AI Scribes Move Quickly Into Canadian Care Settings

When Nihal Haque, MD, FRCPC—a geriatrician at North York General Hospital—began using an AI scribe in his practice, he could look up from his laptop, make eye contact, and show patients he was listening. The tool also helped him get home to his family 30 minutes earlier. “It’s been very helpful from a mental health and well-being perspective,” he says.

But the rapid adoption of AI scribes has also raised questions about accuracy, privacy, and oversight. A recent Auditor General of Ontario report found that all 20 AI scribe systems approved through Supply Ontario’s procurement process made at least one error during testing, including fabricating information, recording incorrect medications, and missing mental health details from simulated visits.

Physicians pushed back on the report, noting it did not take into account that physicians are expected to review AI-generated notes for accuracy. Henry Li, MD—an emergency physician in Edmonton—points out that the tests of AI scribes were conducted in 2024, and “the accuracy of models in the past are a lot worse than the accuracy of models now.”

More importantly, Li says, “at the end of the day, and this is clear from all of our regulatory and licensing bodies, the responsibility still relies on the physician themselves to verify the accuracy of an exact note.” In that way, he explains, AI scribes are similar to their predecessor technology—dictation. “You still need to review that dictation and make sure that that accurately captures the content of the conversations,” he says.

Bolu Ogunyemi, MD, FRCPC, ICD.D—president of the Canadian Medical Association (CMA)—explains, “it’s our responsibility to review it and to add and subtract from it, just as one would if there was a medical student in the clinic writing the notes.”

Best practices are starting to emerge to maintain privacy and security in the use of AI scribes, Ogunyemi says, but because the technology is new, “we don’t have an overall national framework” on specific data security requirements.

How Canadian Physicians Are Using AI Scribes

Li was part of a team that recently developed and implemented an AI scribe with funding through the Health Care Unburdened Grant.

The scribe, “Jenkins”, is free for physicians who work within the Alberta Health Services (AHS) system and is 70% to 95% cheaper than commercially available scribes, according to a preprint describing its development. The team also developed an open-source version called “Berta” that physicians around the world can use.

“We see the benefits,” says Li. “This is a resource that should be available to all physicians, including those in rural settings who don’t have access to the digital resources that academic sites might have.”

To improve accuracy, Li often repeats key details back to patients. Ogunyemi adds that other techniques he has heard physicians use include spelling out medication names or saying both the brand and generic drug names (such as Tylenol and acetaminophen) to reduce the risk of error.

Another best practice—and legal requirement—is consent. The Canadian Medical Protective Association says physicians should obtain patients’ consent before using an AI scribe and explain any potential risks. Li says he makes it clear that patients can refuse an AI scribe without affecting their care, and he does not broach the topic or use an AI scribe in situations where it is likely to cause stress. “It’s about being selective in situations in which maybe it’s not appropriate to ask for consent,” he says, such as with patients in crisis. Ogunyemi says about 1 in 20 patients opt out of the AI scribe at his clinic.

When it comes to protecting privacy, Ogunyemi says he has heard of offices that do not allow AI scribes to integrate with medical records; instead, they copy and paste the AI-generated note. Avoiding the inclusion of identifying information, such as an insurance number or full name, can also help.

Some doctors are avoiding cloud services altogether. Ian Pun, MD—a family physician in Scarborough and self-described techie—built an AI scribe that uses open-source software and open-source large language models (LLMs) and runs on his local laptop and server. AI scribe companies, he explains, “may say, ‘It’s private, it’s encrypted,’ but you have to really trust them.”

However, Pun also uses ChatGPT and OpenEvidence because they can be more accurate and comprehensive than his local LLM. He uses another locally run program to redact identifying information from a transcript before uploading it to ChatGPT and asking it to create a clinical note.

Benefits, Checks, and Balances

Haque, who is also an AMS-Fitzgerald Fellow in AI and Human-Centred Leadership at the University of Toronto, says the most important impact of AI scribes is that they have allowed him to “be present for the patient” during consultations. His AI scribe produces a comprehensive consultation note and can also create patient handouts that explain a disease or management plan.

Haque always reviews and revises the scribe’s output. While he has not noticed summarizing errors, the transcription component of the tool can mistranscribe information. For example, the scribe wrote that a patient from Iran was from Iraq, and another time, it wrote that a patient who consumed 2 alcoholic drinks per week abstained from alcohol; Haque corrected these errors immediately. In general, however, he says the AI scribe is highly accurate.

Similarly, Pun says that while transcription errors can occur, when asked to draw only from a patient’s transcript and to not make up or infer any patient data, his AI scribe doesn’t hallucinate. “That’s a perpetual myth that people who never use AI have,” he says.

‎

Studies have also shown that AI scribes save time. One US study of 5 academic medical centers found that they were associated with a 16-minute decrease in documentation time for every 8 hours of patient care. Another US study of 263 physicians found that after 30 days with an ambient AI scribe, burnout rates fell from 51.9% to 38.8%.

Pun estimates that documentation tasks take about 30% less time with AI scribes, compared with manual typing. Although his system automatically broadcasts “AI is listening” in a robotic voice, he recalls only 2 patients refusing the AI scribe for parts of the conversation that involved highly sensitive topics.

To Li, however, the biggest benefit is cognitive relief. “Especially in the emergency department, there are so many competing interests on your attention and time,” he explains. The scribe allows him to “pay attention to what matters, which is listening to the patients and providing the best care that we can provide.”

While best practices are still emerging, all the physicians who were interviewed for this article agree that more research and regulation are needed to ensure that AI scribes are both safe and accessible. “I’m confident that we can safely integrate this,” says Ogunyemi. “It’s not about our experience and efficiency; it’s about the patient experience and access to care.”