I was initially skeptical about using artificial intelligence in clinical practice. My biggest concern was trust. What if it oversimplified a complex problem or missed important evidence? Medicine is full of nuance, and as most clinicians know, there often isn't a single right answer. 

I eventually realized that AI doesn't have to make clinical decisions to be valuable. Its real strength is helping clinicians get to the evidence faster, leaving more time to interpret what it means for the patient. 

The AI tool I use most often is OpenEvidence. I use it as an intelligent medical search assistant to quickly navigate the recent literature, summarize evidence, and identify relevant publications. I chose it because it is built specifically for physicians and integrates peer-reviewed literature, clinical calculators, guidelines, and links to the original references in one place. Most importantly, I can review the cited studies myself before applying the information to patient care. 

My Step-by-Step Approach to Using AI in Clinical Practice 

Most clinical questions begin the same way. A patient asks something I haven't encountered in a while, or a new study is published, or I find myself wondering, "Has anything changed since the last guideline?"  

Instead of opening five browser tabs and searching multiple databases, I begin with OpenEvidence. I ask my questions in plain clinical language using de-identified clinical scenarios and never include protected health information. I read the summary. Then I review the references it provides. If the answer influences patient management, I read the original paper or guideline before making any decisions. To me, AI replaces the time spent searching, not the time spent thinking. 

Real-World Examples 

Here are two examples of how I’ve used OpenEvidence in my practice. 

Example 1: Recently, I was evaluating a patient whose bone mineral density continued to decline. The patient sustained a fragility fracture despite more than 2 years of intravenous zoledronic acid. The question was whether there should be a waiting period before transitioning to subcutaneous denosumab.  

Instead of manually searching multiple databases, I entered the question into OpenEvidence. Within seconds, it summarized the available evidence, highlighted key practical considerations, and directed me to the original publications. I still reviewed the original article before making my recommendation, but AI dramatically shortened the time it took to find the evidence.  

Example 2: Another way I use AI is for administrative work. Prior authorizations often require citing guideline recommendations and clinical trial evidence. Instead of starting with a blank page, I use AI to generate a structured draft based on current evidence. I then personalize it with the patient's clinical information and ensure every statement is accurate before submitting it. It makes the process considerably more efficient. 

Lessons I've Learned 

Using AI has also made me a better consumer of medical evidence. I've learned that AI can occasionally oversimplify complex studies or overlook important nuances. Sometimes a recommendation is technically correct but doesn't apply to the patient sitting in front of me. That's why I never treat an AI summary as the final answer. 

Another important principle is privacy. I do not enter protected health information; instead, I ask questions using generalized clinical scenarios. 

Perhaps the biggest lesson is that AI is most valuable when paired with clinical judgment. It is excellent at retrieving information, but physicians remain responsible for determining whether that information is accurate, applicable, and appropriate for an individual patient. 

My Suggestion to Peers 

If you're hesitant about using AI, you're not alone. I was, too. In many ways, that hesitation comes from a sense of responsibility, and I think that's a good thing. It reminds us to use AI thoughtfully and always to apply our own clinical judgment. 

My advice is to start small. Ask AI questions you already know the answer to. Compare its responses with the original literature and learn where it performs well and where it falls short. Over time, you'll develop a sense of when AI is helpful and when you need to dig deeper. 

I don't think AI will replace physicians. But I do think physicians who learn to use AI thoughtfully will be better equipped to keep pace with ever-expanding medical literature. 

For me, AI isn't replacing clinical judgment. It's helping me spend less time searching for information and more time caring for my patients.