A person's face being analyzed by an AI skin-scanning overlay
“AI” in beauty and wellness spans three very different things: consumer-facing skin scanners and advisor chatbots, personalization engines, and back-end research tools. Only some of them are backed by independent evidence, and the marketing rarely tells you which is which.

Almost every beauty and wellness brand now has an AI story. Point your phone at your face and an app returns a “skin age” and a percentage score for wrinkles, pores, and spots. Answer a quiz and a “custom” serum is formulated for you. Ask a chatbot which retinol to buy, or how to fix your sleep. The pitch is consistent: artificial intelligence brings dermatologist-grade analysis and personalization to your bathroom, for free.

The problem is that “AI” is doing a lot of work in that sentence. The same two letters cover a validated cancer-triage algorithm published in Nature, a consumer skin-scanner whose “accuracy” comes from a company press release, and a wellness chatbot that a Stanford team found gave dangerous responses in a crisis. Sorting the genuinely useful from the marketing theater — and the quietly risky — is the whole exercise.

What “AI” Actually Means Here

It helps to separate three categories, because the evidence is completely different for each:

The marketing tends to borrow the credibility of the third category (real science) to sell the first (consumer features). Keeping them apart is the key to reading the claims honestly.

Skin-Analysis Scanners: The Credibility Borrow

The foundational result everyone points to is real. In 2017, Esteva and colleagues trained a convolutional neural network on 129,450 images and showed it could classify skin cancer at a level matching 21 board-certified dermatologists (Nature, 2017). That is a genuine milestone. But it is a clinical cancer-classification task on curated images — not the consumer “scan your face for wrinkles and pores” use case, and the two are routinely conflated in marketing.

Two caveats matter. First, algorithmic performance drops sharply once you leave the clean training datasets for the real world, a point made directly in a Journal of Investigative Dermatology “reality check”. Second, most training sets over-represent lighter skin (Fitzpatrick I–III), so tools tend to underperform on darker skin.

As for the consumer scanners themselves: the widely quoted accuracy figures for tools like Perfect Corp’s YouCam come from company press releases, not independent validation. There is some peer-reviewed work on device-based facial-skin analysis in the Journal of Cosmetic Dermatology, but it establishes reproducibility of surface measurements — not that an app equals a dermatologist. Neutrogena retired its Skin360 hardware in favor of phone-camera AI and now licenses an engine from Haut.AI, a B2B provider whose “research” is largely publicity rather than independent trials.

Personalization: A UX Layer, Not a Clinical Upgrade

“Custom” skincare and supplements — from brands like Prose and Function of Beauty — are marketed as algorithmically tailored to you. The evidence that questionnaire-driven personalization outperforms a well-matched off-the-shelf product is thin. A 2025 review in Frontiers in Artificial Intelligence on AI, genomics, and personalized skincare describes an emerging field, not a proven one, and consumer testing (Allure) repeatedly finds the personalization is more about experience than measurable outcomes.

The active ingredients and their concentrations do the work, and those are not unique to “custom” brands. Personalizing the packaging and quiz flow is a legitimate user-experience choice; it is not, on current evidence, an efficacy upgrade.

Generative AI in the Marketing Itself

AI increasingly generates the advertising, not just analyzes your skin. AI-generated “models” appeared in a 2025 Vogue Guess campaign, prompting significant backlash over unrealistic beauty standards and disclosure. Research suggests AI-generated ads can backfire for premium brands.

The bigger consumer-protection issue is fake and AI-generated reviews and endorsements. In 2024 the FTC finalized a rule banning fake reviews and testimonials, with civil penalties, and ran Operation AI Comply against deceptive AI claims. Virtual influencers get no special carve-out; the existing Endorsement Guides still apply.

The Claim

“Our AI gives you dermatologist-level skin analysis, a formula personalized just for you, and a smart wellness advisor — all from your phone.”

(Composite representative claim; reflects skin-scanner, custom-formula, and AI-advisor marketing language across the beauty and wellness category.)

AI Wellness Advisors: Where It Gets Risky

The wellness side is where overpromising stops being merely annoying and starts being unsafe. On mental health, a 2025 Stanford study found that large-language-model “therapy” chatbots expressed stigma and gave dangerous responses, including failing to respond appropriately to crisis cues.

On symptom checking, the track record is long and consistent. A landmark BMJ evaluation (Semigran et al., 2015) of 23 symptom checkers found the correct diagnosis was listed first only about a third of the time and triage advice skewed toward over-caution. A 2022 systematic review in npj Digital Medicine reached similarly modest conclusions. These tools can be a starting point; they are not a clinician.

Where AI Genuinely Delivers: The Back End

The most substantive use of AI in this industry is the one consumers never see. Machine learning for predictive modeling of cosmetic-ingredient safety and tolerability and formulation design is a real, incremental accelerant in R&D. That is ordinary computational chemistry and toxicology — useful, unglamorous, and largely honest, because no one is selling it directly to you. The consumer-facing “our AI discovered this ingredient” messaging is mostly branding layered on top of that routine work.

The Privacy Cost of a “Free” Scan

A free AI skin analysis is not free: you are uploading biometric facial data. Under Illinois’ Biometric Information Privacy Act (BIPA), this has produced real liability — Kenvue agreed to a multi-million-dollar settlement over Neutrogena’s Skin360 app, and a wave of BIPA class actions has hit AR try-on and skin tools. Because a handful of vendors power many brands’ scanners, your face data may travel further than the brand’s name suggests.

Use Case Evidence Level What the Data Show
Clinical skin-cancer triage (regulated) Strong Dermatologist-level classification on curated images; performance drops in the real world
Back-end formulation / toxicology modeling Moderate Genuine, incremental R&D accelerant; peer-reviewed reviews, not consumer-facing
Consumer skin scanners / “skin age” Weak Reproducible surface metrics; no independent proof of dermatologist-equivalence; skin-tone bias
“Personalized” formulas Weak Mostly UX; no strong evidence of better outcomes vs matched off-the-shelf products
AI symptom checkers Weak / Risky ~1/3 correct-first-diagnosis; systematic over-triage (BMJ 2015)
AI mental-health / therapy bots Risky Documented stigma and dangerous crisis responses (Stanford 2025)

What the Evidence Actually Shows

AI in beauty and wellness is genuinely useful in two places most consumers never touch: regulated clinical triage and back-end R&D. The consumer-facing features that carry the loudest claims — skin scanners and “custom” formulas — are the least validated, and they lean on the credibility of the clinical research to sell what is largely a reproducible surface reading or a UX layer. On the wellness side, symptom checkers and therapy chatbots have documented accuracy and safety problems. The technology is real; the evidence for the specific consumer promises is not there yet.

What This Means for Consumers

Treat an AI skin scan as a mirror with a scoreboard, not a diagnosis. It can be a fun, roughly consistent way to track surface changes over time, but its “skin age” and percentage scores are not clinically meaningful, and anything that looks like a medical concern belongs with a real dermatologist.

On “personalized” products, buy them if you like the experience, but judge them the way you would any product — by their actual ingredients and concentrations, which you can compare against cheaper standardized options. On AI wellness advisors, use them for general information and convenience, never for crisis support or diagnosis; the published evidence says they are not ready for that. And before you upload your face for a free scan, assume the image is biometric data that may be stored and shared, and decide if the novelty is worth it.

Verdict: Mixed Evidence

AI earns real marks for validated clinical triage and for the unglamorous back-end R&D that genuinely speeds formulation and safety screening. It loses points because the consumer-facing promises that dominate the marketing — dermatologist-level scanners and efficacy-boosting personalization — lack independent validation, carry skin-tone bias, and collect biometric data, while AI wellness advisors have documented accuracy and safety failures. A powerful set of tools, unevenly matched to the claims made for them. Evidence rating: 3/5.

References & Further Reading