US Business Registries: All 51 Ranked by Data Accessibility (2026)

Published August 2026. Every registry below was checked against its own primary source on 2026-08-18. Fees are quoted only where the state publishes one.

Say you are standing up a company database that covers the whole United States. You start with one state, find its business register, write the parser, and the work goes fine. The second state goes fine too. By the eighth you have eight parsers. By the twentieth you are maintaining a spreadsheet of which offices take a credit card, which take a cheque in the post, and which will not sell you the file at all.

That is not a tooling problem. It is the actual shape of US company data, and most teams meet it halfway through a build rather than before one.

So we did the boring version properly. In August 2026 we audited all 51 US business registries, the fifty states plus the District of Columbia, against their own official pages: what each one holds, what it will hand over, what it charges, how fresh it is, and what happens when an automated client knocks. Two of the assumptions we started with did not survive contact with the evidence. Both are flagged below, because the wrong ones are usually more useful than the right ones.

The Same Question, Fifty-One Answers

Below is the full audit. Each registry is scored out of 20 across six criteria: bulk availability, API, cost, field depth, freshness and licence. Tier A is open, B workable, C restricted, D closed or effectively closed.

Read the office column first. It is the least interesting column and the one that costs teams the most time, because ten of these registers do not sit with a Secretary of State. Arizona keeps its at a Corporation Commission. Michigan uses its licensing department, Wisconsin its Department of Financial Institutions, Maryland its tax assessment agency, Virginia a State Corporation Commission, New Jersey the Treasury. If your ingestion plan assumes fifty offices with the same name, it is already wrong.

Officers means the people who run the company appear in data you can obtain at scale, not merely in a record you can open one at a time.

Five Things That Will Break Your Plan

Every one of these is invisible from the outside. You find them after you have committed to a schema, a budget, or a refresh cadence, which is the expensive order to find them in. Each is anchored to a named state you can go and verify yourself.

1. Budget by state, not by country

A single national line item for registry data is the first thing to delete from the plan. The spread is not a rounding error.

It gets sharper when you read the terms rather than the headline. South Carolina’s fee runs against a July to June state fiscal year and is not prorated, so a subscription bought in April costs a full year for three months of data. Indiana charges $9,500 up front and then $500 for every monthly refresh, which is about $15,500 for one year current. Wyoming is $10,200. Nevada, for contrast, sells its entire unabridged database for $125.

Ten jurisdictions charge nothing at all. Nothing about the underlying filing differs between them. It is the same kind of statutory record describing the same kind of entity, priced by fifty-one separate administrative decisions.

2. Half the country will not tell you who runs the company

This one surprises people, and it is not an access problem. It is a collection problem. For a large share of the country, the state never asked.

• Wyoming does not collect LLC members or managers. Its Articles of Organization ask for five things and none of them is an owner.
• Colorado excludes officers, directors and owners from every product it sells, free or paid, and says so on its own extract form.
• Ohio collects incorporators and organizers. Not officers.
• New York records a chairman rather than a roster.
• Wisconsin keeps officers out of the database by design, so they survive only on annual report images you buy one at a time.
• Missouri collects officers for corporations but requires no LLC member disclosure at all.

Add it up and 22 of 51 give you officers at scale. Fifteen categorically will not. The rest are partial or sit behind a schema nobody publishes until you have paid.

The practical consequence is worth being blunt about. A nationwide, officer-level view of US companies assembled purely from state registers is not expensive. It is impossible. If your product roadmap has that as a milestone, it needs a second source before it needs a budget.

3. Delaware is a statute problem, not a procurement problem

Most closed doors in this audit are administrative. Somebody has to sign something, or the price is high, or the file arrives monthly when you wanted it weekly. Delaware is a different category, and it is worth understanding precisely because escalation will not move it.

Read what it actually does. It removes the duty to supply bulk data. It does not ban it, and the Secretary of State stays free to provide it voluntarily. That distinction matters if you are writing to them, and it matters more if you are writing about them.

What closes the door in practice is the second half of the provision, which prices access per record associated with a file number. There is no volume at which that becomes sensible. And Delaware is where the companies are: its own 2025 annual report counts 2,287,728 entities, with more than two thirds of the Fortune 500 and close to 70 percent of that year’s US IPOs incorporated there. The state that matters most is the one you cannot query in bulk.

4. Plan for bot management, not for geography

Here is the first assumption of ours that did not survive. We went in expecting to find registries blocking non-US traffic, and we designed the test to catch exactly that, running it from a non-US network where a country block would be most visible.

We found none. Not one of 56 endpoint tests returned country-block wording.

What we did find was bot management, on 21 of the 51: Cloudflare, Akamai, Imperva, F5, DataDome, Radware and AWS WAF between them. Four more serve their pages normally and gate only the search action behind a challenge, which is a different failure domain and should not be counted the same way. Two returned a bare 403. Seven were unreachable, which we record as unproven rather than as a block, because a silent timeout is not evidence.

The useful version of this finding is architectural. New York runs bot defence on its search interface while its open-data host serves the full 4.27 million row dataset over an unauthenticated API to the same client in the same second. The control sits on the interface, not on the data. Ohio is the cautionary twin: its challenge guards the data host itself and returns a 1.2 megabyte HTML page with a 403 status, so a fetcher that only checks for a response will ingest the challenge page and report success.

5. Check the file, not the metadata

Free data has its own failure mode, and it is quieter than a paywall.

Virginia publishes fourteen free datasets on its open-data portal, including a separate officer file, which on paper makes it one of the better states in the country. The metadata claims monthly updates. The underlying files have not been regenerated since November and December 2024.

The dataset timestamps move. The resources do not. Anyone who trusted the portal and skipped the file check is running on data approaching two years old and has no signal telling them so. Fourteen of the 51 either publish no cadence at all or publish files that are demonstrably stale, which is why freshness is scored separately here rather than folded into availability.

How We Scored This

Nine facts were established per jurisdiction against primary sources: the office holding the register, the public search URL, whether bulk data exists and how you obtain it, its cost, whether an API exists, whether the state publishes to an open-data portal, which fields are exposed, the refresh cadence, and any stated restriction on commercial use, redistribution or automated access. Those roll into six scores.

Fees appear only where a state publishes them. Vendor summaries were not treated as pricing sources, and that decision changed at least one answer: the California bulk price commonly quoted across the industry turned out to be partly wrong, with the real figure sitting in the Secretary of State’s help guide rather than its fee schedule. Automated-access behaviour was observed directly, from a single non-US network, with every host resolved through two public DNS resolvers before any network failure was recorded.

Where This Audit Is Thin

The second assumption that did not survive was our own confidence. A first pass produced a blocking figure we could not defend once we checked the sample properly, so it was rebuilt. Here is what is still soft, stated plainly so you can weight it yourself.

• Six jurisdictions publish no bulk price anywhere public. We record that as unpublished rather than estimating it. Several need a phone call to settle.
• Officer presence in a delivered file, as opposed to the public record, is unverified wherever the schema sits behind a login or a signed agreement. You cannot see the field list before you pay.
• Seven registries were unreachable during testing and are recorded as unproven, not as blocks.
• Wisconsin’s search host was not reached in the automated-access retest, and Delaware’s returned an origin error, so its terms of use were confirmed against two archived captures rather than the live page.
• Four jurisdictions changed registry systems in the fourteen months before publication. DC replaced its portal eight days before we ran this. Treat any registry audit, including this one, as perishable.

This guide publishes the map, not a route around anyone’s controls. There are no workarounds here for the states that restrict automated access, and no state’s dataset is reproduced.

Questions We Get Asked About This

Is there a national US company register?

No. Company formation is exclusively a state act, so there are 51 separate registers and nothing above them. SEC EDGAR covers only companies with a securities-law filing obligation, and the IRS exempt-organizations file covers only nonprofits. Neither is a company register, though both get mistaken for one.

Which state has the most accessible business registry?

Connecticut, at 20 out of 20. It publishes six datasets covering the master register, agents, principals and full filing history, refreshed daily, free, through a public API, with the master dataset in the public domain. Colorado follows at 19, then the District of Columbia, New York and Oregon at 18. The catch with Connecticut is structural rather than commercial: the data is normalised across four datasets you have to join, and the filing history is rebuilt monthly, so snapshot it if you need continuity.

Can I download all US company data from one place?

Not from any official source. You would be assembling 51 feeds across at least six delivery mechanisms, paying for roughly half, signing agreements for several, and accepting that nine states sell nothing in bulk at any price. That fragmentation is the reason commercial company-data products exist, and it is also why they disagree with each other.

Do state registries include company owners and directors?

Sometimes, and less often than most people assume. Twenty-two of the 51 expose officers in data you can obtain at scale. Fifteen categorically do not, because the state does not collect them. Wyoming never asks an LLC for its members or managers, so no amount of money produces that data from the state. If you need people rather than companies, the register is a starting point and not a source.

The Fragmentation Is Permanent. Plan For It.

None of this gets tidier. There is no bill coming that harmonises fifty-one filing offices, and the federal layer moved further away this month rather than closer. The registries will keep changing systems, keep repricing, and keep publishing metadata that disagrees with the files underneath it.

Which is a workable position, as long as it is the position you planned for. Score the states you actually need instead of the country. Start with the ones that will hand you a file, and treat the closed ones as a separate, smaller, and more expensive problem rather than as a blocker to the whole build. Budget per state. Check timestamps, not metadata. Assume the office name is wrong until you have looked.

We maintain extraction pipelines across sources that behave exactly like these: fragmented, inconsistently priced, and changing underneath you without notice. If you would rather not maintain fifty-one of them yourself, talk to our team. And if you find a state where this audit has gone stale, tell us. That is the nature of the material, and the table is easier to keep honest with more eyes on it.