LEI, CIK or Wikidata ID: which company identifier?

Ask an AI agent about a company and it will usually reach for a name. Names are where identity goes wrong: they change, they repeat, and they are spelt differently in every system. Identifiers fix that, but only if the agent knows which identifier answers which question. Here are the three an agent meets most often when it researches companies, and the rule for linking them.

The three identifiers at a glance

Identifier	Issued by	Format	What it identifies	Access
LEI	The global LEI system, overseen by GLEIF	20 characters, ISO 17442	A legal entity, anywhere in the world	Free; published under CC0
SEC CIK	The US Securities and Exchange Commission	10 digits with leading zeros	A person or company that files with the SEC	Free; no key; fair-access rate limits
Wikidata ID (QID)	Wikidata, a community-edited knowledge base	“Q” followed by digits	An item, which may describe a company	Free; structured data under CC0

They overlap, but they are not interchangeable. A company can have an LEI and no CIK (most companies outside the US securities markets), a CIK and no LEI, or both. A Wikidata item can exist for any of them, and its claims are only as good as the edits behind them.

When to use which

• “Which legal entity is this?” Start with the LEI. Its record gives the legal name, the jurisdiction the entity was formed in, its legal form and, for nearly every entity, its parents.
• “What has this company filed?” Use the CIK. The SEC’s data APIs need no authentication or key, update throughout the day as filings are disseminated, and address companies by the 10-digit CIK, for example https://data.sec.gov/submissions/CIK0000320193.json. The SEC asks automated clients to declare a user agent and to stay under 10 requests a second.
• “What else is known, and does anyone disagree?” Use the Wikidata item. It is a fast way to find candidate identifiers: it stores LEIs under property P1278 and CIKs under property P5531. Treat what it says as corroboration, not proof.

Can you link them?

Only when a primary record names both. This is the rule that separates a careful agent from a confident one.

In IKAG’s tests on captured public records, Apple’s LEI record and its SEC record did not name each other. Wikidata linked them, but Wikidata is community data, so the capability kept the two identifiers apart and said why. The SPDR S&P 500 ETF Trust was different: its LEI record names its CIK, so the two could be joined on the strength of a primary record.

Situation	Join them?	Why
A primary record names both identifiers	Yes	The issuer of the record made the link
Only a community source links them	No, not yet	A lead to confirm, not a fact
The names match	No	Different entities often share names

Pitfalls that produce wrong answers

1. Unpadded CIKs. “320193” and “0000320193” are the same CIK, but the SEC’s APIs expect the ten-digit form.
2. Names as identifiers. A name search returns candidates. In one of our tests, a fund’s name matched three entities in the LEI register, one of them inactive.
3. Lapsed is not closed. A lapsed LEI means the registration was not renewed, not that the company stopped operating.
4. Silent conflicts. When sources disagree, the conflict is the answer. Report both values and their records rather than picking one quietly.

What an agent should return

For each identifier it uses, an agent should say where it came from: the issuing source, the record, and whether a primary record or only a community source supports any link to another identifier. That small amount of bookkeeping is what lets a reader, or another agent, check the answer in seconds instead of starting again.