Methodology

How BrandCheck verifies authenticity, what the confidence score means, and where AI falls short.

The pipeline

  1. OCR pass. Your photos go through Claude Vision in OCR mode to extract any visible serial numbers, date codes, or identification stamps.
  2. Brand detection. If you tell us the brand in the description, we use it. Otherwise the AI identifies the brand from logos, monograms, and labels.
  3. Serial validation. The extracted serial is checked against brand-specific format rules (e.g., Louis Vuitton uses 2 letters + 4 digits; YSL uses 3 letters + 6 digits). Mismatches are flagged.
  4. Visual analysis. Photos are evaluated against six authentication points — serial, stitching, hardware, logo, material, labels — with brand-specific hints fed into the prompt (e.g., "Goyard chevron is hand-painted, slight irregularity is normal — perfect symmetry is suspicious").
  5. Verdict. The model returns one of three outcomes — likely original, likely fake, uncertain — with a 0–100 confidence score and a per-point breakdown.

What the confidence score means

Where we fall short

BrandCheck is intentionally conservative about its limits:

What it's not

BrandCheck does not claim legal authority. We don't certify items. Our reports are intended as a pre-purchase or pre-sale informational signal — not a substitute for forensic, brand-issued, or accredited third-party authentication services for high-value transactions.

Methodology — How BrandCheck verifies brand authenticity