AI-Era Skincare Report Finds Widespread Gaps in Verifiable Product Claims

As consumers increasingly turn to artificial intelligence (AI), social media, and online search to guide skincare purchases, a new analysis suggests the information underlying those recommendations may be difficult to verify.

M11 analyzed 633 skincare brands and 1,470 high-impact product claims for its Trust Gap in Skincare report, evaluating whether claims could be verified against evidence accessible from the brand's product pages and surrounding public materials. According to the analysis, 47% of brands had no independently verifiable evidence behind any of the high-impact claims evaluated.

The findings highlight a potential challenge for clinicians as patients increasingly encounter confident product claims online: An AI-generated recommendation may be able to retrieve and repeat information about a product without necessarily determining whether the underlying claim is adequately supported.

Claims About Skin Conditions and Safety Raise Concerns

Among the 1,470 claims evaluated, M11 reported that 849 pointed to evidence that existed but was not published where the claim was made, 454 had no published proof or were contradicted by information on the product page, and 128 used regulated wording. Only 39 were categorized as having proof available on the page.

Importantly, the investigators stressed that a flagged claim should not automatically be interpreted as false. Rather, it indicates that a consumer, or an AI system reading the same publicly available information, could not verify the claim from the materials the brand made accessible.

Some of the findings involved claims with direct relevance to dermatologic care. Thirty-five percent of analyzed brands made claims about treating, curing, or preventing a condition that the report flagged as unverifiable from the page. Another 23% made allergen or “free-from” claims that could not be verified, while 12% made sun-protection claims without accessible testing.

One example involved a balm marketed as providing relief for eczema and psoriasis. M11 classified the language as crossing a potential drug or medicinal-product boundary because it referred to treatment of named medical conditions while the analyzed page did not provide evidence of medical authorization.

The report did not identify the brand associated with the example, as individual findings were anonymized.

Safety Information Was Also Frequently Missing

Claim substantiation was not the only issue identified.

M11 documented 250 serious product-data gaps across 214 of the 633 brands analyzed. These included missing certifications, incomplete ingredient lists, inconsistencies across sales channels, undisclosed concentrations of active ingredients, and unbacked claims.

Approximately one-third of brands had at least 1 serious data gap classified as having high or critical impact.

For dermatology clinicians, these gaps may be particularly relevant when patients are selecting products because of allergies, sensitive skin, active ingredients, or chronic inflammatory skin conditions. A recommendation engine can only evaluate information it can access; incomplete or contradictory product information may therefore carry forward into the recommendations consumers receive.

The analysis also found that information about the same product was not always consistent across platforms. According to M11, 35% of mass-market brands, 23% of mid-market brands, 20% of indie and challenger brands, and 19% of luxury brands had discrepancies in how a product was represented across a brand website, retailer, and marketplace.

Luxury Pricing Did Not Guarantee Stronger Evidence

Higher price points did not necessarily correspond with greater verifiability.

M11 created a 100-point Verifiability Index incorporating claims, evidence sources, and completeness of product data. The overall skincare category scored 52 out of 100, with average scores differing by only 4 points across mass-market, mid-market, indie and challenger, and luxury categories.

Luxury skincare had a particularly notable evidence gap. Sixty-eight percent of flagged luxury claims referred to a result without providing the supporting study, the highest proportion among the cohorts analyzed. Meanwhile, 37% of flagged claims among indie and challenger brands provided no evidence at all.

Across the entire analysis, only approximately 1 in 20 of the 2859 evidence links associated with evaluated claims could positively confirm a claim, according to M11.

What the Findings Mean in the AI Era

The findings become particularly relevant as consumers increasingly use generative AI systems to research products.

Unlike a dermatologist evaluating a patient's diagnosis, medical history, allergies, concurrent treatments, and individual treatment goals, consumer-facing AI tools may rely heavily on information available across brand websites, retailers, marketplaces, and other online sources. If those sources contain unsupported, inaccessible, outdated, or contradictory information, an AI-generated answer may reproduce those limitations.

However, the report does not demonstrate that AI shopping tools routinely provide incorrect skincare recommendations. M11 primarily evaluated the accessibility and verifiability of the information available online rather than systematically testing the accuracy of AI-generated recommendations.

That distinction is important. The findings instead identify a potential weakness further upstream: the information environment from which both consumers and AI systems retrieve product claims.

Methodology and Limitations

The analysis included brands with an established presence in the UK skincare market and divided them into mass-market, mid-market, indie and challenger, and luxury cohorts. M11 used a multimodel scanning approach incorporating models from Anthropic, OpenAI, Google, Perplexity, and xAI to analyze brands' primary products and related public materials.

The analysis was depth-normalized rather than designed as a census of every claim made by every brand. Approximately 2 high-impact claims were evaluated per brand. A blinded sample of 180 automated classifications was subsequently reviewed manually, producing 77.8% agreement across the 5 classification labels and 86.7% agreement for the high-risk binary classification underlying the report's headline finding.

The authors also acknowledged that the study represents a point-in-time assessment and that public webpages can change. Brands were not contacted for comment, and brand-level findings were not published. Most importantly, the analysis evaluated verifiability rather than truthfulness; an inaccessible or inadequately substantiated claim was not necessarily determined to be false.

For dermatology clinicians, that limitation may also capture the report's most relevant takeaway: When patients encounter skincare recommendations online—whether through social media, search engines, retailers, or AI—a confidently presented claim should not necessarily be interpreted as an evidence-backed one.

References

1. Advertising Standards Authority. (2026, April 22). ASA ruling on L'Oréal (UK) Ltd t/a Garnier. https://research.m11.ai/trust-gap-in-skincare/
2. M11 Labs. (2026, September). The trust gap in skincare.https://www.asa.org.uk/rulings/l-oreal-uk-ltd-a25-1316838-l-oreal-uk-ltd.html
3. U.S. Food and Drug Administration. (2019, December 6). Public safety notification on exosome products. https://www.fda.gov/vaccines-blood-biologics/safety-availability-biologics/public-safety-notification-exosome-products