Admin_Adham
Some doctors use social media to share credible lifestyle tips, discuss new research, and spread sound health information — sometimes colorfully and to huge followings. The latest crop of medical influencers, however, aren’t even human.
Deepfake doctors — artificial intelligence (AI)-generated videos mimicking medical professionals — are increasingly being spotted on social media. These faux doctors peddle everything from false medical info to dubious products, like oils to treat hair loss or supplements for endocrine disorders.
The nonprofit organization Media Matters recently called out eight TikTok accounts for sharing suspected deepfake doctor videos. Last year, a BMJ investigation found that deepfakes of three famous doctors in the UK were being used to shill products, and the Baker Heart and Diabetes Institute in Australia warned that videos of their doctors promoting a supplement for diabetes weren’t real.
The avatars can be facsimiles of noted physicians or fictional figures generated by bots.
“What strikes me is how convincing these synthetic doctors are, with the visuals combined with compelling but potentially fake narratives,” said Ash Hopkins, PhD, associate professor of medicine and public health, who studies AI and healthcare at Flinders University in Adelaide, Australia.
“The potential for harm is concerning because when misinformation is presented in this form, it can be difficult to distinguish that it is not coming from a genuine healthcare professional,” he said.
Why Medical Deepfakes Are on the Rise
In one sense, this phenomenon is no different from other scams or sham operations that arise with new tech, particularly anything that can be amplified and exploited by the internet and social media.
“The recent surge in deepfakes reflects rapid advances in technology, making it possible for almost anyone to generate highly realistic synthetic content quickly,” said Hopkins.
He also suggests that it’s inevitable. “Stuart Russell, in his 2019 book Human Compatible, outlined concepts like if a technology like AI has the potential to be misused for harm or financial gain, it likely will be.”
What do scammers get out of it? People who peddle medical disinformation might aim to cause confusion, fear, discrimination; encourage people to reject evidence-based treatments; or promote “miracle cures” for financial gain. Physicians are the perfect foil. By presenting a deepfake as a trusted expert, they amplify the credibility of whatever they’re pushing.
“Their narratives are designed to appear convincing by incorporating fabricated logic [and] false testimonials or aligning with popular concerns,” said Hopkins. “This content can be especially harmful to individuals with limited health literacy or those already skeptical of conventional healthcare as they may be less able to discern that the source is illegitimate.”
Health-related deepfakes are part of a growing phenomenon of AI-generated content aimed at deceiving consumers. Daniel S. Schiff, PhD, co-director of the Governance and Responsible AI Lab at Purdue University, West Lafayette, Indiana, says he and his colleagues are seeing an uptick in false advertisements emulating celebrities and other public figures, too.
They don’t even have to be good. They just have to be convincing enough.
How Deepfakes Work
The rise of deepfakes began around 2018. Early deepfakes were created using generative adversarial networks. These machine learning frameworks pit two neural networks against each other. One, the generator, attempts to mimic an image. The other, the discriminator, compares the creation to the real thing and tells the generator how to keep improving.
“Back at this point, the images were already pretty compelling,” said Schiff. Then, the field of deepfakery cracked wide open with the advent of diffusion models, like OpenAI’s DALL-E, released in 2021; and Google’s Imagen; Discord’s Midjourney; and Stability AI’s Stable Diffusion, which hit the market in 2022.
Diffusion models start with a noisy image, like static on a television screen, associated with a text caption describing the desired result. They continuously add and remove noise, receive feedback, and recognize and repeat patterns until the result looks like the real thing described by the text caption. These models can create sophisticated images in seconds.
“We’re at a point where neural networks trained on massive datasets of images and videos can produce extremely detailed and realistic visuals, including mimicking human faces and facial expressions,” said Hopkins. “At the same time, large language models can generate coherent, context-aware text and speech. When these two are combined, the result is AI avatars that can speak, emote, and move with remarkable realism.”
Bonus for scammers: These technologies require minimal input to create convincing videos.
The tech isn’t just getting more advanced. It’s also becoming increasingly accessible. Many diffusion models for AI imagery are open-sourced and easy to find. And you don’t need to be a computer whiz or have an expensive computer to do it. A decade ago, anyone who wanted to create a deepfake would have needed a lot of computing power to train and run AI models, says Christopher Doss, PhD, senior policy researcher and professor of public policy at RAND Corporation.
“As the models have gotten better and more efficient, the computing power needed has decreased,” said Doss. “People create applications that kind of take those underlying models and make them easier to create for anybody who doesn’t have a technical background.”
Anyone with a computer or smartphone can now make a deepfake. In a recent TikTok video, cosmetic chemist Javon Ford said that the Captions app, which features easy-to-use AI avatars that say anything you want, was probably used to make medical deepfakes recently spotted on the platform.
(The WebMD/Medscape video team created the AI footage for the attached video in 3 minutes using Google Veo3. For comparison, the graphics bookending the video took hours.)
AI Avatars in Healthcare
Social media isn’t the only place where you might see avatars with an uncanny resemblance to healthcare professionals.
While tracking AI trends, Schiff noticed an increase in healthcare companies using virtual chatbot avatars or virtual nurses. Companies like Hippocratic AI offer virtual providers for tasks like follow-up care and pre-procedure prep. Mediktor offers a virtual medical assistant to help patients assess symptoms. These aren’t deepfakes since patients are told they’re interacting with AI. But experts still have concerns.
Text-to-AI video is also used for educational purposes, distilling complex medical info to make it more easily learned — both for patients and medical students.
“Some of these tools may be good and helpful, [but] many of them are not well evaluated,” said Schiff. “Many of them, from my perspective, aren’t well aligned with careful, ethical thinking.”
The bots might boost efficiency, but will they improve care?
“My cautionary tale is that if you use AI to autogenerate messages to your patients or to auto-do therapy or whatever, it doesn’t necessarily mean that now you have twice as much time to talk to your patients and build relationships,” said Schiff. “It just might mean that they hire fewer people or they increase your workload.”
In January, the National Nurses Union, representing more than 225,000 registered nurses, organized a day of protests against the adoption of AI tech without sufficient staffing levels and patient safeguards.
How to Avoid Being Fooled
First, don’t assume you’re too clever to believe a deepfake.
In 2023, Doss’s team published the results of a study showing that a deepfake video about climate change deceived 27%-50% of people. He says that with the fast pace of AI advancement, even more people would probably be tricked today. He suspects the results would be similar across other scientific subjects, like health disinformation, especially when related to hot-button issues like vaccines.
We may be even more vulnerable since our usual human biases seem to apply. For example, if the avatar appears older, you might knee-jerk trust it more. Ageism towards doctors is common and people viewing older-looking AI avatars trusted those “doctors” more than younger-looking ones, according to a June 2025 study in Frontiers in Public Health.
Some folk wisdom suggests you can spot a deepfake if the person in the video has too many fingers, blurry hands or eyes, or a face that’s too symmetric. More subtle clues could be too-perfect hair, clothing texture, or flat or odd-looking jewelry. Schiff doesn’t recommend putting stock in those tips — the tech is getting too good.
“I wouldn’t suggest that individuals try to be deepfake detectors,” he said.
His team is benchmarking deepfake detection tools made by companies, academic researchers, and government agencies. Even these tools are not even close to reliable, he says.
Your best defense isn’t your eyes — it’s your brain. Keep practicing media and digital literacy, and encourage your patients to do the same.
The biggest tell according to Doss? The “doctors” don’t exist anywhere else. Before you put faith in a TikTok doc, look them up on other platforms. “If you can’t find anything about that doctor other than that video, then that’s probably a clue that doctor may not exist,” he said.
Schiff concurs, especially when the medical subject in question is high stakes. “When you see something pop up on social media for something high stakes — a medicine or a cure or whatever it is — instead, go to a reliable medical source,” said Schiff.
While not impossible, it would take a highly sophisticated scam to create a complete digital presence — websites, news articles, and other social media accounts — for a fake person.
“I think probably the biggest thing people can do is start instilling in themselves a healthy skepticism, and then let that healthy skepticism drive your search for more knowledge and for ways to verify what you have seen,” said Doss.
Deepfakes aren’t the only fake doctors out there, after all. “Sometimes we have deepfakes, sometimes we have cheap fakes or shallow fakes, so you can still fool people with Photoshop or basic splice stuff,” said Schiff.
Hopkins says there’s a pressing need for safeguards to minimize the harms accompanying the benefits of emerging AI technologies like deepfakes. Some of the onus is on the companies creating AI tools. They need to build in transparency to help viewers recognize AI-generated content, says Hopkins.
“AI deployers must ensure that AI-generated content is traceable,” said Hopkins. “This could involve watermarks, explicit disclosures, or clearly branded visuals on AI-generated material.”
The companies behind today’s most popular social media platforms could also step up — but will they?
“Social media companies generally are backing away from preemptively labeling these things, and so it really is becoming more incumbent on the person themselves to try to protect themselves because it does not seem like social media companies are seeing that as part of their core responsibility at this moment,” said Doss.
This article is part of an ongoing, year-long Medscape series examining the decline of public trust in medicine. More content is available here.
