How TrustPick scores products
Exactly how a label becomes a result: where the information comes from, what each result means, what we leave out, and the full Health Score formula.
Last updated 2026-10-05
1. Sources
Each ingredient is described in terms of how these regulators treat it:
- India: FSSAI (Food Safety and Standards regulations, including additive limits), and the Drugs and Cosmetics Act for cosmetics.
- European Union: Regulation (EC) 1333/2008 on food additives, informed by EFSA opinions, and the Cosmetics Regulation (EC) 1223/2009.
- United States: FDA food additive and GRAS rules (21 CFR).
- Canada: Health Canada's lists of permitted food additives.
How the information is produced. The per-ingredient assessment is written by an AI model (OpenAI GPT-4o, run on Microsoft Azure). The model is given the rules on this page and asked to report each ingredient's regulatory status as it understands it. It does not look anything up live in regulators' databases. For common additives, a curated ingredient guide takes over: each guide entry is checked against primary sources by a person, and when an ingredient on a label matches one, its country-by-country status and result come from the guide, not the AI. For everything else the AI's assessment stands — it can be wrong or out of date, and a result is a summary, not a legal opinion.
How current it is. The model's knowledge stops at its training cut-off, so a regulatory change after that date may not be reflected. We review the instructions given to the model when we learn of significant changes. A product's analysis is stored and reused for up to 12 months, or until it is corrected (see section 9).
2. What each result means
We don't label ingredients “safe” or “unsafe”. Whether something is a concern usually depends on the amount, the type of product, who is eating it, and the country. So each ingredient gets a regulatory status, and anything else worth knowing is shown separately, as context.
Permitted for this kind of product in India, the EU, the US and Canada, with no notable conditions and no notable caveat for particular groups or amounts.
Permitted, but with conditions: maximum-use limits, food-category restrictions or mandatory label warnings. Also used when a caution applies to particular groups (such as children or pregnancy), when a concern depends on the amount, when the scientific evidence is genuinely debated, or when the regulatory position is unclear.
Not permitted, or substantially restricted, for this kind of product in at least one of India, the EU, the US or Canada.
A health or evidence concern on its own never puts an ingredient in “Restricted or not permitted”. That result is kept for regulatory restrictions. Concerns like that are shown as “Worth noting”, with the reason given in context.
Where the AI is confident, an ingredient's status is also shown country by country, using these terms:
| Status | Meaning |
|---|---|
| Permitted | Allowed for this use with no notable conditions |
| Permitted with limits | Allowed, with maximum levels, category restrictions or label warnings |
| Restricted | Substantially restricted for this use |
| Not permitted | Banned or not authorised for this use |
| Status unclear | The AI isn’t confident of the position |
3. When countries disagree
An ingredient's overall status follows the strictest position among India, the EU, the US and Canada. For example, titanium dioxide (E171) is no longer authorised as a food additive in the EU, but is permitted with limits in India. It is shown as “Restricted or not permitted”, and the country-by-country breakdown makes clear it is still permitted in India. We show every country's position rather than picking one, because TrustPick is used in several markets.
Disagreements in the science, as opposed to the law, go into context. They never change the status.
4. What we don't consider
- Amounts. Labels rarely say how much of an additive a product contains, so we can't tell whether it is within a legal limit. A concern that depends only on quantity is shown as “Worth noting”, never as restricted.
- Your health. Results are the same for everyone. Allergy and diet flags in the app are matched against your own settings, separately from the regulatory status.
- Countries other than these four. Rules elsewhere may differ.
5. Uncertainty and unidentified ingredients
- Each analysis comes with a confidence value from 0 to 100%, reflecting how legible the label was and how well known the ingredients are. Below 60%, the app tells you to check the label yourself.
- When the AI isn't sure of an ingredient's regulatory position, it marks it “Status unclear” (shown as “Worth noting”) instead of guessing.
- Ingredients that can't be read or recognised are listed separately as unidentified. They are left out of the result and the score. In the app, you can correct the spelling and check again.
6. The overall result and star rating
The AI assigns each ingredient its status. The product's overall result and star rating are then worked out by fixed code, not by the AI, from the number of ingredients in each status:
| Overall result | When |
|---|---|
| Contains restricted ingredients | Any ingredient is “Restricted or not permitted” |
| Some things worth noting | Nothing restricted, but at least one ingredient is “Worth noting” |
| No concerns found | Nothing flagged |
| Stars | When |
|---|---|
| 5 ★ | No ingredient flagged |
| 4 ★ | Nothing restricted, and no more than 15% of ingredients worth noting |
| 3 ★ | Nothing restricted, but more than 15% of ingredients worth noting |
| 2 ★ | At least one restricted ingredient, but no more than 25% of ingredients |
| 1 ★ | More than 25% of ingredients restricted |
7. The Health Score
TrustPick Plus adds a Health Score from 0 to 100 with a grade from A to E. It combines three parts, each scored out of 100:
| Part | Weight | Weight with no nutrition panel |
|---|---|---|
| Ingredients (from the star rating) | 50% | 77% |
| Nutrition | 35% | not used |
| Processing level | 15% | 23% |
Ingredients part
| Stars | Points |
|---|---|
| 5 ★ | 100 |
| 4 ★ | 80 |
| 3 ★ | 60 |
| 2 ★ | 35 |
| 1 ★ | 10 |
Nutrition part
Starts at 100, then is adjusted using the values printed on the nutrition panel. For each nutrient, only the larger adjustment applies. The total is kept between 0 and 100.
| Nutrient | When | Points |
|---|---|---|
| Sugar | more than 25 g | -35 |
| Sugar | more than 15 g (up to 25 g) | -20 |
| Sodium | more than 600 mg | -35 |
| Sodium | more than 400 mg (up to 600 mg) | -20 |
| Saturated fat | more than 8 g | -20 |
| Saturated fat | more than 5 g (up to 8 g) | -10 |
| Fibre | 3 g or more | +10 |
| Protein | 5 g or more | +10 |
A known limitation: these thresholds are applied to whatever basis the label uses. That is usually per 100 g, but sometimes per serving, and we don't convert between the two. That makes the score less precise than formal systems such as Nutri-Score, and we don't claim to follow them.
Processing part
| Processing level | Points |
|---|---|
| Minimally processed | 100 |
| Processed | 60 |
| Ultra-processed | 25 |
| Unknown | 60 |
Grade
| Grade | Score |
|---|---|
| A | 80 or more |
| B | 65 to 79 |
| C | 50 to 64 |
| D | 35 to 49 |
| E | 0 to 34 |
Worked example
A biscuit with 4 stars (ingredients = 80), 18 g sugar, 450 mg sodium and 6 g protein (nutrition = 100 − 20 − 20 + 10 = 70), classed as processed (processing = 60):
80 × 0.5 + 70 × 0.35 + 60 × 0.15 = 73.5 → 74 (grade B)
There is no Health Score for a product with no ingredient list. Some products imported from public datasets show a separate, nutrition-only score labelled “Nutrition score”. It is not the Health Score. Health Score notes about your own diet settings (for example, “sugar-conscious”) only change the wording you see, never the score.
8. Processing level
The AI classifies each product, using criteria in the spirit of the NOVA system, as minimally processed (whole or lightly altered ingredients, no artificial additives), processed (preservatives, added sugars, refined oils or moderate additives) or ultra-processed (markers such as artificial colours, flavour enhancers, emulsifiers or several classes of additive). This is a judgement from the ingredient list, not an official NOVA classification.
9. Human review and corrections
- Products that users add to the catalogue go into a review queue for the TrustPick team. Product data imported from retailer listings and public datasets is not individually reviewed.
- Individual scans are not reviewed by a person. That is why we publish this method and show the AI's confidence.
- Spotted a mistake? Use “Report an issue” on any product page, or contact us. Once a product's record is corrected, the corrected analysis replaces the old one for everyone who looks it up afterwards. Saved scans in your own history keep the result from the day you scanned.
- Advertising and affiliate relationships never change a result or a score.
Not medical advice
TrustPick is not a medical device and does not diagnose, treat, cure, or prevent any medical condition. It is for consumer education and decision support, and is not medical, nutritional, legal or regulatory advice. For allergies, medical conditions or pregnancy, talk to a qualified professional.