nemi uses AI (artificial intelligence) for some features. This page is an honest breakdown of where we use it, where we don't, and what to expect (or not expect) from AI-generated content in the app.
1. Where we use AI
Ingredient explanations
The "description" and "research notes" sections for each ingredient are partially generated by Anthropic's Claude language model. The model works from public sources — EWG Skin Deep, cosmetic-ingredients.org, the Japan Cosmetic Industry Association's INCI registry, and other public data — to produce both Japanese and English text.
Product classification
For new products being added, AI drafts the category (skincare / makeup / etc.) and English name. Final classification is reviewed internally.
2. Where we don't use AI
- Safety score calculation: scores are computed by a deterministic algorithm using public toxicity data and rules we've defined. AI does not assign points.
- Ingredient safety rating (Safe / Caution / Avoid): based on EWG Skin Deep scores and clinical data, not AI judgment.
- Analyzing your data: your scan history, collections, searches — none of this is used to train AI models. We do not send your personal data to Anthropic.
3. Accuracy and limitations
AI-generated content can occasionally include:
- Explanations based on outdated research
- Confusion between similar-sounding ingredients
- Translation nuances that lose detail
We review generated content regularly and fix obvious errors, but we can't guarantee 100% accuracy. For important decisions (use during pregnancy, allergies, etc.), please consult a doctor or pharmacist.
4. Transparency
- If you find an error, please report it via the in-app "Report" button.
- To know what sources informed a specific ingredient's content, email support@nemi.beauty.
- We'll notify you on this page before any change to how we use AI.
5. Contact
Questions or concerns about AI use: support@nemi.beauty.