Recall-notice supplier names: 167 of 310 matched no registry record.

The numbers.

1. More than half the names returned nothing.

Of 310 distinct names, 167 (53.9%) produced no candidate record. The API returns a specific code for this, error_code 300000, meaning no results found. That is distinct from a permission or quota failure. A nil return says the source held no entry matching that string on the query date. It does not say the company does not exist.

2. Fewer than a third resolved to a single company.

91 names (29.4%) returned exactly one candidate. 52 (16.8%) returned two or more, which on a name alone cannot be narrowed further. If your supplier's name falls in that second group, picking the first search result is a coin toss between real registered companies.

3. Records that came back usually carried the code.

Of the 143 names that returned candidates, 134 top records (93.7%) carried an 18-character unified social credit code. That code has a check character under GB 32100-2015 and maps to one registered legal person. It is the only identifier in this chain that does.

4. Most matches came from an English-name field.

The source reported the match came from a registered English-name field in 125 of 143 cases (87.4%). Chinese registries do hold such a field alongside the licence. It is not part of the licence itself. It is also not always the name a foreign notice used.

5. The share of recalls naming China keeps rising.

Within the same frame, notices listing China or Hong Kong among manufacturer countries were 185 of 305 notices in 2024 (60.7%), 301 of 420 in 2025 (71.7%) and 346 of 434 in 2026 to 10 September (79.7%). Separately, 1,117 of 1,159 notices (96.4%) listed exactly one origin country.

Those three percentages describe this frame of notices. They are not a measure of how much of any category is made in China. A rising share of recalls is not by itself a statement about product safety.

Scroll the table sideways to compare each count against its denominator.

How this was measured.

Recall records came from the public SaferProducts REST service run by the US Consumer Product Safety Commission. They were pulled by calendar year for 2024, 2025 and 2026. That saved frame holds 1,159 notices covering 4 January 2024 to 10 September 2026.

From every notice we read the manufacturer, importer, distributor and retailer party fields. We kept each party name carrying an explicit China or Hong Kong marker. Deduplicating those gave 310 distinct names. Each name was then submitted, exactly as printed, to a licensed commercial registry search API on 16 September 2026. For each one we recorded four things. How many candidates came back. Whether the top record carried an 18-character credit code. Which field the source said the match came from. The registration status.

Person-name fields were not published. The CSV linked below holds the aggregate counts only, not company-level rows.

What these numbers do not say.

• A nil return is not evidence of a fake company. It is the expected outcome when a Latin-script name never matched a registered field in the first place. Spelling differences alone are enough: one company in this frame was reachable only after switching Electron to Electronic and Manufacture to Manufacturing.
• These are commercial database results, not an official portal check. The national portal is GSXT. Querying it is a separate step carried out manually inside China, and it was not done for this measurement.
• The figures are a snapshot. Registrations change, and a lookup repeated later will not necessarily reproduce these counts.
• Resolution is not verification. Matching a name to one registered company tells you the company exists on the register. It says nothing about whether it manufactures anything, whether it can perform, or whether it is safe to pay.
• The frame is US recalls only. It describes companies that appeared on a CPSC notice, which is not a random sample of Chinese exporters.

What a buyer can do with this.

The practical consequence is narrow. Put it plainly: ask your supplier for the 18-character code, in writing, before you pay. Everything else is weaker. The name on the invoice, the name on the website, the name a government notice printed: on these numbers, more often than not, none of them leads anywhere.

The code itself can be checked for internal consistency without any database at all: the final character is computed from the other seventeen. A free check-digit tool runs that calculation in your browser. A correctly formed code can still be invented, so this establishes form rather than existence. A malformed one is a fast, free answer.

If you want the registered record behind a code rather than the arithmetic, that is what our supplier reports cover.

Questions about this measurement.

Why would an official notice name return nothing?

Because the notice is not quoting a registered field. A Chinese business licence carries a Chinese name. Whatever appears in Latin script on a foreign document was transliterated, translated or supplied by a party downstream, and none of those routes guarantees a match. One name in this frame was printed as unspaced pinyin running the entire Chinese company name together as a single word.

Does ambiguity mean someone is hiding something?

No. Chinese company names reuse a small pool of auspicious characters, and several genuinely unrelated companies can share a plausible English rendering. Ambiguity is a property of the name, which is why the credit code exists.

Can I reproduce this from the CSV?

You can reproduce the aggregates. You cannot reproduce individual lookups, because the published CSV holds counts rather than company-level rows. The recall notices themselves are public at saferproducts.gov.

Why publish a negative result?

Because the alternative is buyers assuming that a name on an official notice is a working identifier. On this frame it is not, most of the time, and that is a more useful thing to know than any single company's file.

Citing this measurement.

Zhou, B. L. (2026). Resolving Chinese company names from US CPSC recall notices to registered legal entities (2024–2026 frame, n = 310). Currawong. https://currawongweb.com/verify/cpsc-recall-name-resolution-study/ · https://doi.org/10.6084/m9.figshare.33845302

Aggregate counts: cpsc-recall-name-resolution-2026-09-16.csv. Published under CC BY 4.0. Source recall records: the public SaferProducts REST service at saferproducts.gov.