Organization schema is the structured block on your site that tells a machine who you are: name, address, founders, identifiers, and the other profiles that belong to you. Most sites either skip it or paste a generator's output and never look again. Both leave an answer engine guessing when it tries to decide whether the company in a Reddit thread, the company on G2, and the company on your homepage are the same thing.
This post walks the fields in the order we check them during an AI visibility audit.
name and legalName
name is how you want to be referred to. legalName is what appears on your incorporation papers. Keep them separate. If you trade as "Acme" but the entity is "Acme Software Inc.", put both. Models tend to collapse variants, and a single canonical name gives them the variant to collapse toward.
Match it exactly to the name on your LinkedIn page, Crunchbase, and review profiles. "Acme", "ACME", and "Acme Software" are three spellings that look like three companies to a disambiguation step.
url and logo
url should be the canonical homepage with the scheme and trailing slash you actually serve. logo should point to a square or near-square image that does not change between deploys. These are low drama, but a logo URL that 404s is a trust signal in the wrong direction.
sameAs
This is the field that does the most work and gets the least attention. sameAs is an array of URLs for profiles that are provably yours: LinkedIn, Wikidata, Wikipedia if you have an entry, Crunchbase, GitHub, G2, Capterra, YouTube, X. Each one is a link in a chain that lets an engine confirm the entity across the sources it cites.
Two rules. Only list profiles you control or that are unambiguously about you. And make sure the facts on those profiles agree with the facts in this schema. A sameAs pointing to a LinkedIn page with a different founding year creates a contradiction rather than resolving one.
description
One or two sentences, factual, in the phrasing you want quoted. "B2B payroll software for US companies with 10 to 200 employees" beats "the modern platform for growing teams". In our testing, the description field tends to show up in how engines summarize a company when nothing better exists on the page itself.
address and contactPoint
Use PostalAddress with every component filled: street, locality, region, postal code, country. A model reading "Austin" without a country has to infer. contactPoint with a contactType of "customer support" or "sales" and a real email or phone adds a verifiable detail. Fake or placeholder values are worse than omitting the field.
founder, foundingDate, numberOfEmployees
founder takes a Person object with a name and ideally a sameAs to their LinkedIn. foundingDate is an ISO date. numberOfEmployees is a QuantitativeValue, which lets you give a range rather than a false precision. These three are the fields that show up in "who is behind X" and "how big is X" questions, so they are worth getting right.
Identifiers: duns, taxID, leiCode, naics
Not every company has these, and that is fine. If you do, include them. An identifier is the strongest possible disambiguation signal because it cannot be shared by two companies. A DUNS number or LEI code is a single value that resolves the entity with no interpretation.
What to skip
slogan, award, and knowsAbout are rarely read in a way that changes an answer. They are not harmful, but they pad the block and are easy to let drift. If a field is marketing rather than fact, leave it out.
Keeping it consistent
The schema is only one copy of the truth. The same values need to appear in your llms.txt, your footer, your LinkedIn page, your Wikidata entry, and your review profiles. At AEO GEO Labs we generate Organization schema from a single source document and diff it against every third-party profile, because the most common failure is not a missing field but a field that disagrees with another source.
Check yours
Pull your Organization schema with the free schema markup generator at AEO GEO Labs or any validator, then open your LinkedIn page and your top review profile beside it. Every field that differs is a place an engine has to pick a winner. Fix the disagreements first, then add the fields you are missing, in the order above.