Online health information is considered reliable when it is based on data published in peer-reviewed scientific journals, clearly identifies its author and sources, and does not pursue any disguised commercial objectives. This operational definition allows for a quick sorting of the majority of content encountered on the web, social media, or video platforms.
The volume of dubious medical content has significantly increased in recent years, driven by short videos and automatic text generation tools. According to a survey by the Guardian published in July 2026, AI-generated health content appears in 40% of the top health videos on TikTok. In the face of this acceleration, knowing how to evaluate a source becomes an individual protective skill.
AI-generated health content: a massive problem
Competitors approach misinformation from the angle of classic fake news, shared by humans on Facebook or Twitter. The phenomenon has changed in nature. AI now allows for the production of videos featuring fake digital doctors reciting invented medical advice in a credible format.
An investigation by Maldita.es, published on July 31, 2026, identified 1,046 YouTube channels disseminating dubious or misleading health content. The majority of these channels were created between January 2025 and July 2026, confirming a recent acceleration rather than a residual phenomenon.
The problem is not limited to videos. Entire articles, written by language models, circulate on sites that mimic the layout of recognized medical portals. Checking health information on Skeptic North helps identify some of these fabricated contents, but individual vigilance remains the first filter.
Identifying the author and their medical qualifications
The first reflex when encountering health content is to look for who produced it. An article signed by a doctor, pharmacist, or researcher affiliated with a verifiable institution carries more weight than an anonymous text or one signed with a pseudonym.
Three concrete checks can help determine this:
- Is the author identifiable by their full name and specialty? An RPPS number (for French health professionals) or a university affiliation are strong indicators.
- Does the site display an editorial policy or a scientific committee? Institutional portals (HAS, Inserm, sante.fr) systematically publish this information.
- Does the author have financial ties to a dietary supplement manufacturer, a laboratory, or a coaching platform? An undisclosed conflict of interest disqualifies the source.
The National Council of the Order of Physicians published a charter in January 2025 with 10 principles governing the presence of doctors on social media. It mandates clear identification, citation of sources, prohibition of personalized online advice, and prohibition of promoting one’s own activity through these channels. A doctor who does not adhere to these principles is outside the ethical framework.
Checking the sources cited in a health article
Reliable content refers to identifiable publications. Not to “a recent study” without a link, nor to “researchers” without names. The absence of a precise source is the most reliable warning signal.
When a reference is provided, it should be traceable. An article that cites “a study published in a major medical journal” should allow for that study to be found on the journal’s site. If the link is dead, if the journal does not exist, or if the study does not say what the article claims, the content is disqualified.
The trap of misrepresented studies
Some content cites real studies but distorts their conclusions. A preliminary study on 20 patients becomes “science has proven that.” An in vitro result (in the lab) is presented as a validated treatment in humans.
Checking whether the study involves humans, the sample size, and the level of evidence helps to thwart this type of manipulation. Meta-analyses and randomized controlled trials carry more weight than an isolated observational study.
Warning signals in the tone and structure of the content
Beyond the author and sources, the form of the content provides clues. An article that promises the cure for a chronic illness, uses emotional vocabulary (“scandal,” “conspiracy,” “miracle cure”), or consistently pits conventional medicine against “natural alternatives” deserves increased skepticism.
Several characteristics signal problematic content:
- The text offers a product to buy at the end of the article or via an affiliate link. The commercial intent skews the information.
- The publication date is missing or very old. Medical recommendations evolve, and an article from 2016 on an active topic may be outdated.
- The content contradicts medical consensus without providing solid data to justify this divergence. Consensus can evolve, but it does not overturn based on a YouTube video.
The case of personal testimonies
An individual testimony (“this treatment cured me”) has no probative value in medicine. It may be sincere without being generalizable. Survival, confirmation, and regression to the mean biases explain most of these accounts. A testimony never replaces published clinical data.
The proliferation of AI-generated content makes these checks more necessary than ever. A fluid, well-structured, and visually professional text no longer guarantees anything about the quality of the information. The tools have changed, but the evaluation criteria remain the same: qualified author, traceable source, absence of conflict of interest, consistency with available data. It is these four filters, applied systematically, that separate useful information from potentially dangerous content.
