Skip to main content
VerifySkinCHECK • KNOW • USE SAFER

Methodology: how results are decided

VerifySkin assesses risk from the evidence available. Where there is no evidence we do not guess — we say the information is insufficient.

Fixed, versioned rules

The same input always gives the same result, using written rules. Every result stores the rule version so you can later see why it came out that way. Current rule version: 2026-10-24.1

The five results and when each is given

  • Verified supply chain

    Only when the product is identified and a high-reliability source such as the manufacturer or an authorised distributor has confirmed that specific batch — or our reviewer has decided citing such evidence.

    What it does not mean: This does not mean no single unit could ever be fake.

  • Potential risk detected

    When at least one important risk sign is found: a match with a verified official notice (banned, recalled or warning), a recorded recalled or counterfeit batch, a mismatch with a reliable LOT format rule, an expired or impossible printed date, or specific visible discrepancies in the photo.

    What it does not mean: This does not mean the product is definitely fake.

  • No obvious red flags

    When at least one meaningful check (such as a readable photo analysis or a reliable LOT format rule) was completed and no warning sign was found.

    What it does not mean: This does not mean the product is genuine.

  • Insufficient information

    When the product could not be identified, the batch code is not in our records, the photo could not be read — or only the barcode matched (a barcode identifies a product, it does not prove it is genuine).

    What it does not mean: This means neither genuine nor fake.

  • Under review

    When a person needs to look — for example a match with a list still under review, or you asked for a review.

    What it does not mean: This is not a final result.

Identification is a separate question

“Which product is this?” (identification), “is it genuine?” (authenticity) and “are there known concerns?” (safety and official notices) are shown separately. A barcode can identify a product in our catalogue, but it does not prove the product in your hand is genuine — fake products can carry real barcodes.

LOT numbers and dates

A LOT number is checked against the brand’s documented LOT format and our batch records. Manufacture and expiry dates are shown only when a cited source states them, or as read from your photo — never worked out from the LOT code.

Types of evidence and how reliable they are

Every piece of evidence is stored with its source, source type, check date and reliability level (high / medium / low).

  • High: written confirmation from the manufacturer or an authorised distributor; notices from a regulator
  • Medium: official references, supplier documents
  • Low: public barcode databases, customer photos, general public information

A customer photo or a public barcode database on its own is never treated as proof that a product is genuine.

Photo analysis (AI) is only an input

When you add a photo, an AI model looks for visible discrepancies in spelling, fonts, printing, logo, barcode and dates on the pack, and reads the barcode number, LOT code and printed dates. The model does not decide the result: it only says whether the photo could be read and which specific discrepancies it saw. A model score is never turned into “genuine”. If the photo cannot be read the result is “Insufficient information” and your check is returned.

Web search results are public information of low reliability, found by AI and not yet checked by our team.

Human review

If a result seems wrong you can ask for a review. The reviewer records a reason and the evidence used. The earlier result is never deleted — a new version is added and the result shows “updated”. Nobody can set a result to “Verified supply chain” without high-reliability evidence.

What we never claim

  • That a product is “100% genuine” or safe to use
  • Any diagnosis or medical advice about skin conditions
  • That a seller is a fraud — we never publicly accuse a seller based on customer reports or automatic assessments
  • A percentage “accuracy” for detecting fakes — until a verifiable evaluation set exists