How IngrediScan decides
Short answer. IngrediScan reads the label text on your phone. It matches each ingredient to a fixed dictionary and applies fixed rules for the profile you chose. There is no AI verdict. The same text always gives the same result, and each result shows its reason and sources.
1. Read the text
The camera reads the ingredient list with on-device text recognition. The app does not save camera frames and does not send them anywhere. The app joins words across several frames, so a shaky hand or a curved pack still gives one clean line.
2. Match ingredients
The app first tries exact matches against names, E-numbers and INCI names in 21 Latin-script languages. If a word does not match, it tries recovery passes for common reading errors: swapped letters, split or merged words, clipped edges, and garbled codes such as "E62l". The longest match wins, so "sodium benzoate" counts as one ingredient and not as two.
3. Read the context
- Ingredient list. Text before the list header, such as a product name, is treated with care. A hit there is shown only if it is an allergen from your profile.
- Negation. Phrases such as "no added sugar" or "gluten-free" do not count as a hit.
- "May contain". Entries after such a phrase get a Trace tag. If you have an allergy, treat them as present.
- Allergen families. One allergy in your profile covers all names in its family. For example, "casein" counts for a milk allergy.
4. Rate per profile
Every ingredient has fixed rows for these profiles: adult, child, infant, dog and cat. A row has:
- a level: Low, Medium, High or Critical,
- a basis: a ban, limit or official classification (Regulatory), published studies (Evidence), or limited and mixed data where the level errs on safety (Precautionary),
- the possible concerns and a short reason,
- the sources behind it.
A younger profile never gets a lower level than an adult for the same ingredient. A pet row can differ from a human row only when a documented reason exists.
5. Show sources
Sources are official texts, regulators, scientific papers and practical references such as veterinary pages. Each link is tagged by type. You can open it from the ingredient detail screen. The website pages for sodium nitrite, titanium dioxide and xylitol show the same data.
What the app does not do
- No diagnosis or medical advice. It rates ingredients. It does not tell you what is wrong with a person or pet.
- No guarantee. Text recognition can miss part of a label. No result does not mean the product is safe.
- No brand score. The app does not rate products or brands. It rates what is in them.
- No cloud lookup for scans. Scanning works without internet. A barcode lookup goes online only when you ask for it.
Sources and review
The ratings follow EU regulations, opinions of EFSA and other regulators, IARC classifications and published studies. We review the rating data on a schedule and show the review date on these pages. Questions or corrections: [email protected].
Ingredient pages
The assessment is for information only. It does not replace the manufacturer's label or advice from a doctor or vet. Text recognition (OCR) can miss part of a label. No result does not mean the product is safe.