How it works

1. It finds the product

You type a product name. It searches the US National Institutes of Health's Dietary Supplement Label Database and picks the closest label, using the words you typed, the brand, the strength, whether the label is still on the market, and how recent it is. The page always shows which label it picked, with a link to it, so you can check it's the right one. Only products sold in the US are covered.

2. Two AI models describe each ingredient

For each ingredient you tick, two AI models (from Anthropic and Perplexity) each fill in the same fact sheet: how strong the evidence is for each goal, the dose range studies used, common side effects, which kinds of medication it can interact with, and whether to avoid it in pregnancy.

• When the two sheets differ a little, the more cautious answer is taken.
• When they differ a lot, a stronger model decides. If it can't, that fact is marked "uncertain".
• The facts are the same for everyone. The models never see your profile.

Facts are remembered, so an ingredient that was checked before is instant.

3. Fixed rules fit it to your profile

No AI is involved in this step. Plain rules work out, for each ingredient:

• Your daily amount: the amount per serving times your servings a day.
• Whether that is within, below or above the range studies used.
• Whether it passes the safe upper limit for your age, sex and pregnancy. These come from the NIH Office of Dietary Supplements fact sheets, and exist only for vitamins and minerals.
• Whether it can interact with a kind of medication you take.
• Whether to avoid it if you are pregnant or breastfeeding.
• How strong the evidence is for the goals you picked.

4. The score

• Only ingredients with "some" or "strong" evidence for at least one of your goals count. The rest are listed as not linked to your goals.
• Each counted ingredient gets 0 to 100: evidence counts 50%, how the dose compares 30%, and how certain the facts are 20%.
• The product's score is the average of those.
• If any ingredient is flagged (over its upper limit, can interact with a medication you take, or to avoid in pregnancy), the score can't go above 20.
• Every score shows its reasons, ingredient by ingredient.

What it can't do

• AI can be wrong. Two models can share the same mistake, and an agreement isn't proof.
• Labels can be incomplete. An amount that isn't listed, or in a unit that can't be converted (such as IU of vitamin A), is left out, so a total can be too low.
• Upper limits are partial. There are none for most ingredients. Some cover only one form, and a vitamin A flag can be a false alarm, because labels don't always separate the form the limit applies to.
• It doesn't know you. It knows only what's in your profile. It can't replace a doctor or pharmacist, who know your health, your medicines and your history.