NUA

AI Transparency

Last updated: 15 September 2026

NUA Nutrition uses AI for five things, and labels every one of them: meal reviews, meal suggestions, nutrition estimates from free text, photo recognition, and the weekly reflection. Wherever a text was written by AI, the app marks it visibly as "AI-generated · may contain errors" (EU AI Act, Art. 50). All AI features can be switched off in one tap in the app's settings — the app then falls back to its built-in food database and curated content, labelled honestly as such. AI requests are answered by a stateless server that stores nothing and never learns who is asking.

Why this page exists

The EU AI Act (Regulation (EU) 2024/1689) expects apps to be open about where AI is involved, so you are never left wondering whether a machine wrote something.

Where it is used

recognising what is on a plate from a photo you take;
estimating nutrition from a sentence you type;
writing short pieces of text about what you logged;
proposing something to eat when you ask for an idea.

The model behind these is Claude, by Anthropic. What is sent for each is set out in the privacy policy.

Where it is never used

Vitamin and mineral values are looked up, not predicted. They come from the German Federal Food Code (BLS 4.0, Max Rubner-Institut). Reference values follow the published DGE/DACH figures for adults. All calculations run on your device by fixed rules.

Calories and macronutrients are looked up too wherever the food database or a product label has them. Only where it does not — a sentence you type while the AI features are on, or a photographed ingredient the database cannot find — does the AI estimate them, and the app labels those figures as AI estimates.

How it is marked

Every sentence an AI wrote carries a visible label in the app saying it is AI-generated and may contain errors. Figures derived from a photo or a typed description are marked as estimates wherever they appear.

Checked before you see it

When the AI suggests something to eat and says what it gives you, the app checks that claim against its own food database and drops the suggestion if the data does not support it. Written meal reviews and the weekly reflection are checked against a list of things they must never say — such as health claims, calorie talk or judgements about food — and are replaced by a plain line if they break those rules. These checks are careful, but they are word checks, so the "may contain errors" label stays on every AI text.

What it never does

No automated decision-making with legal or similarly significant effects about you (Art. 22 GDPR).
No emotion recognition and no biometric categorisation.
No profiling and no prediction of your health outcomes.
No diagnosis and no medical advice. See the terms of use.
Nothing you log is used to train any model, by me or by Anthropic.

Switching it off

One switch in the app's settings turns all of it off. Nothing is then transmitted, and the app continues to work on your device alone. If the service is simply unavailable, the app falls back the same way rather than showing an error.

Questions

Write to anna.gaertner@whu.edu.