How we figure out where a product is really from

A plain-English guide to how Product Origin Checker scores products.

Every score we publish answers four questions about an Amazon product: where it was made, where the unit you'll receive ships from, who's actually selling it (the retailer transacting with you), and where the money ultimately goes (the brand parent that captures the profit). This page explains, in plain terms, every place we look — and how honest we try to be about uncertainty.

Why four, not three? Two very different questions used to live under one label. For a Seiko watch sold on Amazon: the retailer is Amazon.com (US), but the money ultimately flows to Seiko Group in Japan. Both facts matter. So we split them.

The short version: we don't just read the Amazon listing. We pull signals from several independent places — the product page, the seller's own business records, the brand's website, a global domain registry, and our own growing database of verified corrections. No single source decides a score. The more independent sources that point to the same answer, the more confident we are.

Everything we check

1. The Amazon product page

Our browser extension reads the publicly visible information on the listing itself:

• The product title, brand, and description bullets.
• The "Country of Origin" field, when Amazon lists one in the product-details table. This is the strongest single signal we have.
• The "Ships from" and "Sold by" lines in the buy box.
• The seller's name and a link to their storefront.
• The manufacturer, the date the listing first appeared, and its Best Sellers Rank.
• Up to 15 customer reviews and 5 Q&A entries, scanned for "Made in…" claims, customs mentions, and shipping comments.
• The star-rating distribution — a lopsided wall of 5-star reviews is a known fingerprint of opaque drop-shippers.

2. The seller's business details page

Amazon keeps a separate "Detailed Seller Information" page for each third-party seller — and it's where the legally meaningful facts live. We fetch that page directly and read the seller's registered legal entity name, business address, and country of registration. A shell company is often only identifiable here, not on the product page.

3. The brand's storefront and website

We also fetch the brand's Amazon storefront page, reading any "About the brand" information and discovering links to the brand's own website. That website is the bridge to source #4.

4. A global domain registry — a third-party check

When we discover the brand's own website, we look that web address up in RDAP — the public, official registry that records who registered every website domain in the world. From the registry we read the country and organisation listed for the domain's registrant. This is an entirely independent, third-party confirmation: it doesn't come from Amazon or from the seller, so it's a valuable cross-check. A company registering its domain through a Chinese registrar with a Chinese registrant address is a meaningful clue that a US-sounding brand name alone would hide.

5. Our own database of verified corrections

Every time our review team confirms a correction (see "When a person overrides the AI", below), that becomes ground truth. If a seller or brand has a track record of confirmed origins in our database, the next new product from them starts with that knowledge built in. We also look across other products by the same seller we've already scored, so a consistent pattern carries over. Every complaint we resolve quietly makes future scoring more accurate.

6. Quick pattern checks

Before involving the AI, we run a handful of fast, cheap checks — calibrated especially for detecting Chinese-origin products, which is where seller obfuscation is most common:

• Whether the seller name contains a known Chinese manufacturing city (Shenzhen, Hangzhou, Yiwu, Guangzhou, …).
• Whether the seller name ends with "Co., Ltd." — a common Chinese business-registration suffix.
• Whether the brand name is the kind of random letter-string drop-shippers tend to use.
• Whether the listing was created within the last twelve months.
• Whether any text on the page literally says "Made in…" or "Manufactured in…".

Putting it together — the AI step

All of the evidence above — the product page, the seller's business records, the brand website and its domain-registry record, our USPTO trademark index, our database of priors, and the pattern checks — is assembled and sent to Anthropic's Claude AI. Claude weighs it alongside its own trained knowledge of brands and companies (it knows TCL is Chinese, that Hydro Flask is American, that Bosch is German), and returns a structured score for each of the four indicators: a top country, its ISO code, a percentage, a confidence level, and a short list of plausible alternative countries.

World knowledge first. Before Claude looks at any structured evidence, we require it to write down what it already knows about the brand from its training — the parent company's country, whether it's a well-known international brand, whether it fits the pattern of a Chinese Amazon marketplace seller using a US shell company for its trademark filing. This step exists because our structured pipeline can otherwise be fooled by paper trails: a Chinese seller can register "BRAND INC" in Delaware and file a US trademark, and our USPTO lookup would then report "United States" as the owner country — technically true for the shell, wildly misleading for the beneficial owner. When Claude's world knowledge places a brand confidently in a specific country, that knowledge takes precedence over the structural US signals from shell entities.

The strongest evidence wins. A literal "Country of Origin" field outranks a guess from the brand name; an independent registry record outranks a hunch. We also have a strict rule that a company's headquarters can never, by itself, be used to claim where a product is manufactured — a common and misleading conflation.

When the evidence is thin — live web search

If the assembled evidence still leaves us uncertain, Claude performs live web searches — against business registries, brand websites, LinkedIn, and news archives — to fill the gap. Every URL it relies on is recorded, and every score we show you lists the sources behind it. This step only runs when it's needed, so easy cases stay fast.

How confidence works

Each indicator has a percent (our certainty for the top country) and a separate confidence (our overall confidence in the estimate). These are not the same number — and the distinction matters.

• Percent close to 100, confidence close to 100 → we have strong, multiple corroborating sources. You can trust this.
• Percent 80, confidence 50 → we have a reasonable lean, but the evidence is sparse. Treat it as a working hypothesis, not a fact.
• Percent 50, confidence 20 → we genuinely cannot tell. The number is a placeholder; the meaningful signal is the low confidence.

We will not raise confidence above 85 unless we have either a literal Amazon-supplied origin field or multiple independent web sources confirming the same answer. Confidence honesty is the design principle.

Hong Kong, Macau, and Taiwan

For all four indicators, we score Hong Kong (HK), Macau (MO), and Taiwan (TW) as their own countries — never folded into mainland China. This is a hard rule in our system.

Manual verification (when humans override the AI)

Every score has a Contest this finding button. Reporters fill in a form telling us what they think is wrong, optionally proposing the correct value, and explaining their evidence and credentials. We verify the reporter's email and review the claim.

When our review team validates a correction, we install a manual override for that product. Future users see the verified value with a ✓ Verified mark — and the citations list includes the override as the top entry. Manual overrides can be retired or replaced as new information comes in; the full history is preserved for audit.

What we don't currently use

For transparency, here's what's NOT yet in our pipeline:

• US Customs import records. Panjiva, ImportGenius, and similar services have shipment-level data for who imports what from where. This is the gold-standard evidence layer; the plumbing is in place and it will activate once the subscription cost is justified by paying users.
• OCR on customer-uploaded product photos. "Made in X" is often printed on packaging visible in review photos; computer vision could read it.
• WIPO international trademark database. Adds non-US filings — useful for European brands specifically. The US trademark registry (USPTO) is already in our pipeline.
• OpenCorporates. Cross-jurisdiction company registry data.

Bottom line

Our scores are well-supported probabilistic estimates, not factual claims. When evidence is strong, we're confident. When evidence is thin, we say so loudly. When someone with first-hand knowledge tells us we're wrong, we listen — and the dataset gets better.

If you want to dig deeper, the full technical specification of every data source we consult is published at DATA_SOURCES.md in our repository.