: clinical applications or operations utility. Right now, I’d say probably the most mature and adaptable use of AI (and that’s a general term, I’ll admit) is within operational utility. This ranges from data entry, automation integration, prescription management, and related day-to-day operations that a pharmacy faces with acquiring, managing, and dispensing medications to their patients. Much of this is data-driven, integrated into pharmacy management systems, and has an established workflow that makes AI application a more realistic process.

Clinical applications is a varied topic with AI. I don’t think we are anywhere close to an ‘AI Pharmacist’ that can take on the verification process and management of patient care, with several overarching issues that haven’t been addressed. These include education, training, regulation and governance, ethical considerations, liabilities, patient buy-in, and ultimately cost. There are some grey areas that likely bridge between operations and clinical aspects, whether that’s advanced clinical decision tools, the use of agents for certain clinical topics (e.g., drug dosing, patient outreach); next generation drug information resources that we will likely add into our lexicon and work, but that still leaves the ‘pharmacist-in-the-loop’ and in charge.

Taking these into consideration, it is highly likely that all operational processes will see some level of AI tool integration within the next 5 years, given an established data-driven and identifiable ROI calculation for its use where appropriate per pharmacy organization. The clinical use will trickle in over time but likely will vary given the items previously stated that will pose a barrier.

How reliable are AI-provided outputs when it comes to clinical or drug-related information?

Generative AI tools can be split across several different applications within clinical and drug-related information resources that have led to varying degrees of accuracy based on current research. Currently, general-purpose LLMs (e.g., ChatGPT, Gemini, Claude) have received perhaps the most concern related to their clinical use for asking medical-related questions. These models are based on general and open data sets, with some training related to how to parse that data, and yet still have a risk of hallucinations and accuracy concerns. Other tools (e.g., OpenEvidence) have combined LLMs with the use of RAG and related modeling approaches to reduce the risk of hallucinations and improve accuracy. Some of these tools have also partnered with medical organizations and journals to have better access to what would be trusted evidence for their models to utilize. Lastly, incumbent drug and medical information resources are also integrating LLMs into their databases, using their current data as the resource to query.

In essence, our outputs we can theorize are limited by the data and model training that these tools are based upon, and current research has found conflicting results. Some studies have found that current AI tools underperform traditional pharmacists’ drug information services and highlight their inability to replace pharmacists trained for such services at this current time. Another recent study by Vishwanath and colleagues has caused a large debate [over] whether even medical-focused AI tools are better or comparable with GenAI tools.

Nonetheless, the largest issue with this debate is that the majority of research has compared the models’ ability against previously established questions compared against what a human answered, or general test banks with a ‘correct’ answer. At this point, it is difficult to say that AI-enabled tools outperform humans as there is limited data or attempts to actually integrate AI into real-life cases given the ongoing concerns they face. It stands to reason that we can anticipate that LLM and related tools will be integrated into drug information resources, but pharmacists will still be expected to conduct quality assessment on the outputs and take responsibility for their use in daily practice.

How important is pharmacist verification when using AI to help support any of their services or processes?

It will come down to risk acuity, and that will relate to previously mentioned concerns. This may range from personal comfort that may be impacted but not limited to colleagues and local culture uptake, their education and training, ethical concerns, and current regulations and governance. I think some areas are easier to conduct, such as the operational side, due to likely easier means to assay their accuracy and integration into workflow that pharmacists can see and identify if issues become present; versus the clinical side [where] pharmacists may see vastly varying uptake in the coming years.

Ultimately, with current practice aspects, the pharmacist would own the AI output, and they would need to decide where they are comfortable with using AI in their duties and tasks. Some very gray areas are the transparency level, where even in our discourse with colleagues that give us a response we may not realize they used AI tools to generate their own response, passing on to us that possible risk for perceived colleague expertise that AI delegated. Verification in this era will have to go through an evolution of sorts as we figure out what is permissible versus untouchable or high-risk due to patient concern and perceived negative outcome generation.

As organizations seek to leverage AI, they will likely trial and integrate in low-risk situations before expanding to larger riskier projects. This will help the cultural integration of AI and internal comfort for pharmacists and help with slowly gaining confidence in their utilization.

What are the major liability risks when it comes to AI use in pharmacy practice that business leaders cannot ignore before adopting the technology?

Issues that currently face pharmacy practice is the area of rules, regulation, and governance of AI and the liability and risk ownership. As of 2026, the White House released a National Policy Framework for AI, which will impact the FDA, CMS, ONC, and FTC with their AI oversight. However, the AI issue is being addressed state-by-state in general with some congressional initiatives, especially when it comes to payer use, liability, transparency, and data protection.

Overall, it’s a very fragmented time for AI that really highlights some of the difficulty facing this space. For those looking at the liability risk of AI, they should turn to their local institution and see what processes have been set up, whether committees have set procedure, and if they follow local regulations that have been implemented. Otherwise, this will be a quickly evolving area that will likely impact the clinical integration of AI at this time.

Expert Resource: Manatt Health AI Policy Tracker

What patient concerns may pharmacists have to quell as AI use becomes more of a normalcy in pharmacy practice and beyond?

This will be a tricky question, especially as we consider patient demographics. The foremost issue will be transparency with our patients. Pharmacy has used technology for decades now to handle the dispensing of medications to our patients and for our general operations. As these tools integrate AI, will we have to announce this to patients, or will it be ‘business as normal?’

Flashpoints may arise, especially with the rise of patient-facing agents, where some patients may find them helpful, and others may see it as a friction point in their relationship with the pharmacy, and may perceive it as an inauthentic experience highlighting the concern related to the human aspect of health and caring. These hurdles are multifaceted, and data and research remain mixed on best practices for the best approach on AI integration. Though, it will likely remain a touchy area dependent on utilization and visibility.

READ MORE: Technology and Data Resource Center