More than 1,000 AI-generated “doctors” spread approximately 100,000 deceptive medical videos, EDMO finds

More than 1,000 profiles presenting themselves as doctors but generated with artificial intelligence have collectively produced approximately 100,000 deceptive medical videos in around 30 languages, according to cases compiled by the European Digital Media Observatory (EDMO). The phenomenon is part of a broader ecosystem in which fictional medical identities, selectively presented evidence and promises of unproven treatments are turned into content capable of generating audiences, sales, donations or platform revenue.

In brief The network tracked by EDMO partners comprises more than 1,000 AI-generated “doctors” and approximately 100,000 videos in 30 languages. Some characters have seemingly realistic identities, wear medical coats and offer advice about specific conditions, even though the people presented as doctors do not exist. Health-related disinformation accounted for 7% of all cases detected by the EDMO network in the EU in both July and August. Disinformation concerning Covid-19 is measured separately and accounted for 2% of cases detected in July and 4% in August. The business model is not limited to selling alleged treatments. Channels can generate revenue from subscriptions, donations, books, supplements and platform monetization mechanisms, while sensational content can produce revenue even when the author does not directly market a product. EDMO cites as examples claims that magnesium treats ADHD and that mebendazole is a treatment for cancer. In both cases, real elements of research are extrapolated beyond what the available evidence demonstrates. The analyzed materials show that the problem is not limited to false information. Artificial characters, a professional appearance, local names and selective references to research can create the impression of genuine medical authority and make deceptive content more difficult to identify.

Investigations compiled by EDMO describe content production that goes far beyond a few isolated accounts. Research carried out by the Nordic hub NORDIS and featured by EDMO identified more than 1,000 channels with AI-generated medical characters on YouTube, collectively responsible for approximately 100,000 videos published in around 30 languages.

Some of these identities are designed for national audiences. In Denmark, characters with names such as Sofie Larsen, Louise Nygaard or Emma Poulsen present themselves in videos as specialists able to offer advice on prostate health, sexual health, nutrition or lifestyle. Sofie Larsen, one of the documented examples, is not a real person, and her image and materials are AI-generated.

The content is not uniform. Some recommendations may seem banal or even resemble general advice concerning sleep, alcohol or smoking. Other channels make scientifically unsupported claims, including promises to treat serious diseases, while some attempt to direct users toward products or supplements.

This combination makes the phenomenon more difficult to assess than an obvious falsehood. A profile may first publish seemingly plausible recommendations, use a credible medical identity and later introduce an unproven claim. EDMO also notes cases in which the reputation of well-known hospitals is used to create an additional appearance of legitimacy.

The scale of the phenomenon is also evident in the European network’s general monitoring. Health disinformation accounted for 7% of cases detected by EDMO in the EU in August, the same proportion recorded in July. These figures do not measure 7% of all health-related information on the internet or 7% of all platform content, but rather the share of this topic within the set of disinformation cases detected by the EDMO network.

Content about Covid-19 is tracked separately. It accounted for 2% of cases detected in July and 4% in August. In the broader health category, EDMO also identifies false claims about sunscreen, mammograms, vaccines or disease outbreaks, alongside content concerning alleged miracle cures.

This latter category most clearly illustrates how real information can be transformed into a conclusion unsupported by research. One example tracked by BENEDMO concerns claims on TikTok that magnesium could control ADHD symptoms or that magnesium deficiency could worsen the condition.

Some studies have found lower average magnesium levels in children with ADHD than in comparison groups. EDMO stresses, however, that this association does not demonstrate that magnesium deficiency causes ADHD or worsens symptoms, and study results are not consistent. Some research finds no differences or identifies higher levels, depending on the participants and the methods used for measurement.

Evidence concerning the use of supplements is also limited. A few small studies have reported improvements in behavior or attention, but some combined magnesium with other nutrients, making it difficult to separate the specific effect. EDMO’s conclusion is that larger and better-controlled studies are needed before magnesium can be recommended as a treatment for ADHD on the basis of this evidence.

A similar mechanism appears in the case of mebendazole, an antiparasitic drug promoted in viral posts as an alleged cancer treatment supposedly concealed by the medical industry. Some posts cite a Johns Hopkins University patent for a modified form of mebendazole and present the existence of the patent as proof that its effectiveness against cancer has already been established.

The real research is more nuanced. Mebendazole has produced results considered promising in laboratory and preclinical research and is being studied for possible repurposing in oncology. These results do not demonstrate, however, that the drug effectively treats cancer in humans, and its efficacy has not been established through sufficiently large clinical trials.

The difference between these two levels of evidence is frequently exploited by disinformation. The existence of a study, a patent or a laboratory result may be real, while the viral claim built on it goes beyond the research findings. In this way, a post does not need to invent every element to create a false impression.

Monetization completes the mechanism. Some channels identified in the investigations compiled by EDMO sell books, supplements or other products, solicit donations or use paid subscriptions. Other profiles may simply benefit from the visibility generated by viral content through platform monetization systems.

AI simultaneously reduces the costs required to multiply these identities. The same type of character can be adapted for different languages and markets, while videos, voices and scripts can be produced at volumes that would be difficult to achieve with real people. The fact that some artificially generated materials are labeled does not solve the problem when the character itself is designed to imitate the authority of a medical professional.

EDMO also points to the way these profiles present evidence as a warning sign. Pseudo-experts tend to offer certainty without acknowledging limitations or uncertainty, use emotional language and may select only studies that appear to support the claim while ignoring contradictory results.

The phenomenon described by the European network thus combines three problems that could previously occur separately. Artificial intelligence enables the fabrication of medical authority, social networks enable its rapid distribution, and commercial systems provide ways to turn users’ attention and vulnerability into revenue.

The figures of approximately 100,000 videos and more than 1,000 profiles do not measure all online medical disinformation and do not allow an estimate of the number of people who followed dangerous advice. They nevertheless document the ability to produce on a large scale a type of content in which users must determine not only whether the information is correct, but also whether the person presenting it actually exists.