One honest number,
from 0 to 10.
Behind every score is the same skeptical question: is this actually safe to use, for the people who actually buy it? Here’s how we answer it — and why no brand can buy a better number.
What the number means
A single rating for the whole product. Higher is safer — and every score on the product page links to the exact ingredients and sources behind it.
Clean formulation, well-understood ingredients, no concerns identified in our review.
A few caution-level ingredients, a notable nutrition concern, or limited transparency. Some groups may want to take extra care.
Multiple warning-level ingredients or a serious concern — an IARC/EU-classified substance, a banned ingredient, or a nutrition profile with several significant deductions.
The four things we score
Every product is scored on four independent lanes. These are the same four bars you’ll see broken out on any product page — the page you’re reading and the score you’re looking at are the same system, described in the same words.
Nutrition
Carries the most weight on food and drink.
Sugar, salt, saturated fat and energy density — judged against a realistic serving of that kind of product, not a flattering per-100g figure a pack can choose for itself. Food and drink only; personal care has no nutrition lane.
Ingredients
Carries the most weight on personal care.
Every ingredient is matched against our curated safety database — each entry carrying a documented concern and a source. Known hazards, restricted substances and irritants pull the lane down. What the label hides counts too: “fragrance”, “colour” or a bare class word with no compound named is treated as an unknown, never as harmless.
Processing
How industrial the formulation is.
Built on the NOVA framework for food: a short list of whole or simple ingredients scores high; a long list of industrial functional additives — modified starches, gums, emulsifiers, artificial colours, synthetic surfactants — scores low. Plant extracts and oils don’t count against it, so a genuinely botanical formula isn’t punished just for a long label.
Marketing
Does the front of pack survive the back?
Every claim on the front — “high protein”, “no added sugar”, “100% natural” — is read back against the actual ingredient list and nutrition panel. Claims that hold up cost nothing. Claims the label itself contradicts pull the score down, because a misleading pack is a consumer risk in its own right.
Sustainability is deliberately not a fifth lane. Sourcing, packaging and carbon impact almost never appear on a label, and we won’t publish a number we cannot evidence. Where such an ingredient is also a health concern, it still counts under Ingredients.
How the four become one
The headline score isn’t a vibe or an average of opinions — it’s a fixed calculation, run the same way every time. Every product page shows you this same arithmetic for that product, step by step.
Each lane is scored on its own
Nutrition, Ingredients, Processing and Marketing are computed independently from the label — no lane can quietly borrow credit from another.
They’re combined by fixed weights
The weights are constant for every product in a category and never vary by brand. Food and personal care use different mixes, because a shampoo has no nutrition panel and a biscuit’s marketing matters less than what’s in it.
One catastrophic lane can’t be averaged away
If a single lane is bad enough, a limiter holds the whole score near it. Excellent ingredients cannot rescue a product that is nutritionally indefensible, and vice versa.
Hard ceilings are applied last
Specific serious findings impose a maximum the product cannot rise above, whatever the arithmetic produced. Ceilings only ever pull down — never up.
Some things can’t be averaged away
Certain findings are serious enough that nothing else in the formula can rescue the score. Each one imposes a ceiling — a maximum the product cannot rise above, whatever the rest of the arithmetic produced. Some ceilings sit in the Poor range, others in Moderate; the ceiling matches how serious the finding is. They only ever pull a score down.
On food & drink
On personal care
Every ceiling above is a rule that exists in code and can fire today — and when one applies, the product page names it in plain English rather than leaving you to guess why the number is low.
Where the machine ends and the writing begins
A score never rides on an AI’s say-so. We’re specific about which part does what.
A curated safety database — fixed rules
Ingredients are resolved against our own substance registry, each entry carrying a documented concern and a citable source. This layer is lookup and rules, not judgement.
A scoring engine — produces the number
The four lanes, the weighting, the limiter and every ceiling are computed in code. The same label always produces the same score. The rules have the last word: where a lane starts from an AI assessment, the engine’s ceilings decide how high it can go.
AI — reads the label and explains the result
AI turns pack photos into structured ingredient and nutrition data, and writes the plain-English explanation. It is given the engine’s verdict and must explain it — it cannot dismiss a confirmed hazard or lift a score past the engine’s limits.
What we check against
Judgements are anchored to recognised, public safety standards — not our own opinions. Where a product page asserts a hazard, it carries the source with it:
The rules we hold ourselves to
Danger is never averaged away
Twenty good ingredients don’t cancel one carcinogen. A serious hazard dominates the score — it isn’t diluted by everything else in the pack.
Unknown ≠ safe
If an ingredient’s risk isn’t established, or the label never names the compound, we treat it as unknown — not harmless. The benefit of the doubt goes to your health, not the product.
The most protective rule wins
If India’s FSSAI permits something the EU or WHO restricts, we judge by the stricter standard — not the most lenient one.
A serving, not a per-100g figure
Nutrition is judged against how much of that product a person actually consumes at once, so a pack can’t look mild by quoting a portion nobody eats.
One bad lane is never hidden
A product with one badly failing axis is never labelled “Great”, even when the weighted average would allow it. A strong average cannot paper over a specific failure.
Who actually buys it matters
A children’s snack is held to child-safety standards; a baby product is judged more strictly again; a product used near the eyes or lips is held to a higher bar. Context changes the verdict.
“Natural” earns nothing on its own
Marketing words are not evidence. Only specific, verified ingredients move the score — and a claim the label contradicts costs the product.
The brand name is invisible
A famous logo and an unknown one with the same formula get the same score. Reputation is not a safety input.
How we keep ourselves honest
A rating system is only as trustworthy as its own quality control. Ours is built in, not bolted on:
Every score is reproducible
We keep a snapshot of every product’s score and the reasoning behind it. Before any methodology change ships, we re-run the entire catalogue and diff it — every single movement has to be explainable, or the change doesn’t go out.
An automated auditor can block publication
Before a product goes live, automated checks look for the ways a score can be wrong — a nutrition panel that contradicts its own ingredient list, a claim the engine and the written explanation disagree about. A serious finding stops that product from publishing until a human resolves it.
The method is versioned and dated
Thresholds and rules live in a versioned store rather than scattered through code, so “what did this product score under which version of the method” is always answerable. The changelog below is the public record.
We remove rules we can’t stand behind
We audit our own rules for ones that look impressive but can never actually fire, and delete them. A capability the methodology appears to claim but does not have is worse than no capability at all — so if you see a rule listed on this page, it is one that genuinely runs.
Why we don’t publish the exact formula
We publish the shape of the method — every lane, every step of the calculation, and every family of hard ceiling. What we hold back are the precise weights and thresholds. That’s deliberate. If the exact maths were public, a brand could reverse-engineer the smallest tweak that nudges a number up without making the product any safer. Keeping the constants in-house keeps the score honest, consistent, and hard to game — which is the whole point of an independent rating.
Questions about the score
What does the 0–10 score actually mean?
It’s a single rating for the whole product. Roughly: 7–10 is great, 4–6.9 is moderate, and below 4 is poor. Every score links to the specific ingredients, nutrition figures and sources behind it.
What are the four parts of the score?
Nutrition, Ingredients, Processing and Marketing. They’re shown as their own bars on every product page, alongside the arithmetic that turns them into the headline number — so you can always see which lane cost the product its score. Personal-care products have no nutrition panel, so they’re scored on the other three.
What does the “Processing” part mean?
It reflects how heavily processed and industrial the formulation is, following the NOVA framework for food. A short list built on simple or whole ingredients scores high; a long list of industrial functional additives — modified starches, gums, emulsifiers, artificial colours, synthetic surfactants — scores lower. Plant extracts and oils don’t count against it, so a genuinely botanical formula isn’t penalised just for a long label.
What is a hard cap?
A ceiling a specific serious finding imposes on the score — a maximum the product cannot rise above no matter how well it does elsewhere. Some ceilings sit in the Poor range, others in Moderate: the ceiling matches how serious the finding is. Caps only ever pull a score down.
Can a brand pay to raise or remove its score?
No — never. There are no ads, sponsored placements or paid rankings, and the rule is enforced in code: the brand name is not an input to the engine at all. A brand can submit a factual correction with evidence, which we verify independently.
Why don’t you publish the exact formula?
We publish the shape of the method, every lane, and every family of hard cap — but not the precise weights and thresholds. That keeps the score honest and hard to game: a brand shouldn’t be able to reverse-engineer the smallest tweak that nudges a number up without making the product any safer.
Is the score just an AI guess?
No. The number is computed by our scoring engine — the same label always yields the same score, and we verify that by re-running our whole catalogue and requiring zero unexplained movement. AI reads the label and writes the plain-English explanation. For foods and drinks with a nutrition panel, the Ingredients score is now fully rule-based: it starts from a clean label and deducts only for what our register and rules recognise. Personal-care products, foods with no nutrition panel, and labels we could not read properly still start from an AI assessment, but the engine’s rules set the limits that assessment cannot exceed, and AI cannot overrule a confirmed hazard. We are moving those to rule-based scoring too.
Does the score change for my health needs?
The product’s public score is always the general-population one and never moves. If you tell us what matters to you — sugar for diabetes, sodium for blood pressure — we show a second, personal verdict alongside it. Your profile never changes what anyone else sees.
Why isn’t there a Sustainability score?
Because we can’t verify it from a label. Sourcing, packaging and carbon impact almost never appear on the pack, so rather than guess at a number we don’t score it at all. Where an ingredient also raises a health or safety concern, it still counts under Ingredients. We would rather leave a dimension out than publish one we cannot evidence.
Is this medical advice?
No. Decode Label is an information tool to help you read labels, not a substitute for advice from a doctor or qualified professional.
What's changed
Our method is actively maintained. When we recalibrate it, scores can shift — here's the running log, so a changed number is visible, dated work, not a mystery. Current: methodology v3.1 (updated 8 Oct 2026).
-
8 Oct 2026
For foods and drinks with a nutrition panel, the Ingredients score is now fully rule-based. It starts from a clean label and deducts only for ingredients our safety register and rules recognise, instead of starting from an AI assessment of the whole product. The AI assessment had been charging sugar, salt and oil a second time inside the Ingredients score, although the Nutrition score already charges them, and charging undisclosed additive classes a second time, although the undisclosed-additive limit already caps them. Removing that double count raises about 30% of food scores, by 0.8 on average. When a label could not be read properly (for example when the nutrition table was captured as the ingredient list) the old assessment is kept. Also: "Approved flavour", "Flavourings" and a natural "Colouring agent" with no substance named are now treated as undisclosed, like "Permitted flavour" already was, and three spellings of the raising agent INS 450(i) are now recognised.
-
3 Oct 2026
Three synthetic colours now lower a cosmetic's Ingredients score: CI 47000 (a recognised contact sensitiser), CI 26100 (an azo dye) and CI 61565. Before this, the product page warned about them but the score did not count them. A product named after a plant ingredient, such as an "Amla" or "Almond" hair oil, whose largest ingredient is mineral oil or another petroleum base now loses Marketing points, and the page says why. Mineral oil itself is not treated as a hazard: cosmetic-grade mineral oil has no carcinogen or sensitiser classification. An unnamed "vegetable oils" blend is no longer listed as a benefit, because the label does not say which oils it contains. Mineral oil is also no longer mistaken for a plant ingredient when we judge how processed a formula is.
-
13 Jul 2026
Products sold for children are now scored as children's products, whatever shelf they sit on. A kids' or baby toothpaste is filed under oral care and a kids' shampoo under haircare — neither was recognised as a children's product, so both were held to the adult bar. They now take the stricter children's thresholds for fragrance allergens and undisclosed fragrance. And because a young child cannot reliably spit, an undisclosed "Flavour" in a children's toothpaste is now treated as something the child SWALLOWS — an unnamed blend can carry sweeteners or essential-oil allergens that are never declared — rather than as an ordinary cosmetic-fragrance footnote.
-
13 Jul 2026
A bar built on a mild synthetic cleanser (a "syndet", e.g. Sodium Cocoyl Isethionate) is now credited for it on the page, instead of being listed as a neutral filler while only its harsher secondary surfactant was flagged. A syndet base sits far closer to skin pH than a traditional saponified soap (pH ~9–10) and is the milder formulation — the score already recognised this; now the ingredient list does too.
-
12 Jul 2026
Synthetic colours are now judged the same way on every product. A colour like CI 11680 or CI 74260 was being listed as a "moderate concern" on one soap and as a neutral ingredient on another — the same dye, read two different ways. Synthetic colours are now pinned to one consistent moderate, sensitive-skin caution wherever they appear. Mineral pigments are deliberately not treated as synthetic dyes: titanium dioxide remains a safe whitener on skin, and iron oxides are inert. Separately, carbon black is now recognised when a label prints it only as a colour-index number ("CI No. 77266") rather than by name — it carries an IARC "possibly carcinogenic" classification, so this can lower the score of a product that declares it that way.
-
12 Jul 2026
Bathing bars are now correctly recognised as wash-off products. The contact-time relief announced on 2 July was described as covering bar soap, but an internal naming mismatch meant it never actually reached bathing bars — so a bar was judged against the stricter standards meant for a product that stays on the skin all day, and could even be described in its own review as "leave-on". Both are fixed, which may raise the score of some soaps. Alongside this: essential oils (sandalwood, turmeric, lavender and the like) are now grouped into the same single fragrance-allergen caution as the allergens they carry, instead of being listed as neutral while the review warned about them; several distinct colours declared under one generic "Colour" heading are no longer shown as the same colour repeated; and the claims a product makes in its NAME ("2% Kojic Acid", "with Niacinamide") are now decoded and checked against the ingredient list, as front-of-pack claims already were.
-
7 Jul 2026
Three snack-batch refinements. BHA (E320, butylated hydroxyanisole) now triggers the carcinogen-additive cap in the same way ammonia caramel does — both carry an IARC Group 2B classification — including when it is disclosed only inside a generic "Preservative" listing. A major allergen such as peanuts is no longer counted as a general-population ingredient hazard that lowers the Ingredients score for everyone; it is shown instead as a clear allergen disclosure, since it is a risk specific to allergic individuals (already surfaced in "Use with caution if"). And a blend of refined vegetable oils is now flagged once as a single "refined oil blend" concern rather than listing each oil as a separate concern, unless an individual oil carries its own distinct issue (e.g. an added BHA/TBHQ antioxidant).
-
3 Jul 2026
Named developmental feeding-stage labels (e.g. "1st/2nd/3rd Foods", "Supported Sitter") are now recognised as age/stage designations — shown as such rather than as an unverifiable marketing claim, the same way "Stage 2" and an age range already were.
-
2 Jul 2026
The "trace amount" label on the ingredient read no longer misfires on short lists or on base staples — a major ingredient like salt or oil sitting a few places down a short list is no longer mislabelled as a token trace. And front-of-pack text that just describes the packaging ("new pack", "easy seal") or the product category ("Indian savoury snack") is now recognised as a neutral descriptor rather than counted against the product as an unverifiable marketing claim.
-
2 Jul 2026
Optical brighteners (fluorescent whitening agents — synthetic dyes that only make a bar look whiter) are now recognised in their own right: correctly classified rather than mislabelled as preservatives, and marked down as an unnecessary additive, with an extra caution for baby products where they have no place.
-
2 Jul 2026
Traditional "true soap" bars (made from saponified oils) are now recognised as alkaline (pH ~9–10) and marked down for infant products, where a baby's naturally acidic skin barrier is stripped by high-pH soap — the practical infant concern that trace additives had been overshadowing. At the same time, wash-off products (bar soap, body wash, shampoo) now get some relief on the fragrance-disclosure ceilings, since a 30–60 second rinse exposes skin far less than an all-day leave-on; genuine safety limits (harsh surfactants, preservatives, the soap-pH concern) are unaffected.
-
2 Jul 2026
Undisclosed fragrance in a baby product now carries a stricter ceiling than in an adult product (delicate skin reacts to fewer allergens). The Ingredients score now spreads further when several high-concern ingredients stack up, so it stays informative even when a cap sets the headline. Market-rank / professional-endorsement claims ("#1 Pediatrician Prescribed") are now shown as their own claim type and weighted more heavily than ordinary unverifiable puffery. EDTA is described by the state of the evidence rather than as a settled endocrine disruptor.
-
25 Jun 2026
Added a cap for topical carcinogen-class ingredients (e.g. talc, classified IARC Group 2B) in leave-on and rinse-off personal-care products, so a product whose own analysis flags a carcinogen of concern can no longer score in the "Good" band. Titanium dioxide remains uncapped on skin, where it is a safe UV filter.
-
16 Jun 2026
Infant formula (baby milk substitutes) is now scored as a regulated breast-milk substitute: its mandated fat and lactose profile is no longer penalised the way adult foods are, while added or free sugar (e.g. sucrose) is still flagged. Disclosed probiotics, prebiotics and DHA are credited, and ingredients such as coconut oil that are deliberate breast-milk analogs are no longer shown as a concern in this context.
-
15 Jun 2026
Shaving & grooming products now use the same Ingredients/Processing/Marketing breakdown as other personal-care items, instead of an unverified overall score. Added a cap for ingredients with endocrine-disruption concerns (e.g. BHT), and a transparency flag for ingredients listed only by a generic class name ("Preservative", "Colour") without naming the specific compound.
-
15 Jun 2026
Clarified that the score is built from Ingredients, Processing and Marketing. Environmental sustainability is not scored as a separate dimension — it can rarely be verified from a label — so it is no longer shown as its own bar (see "How we score").
-
14 Jun 2026
Added a carcinogen-additive cap (e.g. ammonia caramel, titanium dioxide in oral care), name-based ultra-processing detection, and a sugar-density floor for drink powders.
-
12 Jun 2026
Recalibrated Marketing scoring to read front-of-pack claims against the label; opaque, mostly-unverifiable packs no longer score as "honest".
See it on a real product
Search any product and read its score, the four lanes behind it, and the sources we cite.
Search a product