It is easy to finish watching a skincare video with several products in mind and only a vague idea of whether they suit you. Asking AI to explain the routine, find cheaper alternatives, or put the steps in order seems sensible. But the advice it gives still depends on what it's been asked to work with.
Research presented at the European Academy of Dermatology and Venereology Congress in Vienna examined 38 TikTok videos about Korean skincare and “glass skin,” the term used to describe clear, smooth, deeply hydrated-looking skin. According to the congress release, almost 95 percent contained at least one potentially misleading claim.
Researchers selected videos through five search terms in May 2026, using an account without previous engagement history to reduce personalisation. This was a small sample of selected videos. The findings cannot tell us how much Korean skincare content on TikTok is misleading overall.
What was missing from the videos was revealing. The release reported that 84 percent did not account for different skin types, while 32 percent promised instant results. None cited an external source or reference.
“What surprised us most was how consistently basic safety information was absent,” said lead author Dr Maria Lúcia Paysano.
The researchers stressed that Korean skincare itself was not their concern. They were examining how routines were presented and whether viewers were given enough information to understand their limitations.
No AI involvement was established in the study. Its relevance comes at the next step, when someone takes a recommendation from a video and asks AI to make it suitable for them.
That could be useful. A search tool can help find reputable information or explain unfamiliar terms. Someone unsure about a recommendation could use it to prepare questions for a qualified professional, rather than buying everything shown in the video.
A detailed answer, however, can feel more personal than it really is. Give an AI tool an age, a budget, and a description of skin concerns, and it may return a complete routine. Using those details doesn't mean the advice is appropriate. Generative AI can produce plausible but wrong explanations.
The image that prompted the search also matters. AI-generated faces and altered photographs can show a result that no product has achieved. A synthetic testimonial can describe an experience no customer has had. Those possibilities make the origins of a demonstration worth considering before treating it as evidence.
The ASA’s guidance on AI advertising says disclosure is needed where consumers could mistake generated material for a real person, experience, or endorsement. An AI label would not, by itself, resolve a misleading impression.
For beauty brands and retailers, there are useful ways to employ the technology, from organising product information to drafting explanations for review. Checking those explanations against the evidence matters as much as checking the finished image. A small wording change can leave a customer expecting something the product cannot support.
Consumers do not need every skincare search to become a research project. AI should make the information easier to navigate. If it gives us a longer shopping list without explaining why the products belong on it, we may be no better informed than when the video ended.
Read more here.
