What Should Physicians Consider Before Choosing an AI Scribe?
- Innovative ideas to enhance primary care: A guide to selecting and implementing AI Medical Scribes
By: Dr. Daniel Ngui
Sep 22th, 2025
đź“– In a new commentary, Dr. Daniel Ngui, MD, clinical professor at UBC, shares practical insights on how clinics can adopt AI scribes responsibly. See the original post here.
The article “Practice tips to enhance documentation in primary care: Guide for selecting and implementing artificial intelligence medical scribes”, was just published in Canadian Family Physician (September 2025 issue)
Read full article here: https://www.cfp.ca/content/71/9/569/tab-article-info
For a broader perspective on the benefits & risks of AI scribes, check out Dr. Ngui’s earlier blogs:
Your Top Questions About AI Medical Scribes—Answered by Dr. Ngui
How Empathia AI is Revolutionizing Physician Billing & Documentation
Primary care clinicians are facing increasing demands in practice and more burnout. We have large rosters of patients, a growing list of more complicated conditions and finally, the daily burden of tasks. During our encounters, we gather detailed patient histories, review charts, strive to meet our documentation standards while simultaneously engaging meaningfully with patients. Addressing these challenges with innovative technologies like ambient artificial intelligence medical scribes (AI scribes) is timely and critical for sustaining primary care and clinician well–being.
What Are AI Scribes?
AI scribes are software applications leveraging artificial intelligence, natural language processing and automatic speech recognition to generate draft notes by analyzing and interpreting conversations between physicians and patients. Anecdotally, these advantages improve our ability to multitask and lead to higher–order processing and improved decision–making compared to traditional methods of writing, typing, and/or dictating. Emerging evidence in the literature show AI scribes reduce cognitive load, may enhance efficiency, improve patient satisfaction, and restore the joy of practicing medicine.
The authors, with the combined experience of nearly 10,000 AI–assisted notes, have witnessed the positive impact of these tools. This article provides a framework for selecting and implementing AI scribes, explores potential pitfalls, and the appendix is a useful and quick guide.
Benefits of AI Medical Scribes for Clinical Documentation and Patient Care
Allow clinicians to focus on patients’ narratives and patient engagement/education by automating note–taking.
Clinicians utilizing techniques such as verbalizing non–verbal observations and ones’ thoughts during exams, auto–generating patient summaries/handouts, and summarizing the patients’ care plan can improve both the scribe’s documentation and patients’ understanding.
Saving time enables more thorough discussions about diagnosis and treatment and may lead to improved patient–centered care and shared decision–making.
Improved quality of notes can help improve continuity of care, improve documentation for insurance and medical-legal purposes and facilitate team based care.
Challenges and Risks of Using AI Medical Scribes in Healthcare
AI–generated notes must be accurate with minimal errors to ensure safe and effective care. Vigilant proofreading is essential and our professional organizations have outlined our ethical and medico–legal responsibilities.
Hallucinations:
AI scribe programs may create “hallucinations,” which are errors in synthesizing a draft note.
Just as young children can make up facts when telling their stories and their facts are not always based in reality due to their limited knowledge and experience.
These AI hallucinations or fabricated facts are sometimes created to make the note sound reasonable and can be very plausible.
Many AI programs have customizable settings to minimize hallucinations.
Bias in Research Trials:
AI programs have inherent bias.
For example, if an AI program has training data on male populations or young patient populations, only then AI-generated clinical notes could be biased without full data on female and/or elderly populations.
Premature Closure:
AI scribes provide us with efficiency that may allow for more time to look up the latest guidelines and/or evidence-based care during the visit.
However, an over-reliance on AI scribes draft notes could lead to premature closure, when we rush to a conclusion without considering alternative possibilities.
Privacy and Consent
Adherence to Canadian data privacy regulations PIPEDA for privacy, security, and patient consent is paramount.
Current best practices are to get explicit verbal and written consent and ensure your patient knows you are using AI.
We need to ask if the data is used for external training or data mining, and stored in compliance with suggested retention times.
Ask vendors if a privacy impact assessment has been done.
AI Scribe Cost vs. Value: How Physicians Can Decide
There should be a future debate about who should be paying for these primary care enhancement tools.
Health system payers should consider that AI scribes could be a solution to enhance capacity and address physician shortages.
Many AI scribe programs vary in cost and a cost-benefit analysis is important.
The cheapest program may not always be the best.
More costly programs may have unnecessary features.
What matters most is experimenting with programs to find one that maximizes total time saved.
Subpar programs may require more editing, have more hallucinations, or less user-friendly platforms. When choosing the best program, focus on note accuracy and only the features you will use regularly. There is no such thing as a perfect EMR program or a perfect AI scribe. One may have to pilot between 1 to 4 companies to find the right AI scribe for one’s workflow.
Questions to Ask Vendors
Quality, Accuracy, and Reliability
How does your program optimize quality, accuracy, and reliability of AI Notes?
Are you using the latest NLP engines? What plug-ins do you provide to minimize editing time, inaccuracies, and hallucinations?
Efficiency, Navigation, and Access
How is your user platform optimized for efficiency, navigation, and speed of access?
Do you have smartphone/device applications that allow contacting patients while displaying office number?
Can you demonstrate time-saving features, customization of outputs (handouts, referral notes), dictation, URL incorporation, language options, etc.?
Customization and AI Learning
Can I customize visit templates or note style preferences (point form, narrative, brief, comprehensive)?
How can I provide feedback to improve future notes?
How do I train the AI to learn my preferences?
Support and Pricing
How responsive is your support team to queries, and what training materials (videos, case studies) are available?
Does your company involve medical advisory boards or practicing clinicians in product development?
What pricing models, group discounts, or subscription plans do you offer?
Integration with EMR
Is your program integrated into my EMR system?
If so, how does integration impact workflow efficiency and cybersecurity risk?
If not, what are the advantages of using your program alongside a separate EMR?
Best Practices for Implementing AI Scribes
Define Needs and Goals
Clinicians should start by assessing their documentation challenges. Identify whether your focus is producing comprehensive consult notes or short SOAP notes. Consider the most common daily tasks and select AI scribe programs that efficiently address those needs.
Pilot with a Structured Evaluation Plan
A short pilot of one to two weeks per program is typically sufficient to test usability, uncover potential issues, and record structured ratings. Set a clear start date for adoption and allow one to three months for optimization and for the AI to adapt to your style. Continuous quality improvement through team discussions, user groups, and vendor feedback helps refine workflows and maximize efficiency.
Engage Patients and Document Consent
Use clear and standardized patient scripts to explain AI scribe use and obtain explicit consent. Document consent directly in the EMR.
For first-time consent, develop a succinct patient scripts such as: When getting first consent, try “...I need your permission to use an industry standard AI program that listens to our conversation, both privately and securely, which helps me create a draft note, allowing me to be more efficient and improve the care during the visit, ...is that ok?”
For returning patients, at the start of the visit “...As a reminder, both myself and my AI program is listening, generating notes and helping me to better focus on your care… ...is that ok? “
Some additional ideas include asking vendors and professional associations for written consent templates, office posters and/or website updates about AI scribe use.
Conclusion
AI scribes offer a viable solution to reduce administrative burden and improve efficiency in medical practice. Choosing the right AI scribe requires careful and systematic consideration of factors such as EMR integration, privacy compliance, features, and pricing. Future research will be crucial for continued improvement and responsible implementation of AI scribe technology.
