Key takeaways
A chatbot's name for a spot is not a diagnosis.
- In one dermatologist's test, two AI chatbots named the right diagnosis for about 4 in 10 biopsied skin spots.
- Both chatbots called one melanoma on an ear a harmless seborrheic keratosis and said it could be sent home.
- An older image model trained on skin photos got about 6 in 10 right. It is a research baseline, not a tool for the public.
- Asked as a patient, both chatbots said to see a doctor in almost every case.
- This is one doctor's published test, not a peer-reviewed study. A new or changing spot still needs an in-person check.
Not reliably. A skin-cancer surgeon gave two AI chatbots photos of 200 biopsied skin spots. They named the right diagnosis less than half the time. Both also called one melanoma a harmless seborrheic keratosis.
Many people now photograph a spot and ask a chatbot what it is. The test also asked the chatbots a second question: do I need a doctor? Their answer to that one is worth knowing before you trust a chatbot's label.
What did the dermatologist test?
Dr. Magnus Lynch tested whether AI can name a skin spot from photos. He is a Consultant Dermatologist and Mohs surgeon at Guy's Hospital in London. He is also an Honorary Reader at King's College London. He posted the results on his lab's website on October 25, 2026 (Lynch Lab). GIGAZINE and Digital Today reported on it the next day.
It is one doctor's published test, not a peer-reviewed study.
He picked 200 spots at random from MILK10k, a public image set from the International Skin Imaging Collaboration. A biopsy had confirmed every one. Each spot had two pictures: a regular photo and a close-up through a skin magnifier.
Half the spots should not be sent home. They were 30 melanomas, 25 basal cell carcinomas, 25 squamous cell carcinomas and 20 actinic keratoses. The other 100 were harmless: 50 benign moles and 50 seborrheic keratoses.
He tested three systems. Two were general AI chatbots, GPT-6 Astra and Claude Fable 5.1. The third was Google's Inception v3, an older image model fine-tuned on 68,011 skin images.
How often did the chatbots get it right?
The chatbots named the right diagnosis first for 84 and 85 of the 200 spots, about 42%. The specialized image model got 125 right, about 62%.
- Seborrheic keratosis (50 spots): the chatbots named 13 and 6. The specialized model named 26.
- Melanoma (30 spots): the chatbots named 8 and 11. The specialized model named 13.
One case stands out. A thin invasive melanoma on an ear went to both chatbots, and both called it a seborrheic keratosis. Both said it could be discharged. A seborrheic keratosis is a common, harmless growth, the kind many people want gone for looks.
Both chatbots called one melanoma a harmless growth.
Did the chatbots tell people to skip the doctor?
No. Dr. Lynch also asked the way a patient would, "do I need to see a doctor?", with one photo. Both chatbots advised seeing a doctor in almost every case, harmless spots included.
So the chatbots did not steer people away from care. The weak point was the name they gave a spot. If someone trusts a wrong name and acts on it, a skin cancer can end up treated like a harmless growth.
What are the limits of this test?
Dr. Lynch lists the limits himself:
- The numbers are small. There were only 30 melanomas.
- Each chatbot answered once.
- The cases were hard ones, all sent for biopsy.
- The public images may have been in the chatbots' training data.
- The image model is a baseline, not the best available. He notes that regulated products do better.
A bigger study could shift the percentages. In this one, every system got some spots wrong, including melanomas.
What does this mean for you?
A chatbot's label is not a diagnosis. Seborrheic keratoses, moles and early skin cancers can look alike in a photo. Do not use a chatbot, app or AI tool to decide whether a spot is safe.
Get an in-person check before anyone treats a spot that is:
- new, or changing in size, shape or color
- irregular, or dark
- bleeding or itching
- different from your other spots
Any mole is examined in person by a dermatologist before at-home removal is even considered. To see how the three spots in this test differ, read our guide to seborrheic keratosis vs melanoma vs mole.
A quick check before you treat any spot
Most skin spots are harmless. Finding out which one you have is usually a short visit. A dermatologist looks at the spot in person, can view it under a magnifier, and can order a biopsy if it needs one. A photo sent to a chatbot cannot do that.
Identify first, treat second. If you are not sure what a spot is, a dermatologist is the person to ask.
Frequently asked questions
What readers ask most about AI chatbots and skin spots.
AI chatbots, skin spots and when to see a dermatologist
↓ Tap each question to reveal the answer.
The bottom line
In one dermatologist's test of 200 biopsied spots, general AI chatbots named the right diagnosis about 4 in 10 times. Both called one melanoma a harmless growth. Most spots are harmless, and knowing which one you have starts with a person who can look at it.
Source: Magnus Lynch, "A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis," Lynch Lab, October 25, 2026. Reported by GIGAZINE and Digital Today, October 26, 2026.
This article is general information and is not a substitute for medical advice or diagnosis. If a spot is new, changing or you are unsure, check with a dermatologist.
