MerchGuard

MerchGuard / Methodology

How we detect violations.

Eight policy categories, Claude Haiku-first with Sonnet escalation on uncertainty, USPTO + EUIPO trademark cross-reference, and every flag links back to its source policy.

The 8 categories

TM

Trademark

Word marks and logos registered with USPTO or EUIPO that appear in title, tags, or imagery without authorization.

Sources: USPTO TESS / TSDR · EUIPO eSearch

IP/c

Copyright

Unauthorized use of third-party characters, photographs, illustrations, song lyrics, or brand imagery.

Sources: Etsy IP policy · DMCA case law

PROHIB

Prohibited items

Items Etsy explicitly disallows: weapons, regulated goods, hate / violent imagery, recalled or unsafe products.

Sources: Etsy Prohibited Items policy

HANDMADE

Handmade & production partner

Listings claiming handmade or original design that appear resold, dropshipped, or templated. Production-partner disclosure rules.

Sources: Etsy Handmade Policy · Production Partner FAQ

POD

POD artwork rights

Artwork that may be AI-generated without disclosure, or derived from unlicensed sources.

Sources: Etsy AI Disclosure rules · POD provider TOS

FEE

Fee avoidance

Encouraging buyers to transact off-platform, circumventing marketplace fees.

Sources: Etsy Seller Policy §Anti-Disintermediation

LISTING

Listing quality

Keyword stuffing, misleading titles, brand bidding in tags, SEO anti-patterns that trigger marketplace downranking.

Sources: Etsy SEO guidelines · Listing Quality Score signals

AUTH

Authenticity & claims

Unrealistic production times, unsupported free-shipping or material claims (e.g. CPSIA-relevant safety statements), undisclosed duties.

Sources: CPSIA §101–104 · Etsy Materials Disclosure

Scan pipeline

  1. 1

    Listing input

    Paste a listing URL (auto-fill via JSON-LD + OpenGraph extraction) or fill the form manually. Title, description, tags, category, materials, production-partner disclosure, and price feed the scan.

  2. 2

    Rule loading

    Active marketplace rules are loaded from the policy database (cached). Each rule has a structured prompt snippet, source URL, and category mapping.

  3. 3

    Claude Haiku first pass

    A single Claude Haiku call evaluates the listing against all 8 categories at once with cached system prompt. Haiku returns structured JSON with overall risk score, per-violation evidence, and confidence.

  4. 4

    Sonnet escalation (uncertainty-gated)

    Any violation with confidence between 0.3–0.7 is re-evaluated by Claude Sonnet for nuance. Sonnet either confirms with higher confidence, demotes severity, or removes the violation.

  5. 5

    Source citation

    Every violation links back to the original policy clause (Etsy URL, USPTO/EUIPO record, CPSIA section, etc.). No ungrounded claims.

  6. 6

    Result

    Risk score 0–100, confidence 0–100%, ranked violations with evidence excerpt, suggested rewording, and policy URL. Stored against your scan history with model + token telemetry (admin-side).

What we don't do

  • ·We don't guarantee Etsy won't suspend your listing. We surface the same signals their integrity team uses, but final policy decisions are theirs.
  • ·We don't hallucinate trademark hits. Every TM flag is matched against live USPTO / EUIPO records.
  • ·We don't store your listing content longer than the scan window unless you connect a shop for monitoring.

Try it

Run a free scan →