ComfiSkin vs ChatGPT and AI chatbots
A chatbot will answer any skincare question with total confidence, about a bottle it has never seen, for skin it doesn’t know. ComfiSkin answers from the ingredient list of the product in your hand, checked against your skin and your shelf.
Is my new moisturiser OK for sensitive skin?
Great choice! It’s fragrance-free and gentle, so it should suit sensitive skin.
How do you know it’s fragrance-free?
Good question. Many creams like it are fragrance-free.
The label lists “parfum”, fourth from the end.
The short answer
Chatbots sound sure of themselves, but they haven’t read your bottle and don’t remember your skin. Researchers testing 14 AI models found them unreliable on cosmetic chemistry. And the photos you send can be read by human reviewers or used to train the AI.
What that means: you can get a confident wrong answer, and hand over a photo of your face for it. Ask a chatbot general questions if you like. For “is this product OK for my skin?”, use ComfiSkin, which reads the real label and already knows your skin.
Side by side
| ComfiSkin | General chatbots | |
|---|---|---|
| Where the answer comes from | This product’s ingredient list | General knowledge, and a confident guess |
| Knows your skin and your shelf | Yes, every time | Only what you type into that chat |
| Reliable on cosmetic ingredients? | It’s built for exactly that | Researchers found performance “poor”1 |
| Your chats used to train AI | No | On by default in ChatGPT4 |
| Does it check your products work together? | Yes, across your week | No |
| Writes poems | No | Yes |
Sounding sure is the problem
A study published in July 2026 tested 14 AI language models on cosmetic chemistry and skin health. “Overall performance was poor,” the authors wrote, concluding that these models are “currently not reliable sources of cosmetic chemistry information”.1 Doctors are worried too: in a 2024 survey of 192 US dermatologists, 89% named misinformation as a concern about AI chatbots.2
A wrong answer that sounds unsure is easy to ignore. A wrong answer delivered with total confidence gets acted on. Ask about “my new moisturiser” and a chatbot will happily tell you it’s gentle and fragrance-free, or that two products are fine together, without having seen either label. That’s how people end up putting the very thing they’re sensitive to on their face.
ComfiSkin reads the ingredient list of the actual product and checks it against your skin, so the answer is about this bottle, not bottles in general.
What that means for you: a chatbot can tell you a cream is fragrance-free when the label says otherwise, and sound completely sure about it.
The same questions, answered both ways
Because ComfiSkin knows your skin, your products and your week, it gives a straight answer where a chatbot has to hedge or guess.
| Question | A typical chatbot answer | ComfiSkin |
|---|---|---|
| Is there fragrance in this? | Many products like it are fragrance-free. | Yes: parfum, linalool and limonene. You avoid fragrance. |
| Can I use my retinol tonight? | Retinol is usually best at night, a few times a week. | Not tonight, you used your acid toner. Wednesday is clear. |
| Do I need this new cream? | It sounds like a good option for hydration! | No. It does the same job as the cream you own. |
| When did I open my sunscreen? | I don’t have that information. | 12 January. It’s good until next January. |
The answers above are illustrations.
What that means for you: one answer you can act on tonight, instead of general advice you still have to apply yourself.
Think twice before sending a photo
People send chatbots photos of breakouts, rashes and their faces. Google says some Gemini chats are read by human reviewers and kept for up to three years, even if you delete your activity. Its own help page warns: “Please don’t enter confidential information that you wouldn’t want a reviewer to see.”3
In ChatGPT, a setting called “Improve the model for everyone” lets your chats train OpenAI’s models, and it’s on by default.4 So you’re giving a general-purpose tech company very personal information in exchange for a generic answer. ComfiSkin never asks for a photo of your face, and your skin notes and photos stay on your phone.
What that means for you: that photo of your breakout may be seen by a stranger and kept for years.
What to do if you ask a chatbot about skincare
- Don’t trust an answer about a specific product unless the chatbot has seen its full ingredient list, and even then, double-check.
- Turn off training. In ChatGPT, open Settings, then Data Controls, and switch off “Improve the model for everyone”. Don’t send photos you wouldn’t want a stranger to see.
- Use ComfiSkin for the product questions. It reads the real label and checks it against your skin and routine.
Sources
- “Accuracy and reliability of large language models in cosmetic chemistry and skin health: a benchmarking study”, arXiv preprint 2608.14631, July 2026. arxiv.org
- Journal of Drugs in Dermatology, “Physician opinions on artificial intelligence chatbots in dermatology: a national online cross-sectional survey of dermatologists”, 2024. jddonline.com
- Google, Gemini Apps Privacy Hub. support.google.com
- TechRadar, “Your ChatGPT data automatically trains its AI models unless you turn off this setting”. techradar.com
More comparisons
Ask about the bottle in your hand
An answer from the real label, checked against your skin.
- Free, all of itNo plans, trials or paywalls. Every feature, for everyone.
- No sign-upNo account, no email, no password to lose.
- No ads, no trackingNothing about you is sold, shared or followed around the web.
- Your notes stay on your phoneWith a password-protected backup that only you control.
- Never asks for your faceIt reads the label, not you.
- Answers for your skinYour skin type, concerns, avoid list and routine, every time.