Label Lab
AI checks 1,773 supplements from 15 countries for banned advertising claims.
Multivitamin
★★★★★ 4.8 (212)Strengthens your immune system
and reliably protects you
from colds.
Vitamin C and zinc contribute
to the normal function
of the immune system.
1,773 products · 15 countries · 18 banned claims found
How Label Lab checks a whole catalog
The proof
The problem
With protein powder and bars, the mistakes often sit in the data and in the advertising copy.
- Nutrition values are missing or do not add up
- An allergen is in the ingredients but not in the declaration
- Advertising claims the EU health claims regulation does not allow
The solution
- Rules in Python check gaps, contradictions, allergens and nutrition claims, free and always the same
- The AI sees only suspicious sentences and suggests an allowed wording
- One click applies the fix and the traffic light switches
Try it yourself
Click a product and apply the fix. The traffic light switches.
Try it: follow the golden hints.
To try it: click a product, apply the fix. This is the real dashboard with the results of the run. The brand RAWFORM and its texts are fictional. The Open Food Facts data is licensed under ODbL.
The setup
Python does the work that has to be exact. The AI only answers the question about language.
Decisions
- 0001
Rules first, AI only for language
Python calculates and compares exactly and reproducibly. In return, every new rule needs code.
- 0002
Prefilter for allowed claims
Python checks authorised EU claims and nutrition claims such as High Protein itself and never sends them to the models. That saves errors. In return, the list has to be maintained.
- 0003
The model only picks a rule ID
Legal basis and text come from the code, not from the model. In return, the texts sound standardised.
- 0004
Answers in a cache
Every AI answer is stored, no sentence is checked twice. In return, a changed text is checked again.
- Python
- FastAPI
- PostgreSQL
- n8n
- Claude Haiku
- Ollama (local)
- Docker