What AI does in skin care

  • Image guidance: helping you take usable photos: framing, focus, lighting.
  • Initial assessment: suggesting likely conditions from photos and answers, for a clinician to confirm or reject.
  • Triage: flagging cases that look urgent so a clinician sees them sooner, or that need in-person care.
  • Tracking: comparing photos over time to show whether a treatment is working.
  • Paperwork: structuring the history so the clinician spends their time on judgment.

What the research shows

A widely cited 2017 study trained a deep neural network on about 129,000 clinical images and found it performed on par with 21 board-certified dermatologists on two narrow tasks: telling skin cancers from look-alike benign lesions in biopsy-proven images.[1]

A 2022 study then tested leading dermatology AI models on a new, diverse set of images that had been confirmed by biopsy. Performance fell well below what the models had reported, and was worse on dark skin tones and on uncommon diseases.[2]

The lesson from both: strong results on the data a model was built with don't guarantee the same results on your skin, in your phone's photos.

Its limits

  • Skin tone: models trained mostly on lighter skin can do worse on darker skin.[2]
  • Photo quality: blur, shadows and filters change what the software sees.
  • What a photo can't show: texture, depth, warmth, and how a spot has changed over months. A Cochrane review of photo-based skin cancer diagnosis found the evidence limited and a risk of missed melanomas, even with clinicians reading the images.[3]
  • Rare and serious conditions: software is weakest where errors cost most.
  • Regulation: the FDA authorizes some AI-enabled medical devices and keeps a public list of them; a consumer app may not be on it.[5]

Why clinician sign-off matters

The World Health Organization's guidance on AI in health puts human oversight and accountability at the center: people, not systems, stay responsible for medical decisions, and AI should be inclusive and equitable.[4]

The AAD's standards for teledermatology call for care directed by board-certified dermatologists, an adequate history and examination before prescribing, and transparency about the credentials of the clinician advising you.[6] A licensed clinician who reviews the AI's suggestion can catch the cases the software gets wrong, ask a question it didn't, and send you in person when photos aren't enough. They also carry professional and legal responsibility for the decision; the software does not.

Questions to ask an AI service

  1. Does a licensed clinician review every diagnosis and every prescription, or only some?
  2. Can I see the name and license of the clinician who approved my plan?
  3. How was the software tested, and on which skin tones?
  4. What happens when the software or the clinician isn't sure?
  5. Are my photos used to train the software, and can I opt out?

Example: how Nolla splits the work

Nolla describes a clear division. Its software handles the initial assessment of your photos and answers, and a licensed U.S. medical provider reviews every diagnosis and prescription.[7] When the scan is unsure, Nolla says it guides you to the next step, including additional review or in-person care; it also says clear, well-lit photos improve accuracy.[7] Nolla says every diagnosis and prescription is guided by a licensed U.S. medical provider.[8] That is the model this guide recommends: AI for speed and tracking, a clinician for the decision.

Its limits are the ones above: it is built for acne, rosacea and redness, hyperpigmentation, melasma and wrinkles, not for a changing mole or an urgent rash. See where it sits in the directory.

Questions

Can AI diagnose skin conditions?

In research, image-based AI has matched dermatologists on narrow tasks with curated images.[1] On a more diverse image set, performance dropped, especially on darker skin tones and uncommon conditions.[2] It should support a licensed clinician, not replace one.

Is a skin-checking app enough to rule out melanoma?

No. A spot that is changing or matches the ABCDE warning signs needs an in-person exam.[9,3]

Sources

Every numbered claim on this page points here. Providers' own pages were read in November 2026; prices and terms change, so check them again before you pay.

  1. Dermatologist-level classification of skin cancer with deep neural networks (2017) · Nature
  2. Disparities in dermatology AI performance on a diverse, curated clinical image set (2022) · Science Advances
  3. Teledermatology for diagnosing skin cancer in adults (Cochrane review, 2018) · Cochrane Database of Systematic Reviews
  4. Ethics and governance of artificial intelligence for health · World Health Organization
  5. Artificial Intelligence-Enabled Medical Devices · U.S. Food and Drug Administration
  6. AAD Teledermatology Standards · American Academy of Dermatology
  7. Nolla frequently asked questions · Nolla
  8. Nolla Health · Nolla
  9. What to look for: ABCDEs of melanoma · American Academy of Dermatology

Keep reading

How teledermatology worksStore-and-forward, live video and AI-assisted care, step by step.What it can and can't treatAcne, rosacea, eczema, pigment, wrinkles, moles and rashes, and when to go in person.How to compare servicesA six-point checklist and a directory of services judged on it.Costs and insurancePer-visit versus subscription, insurance questions and what to ask before paying.Prepare for your visitPhotos, symptoms, medications, and how to read the care plan you get back.