AI-scribe-appt

Art by Tamara Rees

Since she opened a veterinary practice in 2020, Dr. Summer Hedges has been working to reduce the time she spends writing up patient medical records. She tried training staff at her Durham, North Carolina, clinic to keep notes during exams. She experimented with speech-to-text transcription software. She even hired a medical transcriber in the Philippines who sat in on exams virtually, taking down what was said and done.

Hedges found none of the strategies satisfactory.

In 2023, she signed on to beta test scribing software that uses artificial intelligence. In general, AI scribing software produces an audio record of a doctor's appointment, creates a written transcript, and converts the relevant portions of the transcript — filtering out unrelated chatter — into a formatted medical record.

Such a complicated series of tasks is possible because AI software is able to "learn" from being trained on vast amounts of digital data.

Two years later, Hedges is fully on board with the software. Count her among the early adopters of a tool that appears to be fast gaining traction.

In the past three or so years, around 20 companies have brought veterinary AI scribe programs to market. Some are startups headed by tech entrepreneurs and funded by venture capital. Others have been developed by established veterinary services companies as adjuncts to their practice information management systems.

Aaron Massecar, head of partnerships at CoVet, a veterinary scribing company launched in February 2024, estimates that about 20% of practitioners in the United States are using an AI scribe application. He said initial trepidation among veterinarians about trying the new tools has dissipated.

"A year ago, it was 100% a leap. Now, it's a baby step," Massecar said. "That is the fastest that anything has been in vet med."

Before joining CoVet, Massecar was the executive director of the Veterinary Innovation Council (VIC), a group of veterinarians and industry leaders brought together by the North American Veterinary Community with the self-described mission of promoting innovation in the profession. 

Polls conducted by the Veterinary Information Network, an online community for the profession and parent of VIN News, show a spike among its members in the use of AI scribing over the past 14 months.

When asked in July 2024 how they create and store medical records, about 3.5% of respondents said they used an AI program. In September 2025, the figure had risen fivefold, to 17.5%. (The number of respondents in each poll was comparable — 2,401 and 2,387. The respondents were not necessarily the same individuals, and the polls are self-selected and nonscientific.)

In an acknowledgment that AI technologies can't be ignored, two veterinary organizations put forth separate papers on the topic this year. One paper was by the American Association of Veterinary State Boards (AAVSB), the other by the VIC. 

Taken together, the papers present a picture of a technology with enormous promise for improving veterinary workflows, time use and records quality. At the same time, the authors warned about shortcomings in AI systems. They called for greater transparency from scribe companies and practitioners. Both papers urged veterinarians not to cede control of and responsibility for everything that goes into the medical record.

Reports from early adopters

On VIN member message boards and in conversations with VIN News, veterinarians expressed mixed attitudes toward the new technology.

Dr. Jennie Shih, who works at a practice outside Seattle, recently tested two AI scribe programs as part of an effort to adopt the software. She's found the tool largely helpful. "It's greatly reduced the amount of time typing charts at the end of the day," she said. "So, I'm able to get off work on time or at least minimize extra time spent in the office."

Shih said she appreciates the rich detail in her AI-scribed records, and she's been surprised at how well the product excludes small talk and unrelated information, like when an owner chimes in about a pet at home that's not the patient.

Barking is another matter. "When a dog is barking," Shih said, "[the software] can't distinguish between the human voice and the dog voice, and it cuts out the entire conversation."

Appointments involving more than one pet can also result in scrambled information in the record, according to several veterinarians.

Practitioners said using AI scribe software required them to develop new habits, such as remembering to turn on the microphone — whether in their computer, smartphone or a smartwatch — before starting the appointment and remembering to turn it off at the end.

More significantly for some, they had to get used to narrating the exam for the recording. The scribe tool aside, Shih considers the narration beneficial to the pet owner. "I think it's a good habit to verbalize what I'm seeing to clients who sometimes don't even realize you're doing an exam unless you tell them," she said.

Veterinarians also need to review AI-generated records to spot errors, omissions and random or invented details — the notorious "hallucinations" — before finalizing the records.

Hedges, the beta tester in North Carolina, said the glitches she detected, mostly early on, were easy to catch. "It was really obvious," she said, recalling a time she examined a male dog whose owner was pregnant. The record listed the dog as pregnant. Hedges notified the scribe company of the error so it could make improvements to the tool.

She also found that the tool had difficulty understanding people who spoke with heavy accents — a problem she said can be addressed with further training of the software.

Other practitioners reported drug names and veterinary terms being incorrect or misspelled. One said that the AI would invent new drug names. Another said the tool sometimes inserted information from one patient into a different patient's record.

Dr. Joe Frost, a semi-retired practitioner in Florida, has used three different AI scribe programs, which he described as similar to one another. He said he's rarely seen errors and hallucinations.

In fact, Frost believes AI-scribed medical records are almost always an improvement over records composed directly by humans. His assessment is based on the experience of reviewing about a dozen records a day for three years on patients that had moved to the Sunshine State from all over the country. He described most of the records as "pathetic," often composed of only an invoice that listed procedures, with no information about why the animal was seen.

"AI records are better records — flat out," he said. "Not better than what you could do, but better than what you are doing."

More detail in patient records consistently could have wider benefits for veterinary medicine, said Dr. Brian Hur, president-elect of the Association for Veterinary Informatics, which aims to expand the use of clinical information to improve the quality and safety of patient care. Hur works on AI and data science efforts at VIN.

He explained that veterinary medical records have long been seen as a potentially rich data set for research, but they are often incomplete. This contrasts with human medicine, where more uniform and complete records exist for Medicare, Medicaid and other insurance reimbursement.

"Most of the time, veterinary medical records are written after the appointment and sometimes at the end of the day," Hur said, "They're not commonly being written live during the examination."

Hur was stymied by sparse data when he delved into clinical records to research antimicrobial drug use in cat bite abscess cases. "A lot of times you can't find simple stuff about … like whether or not antibiotics were actually warranted in a case," he said.

Because AI scribes create medical records based on transcriptions of appointments, there's "no guesswork involved as to what was said or what was done," Hur explained. With more thorough narratives, he said, "We might start filling in the gaps of that data." 

'I won't use it'

Reports of errors, glitches and hallucinations in AI scribes are deal-breakers for Dr. Chiara Switzer, a small animal relief veterinarian in Ontario.

AIscribe_disclosure

VIN News Service photo

To help inform clients that appointments are recorded, some companies that make scribing software provide signs that may be posted in examination rooms and waiting areas.

"I won't use it," Switzer said. "This technology is not good enough for these critical documents. Records are a huge part of medical care. It can hinder a future doctor, the care the patient gets and the cost of the care."

Switzer is comfortable keeping handwritten records on paper or typing them into a computer program. "A good record is a good record, regardless of how you make it," she said, adding, "as long as the handwriting is legible."

As for AI saving time, she's skeptical. "I think I can write my record faster from scratch because an AI scribe record would need to be carefully proofread," she said.

She worries that not all veterinarians will be vigilant about checking AI's work, especially as the technology improves. "If nine out of 10 records are correct, you are going to get lazy about reading them," she said.

CoVet's Massecar, who was the editor on the VIC paper, said it's possible. "The models have gotten better at telling the truth, rather than making things up," he said. "Some people have felt so confident that they no longer check, but that is incredibly rare."

Tech moves fast as regulation lags

The three AI scribe company executives who talked to VIN News said the software is built on and uses large language models, sometimes called foundational models, created by companies such as Anthropic, Google and OpenAI. But each company said it has its own proprietary training for veterinary use.

The foundational technology is rapidly evolving. "It's really only been two years that AI has been functioning at the level it is," said Ryan Gallagher, CEO and co-founder of Scribenote, "and every six months there's a big-step change in what it can do." Gallagher started developing the Scribenote software while he was an undergraduate engineering student to help veterinarians like his sister, who was spending long hours at work catching up with her medical records.

The evolution, he notes, is quite different from the last innovation in veterinary record keeping — digitized records stored in practice information management systems, or PIMS.

"PIMS may have existed for years before you actually got onto a PIMS," Gallagher said, referring to veterinarians, "and it wasn't that far removed from concepts you're already familiar with because you're basically just taking all the stuff you did manually, and you're making it digital. Whereas with AI, there's a new intelligence in the flow, and it's happening way faster."

Among those racing to catch up are regulators. In March, the AAVSB weighed in with a paper on the emerging technology. The AAVSB is a nonprofit association of veterinary medicine regulatory boards in Canada and the United States.

Dr. Beth Venit, the association's chief veterinary officer, said the need for guidance popped up on her radar during a conference two or three years ago, when she was in a conversation about the use of AI to interpret radiographs.

"I said, 'Well, I don't know how accurate those are,' " she recalled. "And one person said, 'Oh, they're really accurate.' And another person said, 'It doesn't matter anyway, because the FDA doesn't regulate them, so we can do whatever we want.' I turned to my boss, and I was like, 'We're gonna have to get involved in this, aren't we?' "

Venit was charged with shepherding input from 29 experts for the paper. It opens with a message in bold type: "The responsibility for appropriate use of AI rests entirely with the licensee, and their decisions and actions are regulated."

A key takeaway of the paper, she said, is that although AI scribing software is new to the field, veterinary medical records are regulated, and those regulations apply to AI-scribed records. Practitioners and regulators, Venit said, "don't have to wait for new statutes or regulation."

Rules that apply to conventionally generated records on privacy, storage, security, retention, and patient and client confidentiality also apply to AI-generated records, the AAVSB said.

AI scribe executives who talked to VIN News maintained that they follow safety and data security protocols set for AI management systems that are stronger than the conventional standards and systems used for managing veterinary records.

Rohan Relan, founder and CEO of ScribbleVet, said his company is certified as "SOC 2 Type 2-compliant," which he described as uncommon in the veterinary industry. A computer engineer, Relan started his AI scribe company after witnessing the challenges of veterinary recordkeeping during many hours in hospitals with Potato, the dog he adopted off the streets of Tijuana and who had myriad medical issues.

The compliance standard was created by the American Institute of Certified Public Accountants in response to the growing need for organizations to protect customer data, especially as data moved to the cloud, according to a brief history of SOC compliance posted by a data security group.

To meet the standard, Relan said, "We have to prove over a period of time that we have security and privacy controls in place and that they are being actually enforced. We also, for example, have to do an annual penetration test where we're paying people to try and exploit the system, and if they find any ways to exploit the system, we can go and fix those."

Last-generation veterinary software companies that "touched a lot of client and patient information did not as rigorously have these compliance protocols in place," said Scribenote CEO Gallagher. "Whereas with the current wave of AI tooling … we've had to elevate the standards."

Questions about data use and ownership remain.

"[You] need to know what happens to the data that is being fed into the AI system," Raphael Moore, general counsel of VIN, posted in a message board discussion about scribing software in January. "Does it get owned by the vendor? Do you have to elect to have it erased from their cloud? What happens to personally identifiable information? Is it used to train an AI system? This requires careful delving into the terms and conditions of the service provider."

Executives at CoVet, ScribbleVet and Scribenote told VIN News that they consider medical records and data associated with the medical records to be the property of the practice. They have policies saying either that they do not sell or share that data beyond what is needed to deliver services or only do so with the express consent of the practice. They also will delete any practice's data stored in the cloud upon request.

As for data use in training, the AAVSB guidance says: "All devices require periodic training and may require transmitting practice data back to developers. Licensees should avoid using AI systems that do so without adequate safeguards, such as anonymization or encryption."

Approaches to using data in training vary. At Scribenote, for example, the company will use anonymized data to train its tool from users who access the tool for free. Customers using paid plans may opt out of having their data used for training.

"I really want to stress that in order for training to happen, there's a lot of anonymization that needs to be done so that you're not actually training on a patient's data," Gallagher said.

ScribbleVet's Relan said the company may use a 30-second clip from a recording to train the program if the transcription included an error, such as a misspelled drug name. If that 30-second clip has any personally identifiable information in it, he said, the company won't use it for training.

What about recordings and transcripts?

Though medical records are covered in many jurisdictions' veterinary practice acts, AI-scribed records raise new issues, particularly related to recordings. Requirements for obtaining client consent to record appointments using AI fall under a jurisdiction's rules for any type of audio recording. Open questions include how recordings should be stored and whether recordings — and transcripts — constitute part of the medical record.

Scribe company terms reviewed by VIN News encouraged or required veterinarians to get pet owner consent before recording appointments.

"We consider recordings to be some of the most sensitive information that we have," ScribbleVet's Relan said, referring to the fact that a recording device might pick up a conversation in the background. For this reason, the company deletes recordings from the cloud after storing them for no more than 90 days. Transcripts, Relan said, are stored for longer because personal information can be easily redacted.

As to whether recordings and transcripts should be considered part of the medical record, Gallagher said he's heard arguments for and against.

"One argument is that it's basically transitory information that is used to create a medical record," he said, likening it to notes on scraps of paper. "But the medical record itself is what you've stored and saved."

The other argument, which he's heard from regulators, is that when there is a complaint, all related materials — client communications, text messages, camera footage and so on — are seen as "relevant to the medical record."

One scribing company executive and one practitioner told VIN News that they had first-hand experience of recordings or transcripts being submitted to a veterinary board to refute claims in a complaint.

AAVSB's Venit said she hadn't heard of this happening, but she recalled that when she was in practice 15 years ago, her clinic recorded phone calls. "It aided us so many times where the client said, 'You never told me,' " she recounted. "And we're like, 'We have a timestamp,' " referring to the official recording.

Moore, VIN's general counsel, described a recording as a double-edged sword. "It can be used to refute a complaint. It can also be used to substantiate a complaint," he said.

He also cautioned that transcripts and recordings could be sought in totally unrelated matters, such as divorce cases, to show interpersonal conflicts or accusations. In a conservatorship or undue influence case, a recording could be used to demonstrate that a client appeared confused. They could also be evidence in personal injury cases, insurance and employment disputes, and even criminal cases.

"So rather than a vet saying, 'I don't have a record,' all of a sudden they are pulled into litigation cases," Moore said. "Ignorance is, in fact, bliss sometimes."

Next frontier: AI diagnoses

Perhaps the most controversial part of AI scribe tools is the fact that some scribing tools generate possible diagnoses.

ScribbleVet has partnered with LifeLearn Sofie, which uses AI to provide what it calls "clinical decision support" in the AI scribing tool. According to a press release, Sofie draws from a database that includes proceedings from the North American Veterinary Community annual meeting (known as VMX), Today's Veterinary Practice clinical articles and 25 industry-leading textbooks.

"We've heard that people want to see this," Relan said, "particularly new grads." ScribbleVet users can opt out of using the integration.

Relan stressed that the Sofie integration includes "relevant citations so veterinarians can dive deeper into the research to inform and support their diagnosis." In addition, he said, the integration is intended only for use by DVMs and is disabled when ScribbleVet is being used by anyone who is not a veterinarian.

The proffering of possible diagnoses raises alarm bells for AAVSB's Venit. "Providing a diagnosis is a veterinary-restricted task by law, so only a veterinarian should provide a diagnosis," she said.

"My concern is that if a veterinarian is super busy, they're relying on AI to just do the complete medical record [and there's a] diagnosis on the bottom ... they just glance at it and go, 'Great,' and move on," she posited. "If that patient gets transferred or another veterinarian within the same clinic is dealing with it later on as a follow-up, they're going to review that and say, 'Oh, OK, well, that's a diagnosis' — when, in reality, that was just a diagnosis that the AI made."