Most web addresses in the national doctor directory don't say who they belong to

Some numbers here have moved. This update measures the 2026-05-08 release. At 2026-08-20 there are 110,973 usable web addresses rather than 114,071, and 16,262 of them name an owner rather than 19,334, so attribution fell from 16.9% to 14.7%. The claim that health insurers are not in the directory is also out of date: 27 payer organizations and 233 health plans arrived, though none of them publishes an address. See the 2026-08-21 update.

The US government keeps a big list of doctors, clinics and hospitals. It is called the National Provider Directory.

Part of that list is web addresses. Software uses them to look up a patient's records. We checked 114,071 of these addresses and asked one simple question: does the list say who each address belongs to?

Usually it does not.

What we found

Only 19,334 of the 114,071 addresses say who owns them. That is about one in six.

The other 94,737 are web addresses with no name attached.

Picture a phone book where five out of every six numbers have no name next to them. The numbers might ring fine. You still would not know who answers.

Why this matters

A doctor's office wants to send your records to a specialist. Software looks up the specialist's address in the directory and finds one.

If nothing says whose address it is, the software cannot safely use it. Sending medical records to the wrong place is not a small mistake.

So the missing name is not a small detail. It is the difference between an address you can use and one you cannot.

The good news, and the free list

Nothing here is broken. The names were simply never filled in.

We checked that carefully. Every time the directory does name an owner, that name points to a real organization. Not one was wrong or led nowhere. So this is a blank box, not a bug.

That matters because a blank box is much easier to fix.

We also built a free list of the 19,334 addresses that do name an owner. Each row gives you the web address, the organization, and that organization's ID number. Anyone can download it. No login, no key.

https://ainpi.dev/api/v1/findings/endpoint-org-crosswalk.csv

Some companies fill the name in and some never do

We expected the missing names to be spread out evenly. They are not.

Most doctors' offices do not run their own software. They buy it from a company, and that company publishes the addresses on their behalf. So we grouped the addresses by company.

Sixteen companies have not filled in the name a single time. Together they account for 49,036 addresses.

This is encouraging, oddly. A company sitting at zero has not half-finished the job. It has not started. That is usually a much easier conversation than fixing something half broken.

One warning if you use this data. The company name is not the doctor's name. One company can run software for thousands of different clinics on the same web address. Knowing the software company tells you nothing about which clinic you have reached.

Health insurers are not in the directory yet

The published rulebook for the directory says it is meant to serve health insurers too, alongside doctors and hospitals.

So we checked whether you can look up a health insurer's public doctor list through the directory today. You cannot.

Out of 114,071 addresses, one belongs to an insurer's doctor list. The directory also has no category for "insurer" at all. Every organization in it is filed as a provider, a team, or a government body.

We wanted to be fair here. Maybe insurers simply have not built these lists yet, which would not be the directory's fault. So we picked one and checked. Capital BlueCross publishes a working public doctor list right now, and the government requires insurers to publish one. It is not in the directory.

So the lists exist. The directory just does not carry them yet.

What we still do not know

We can measure what is missing. We cannot yet explain it.

We do not know why some software companies fill in the name and others never do. It could be a setting nobody switched on, a step missing from their instructions, or something harder. Nobody has asked them.

We do not know if the 94,737 unnamed addresses have an owner recorded somewhere else that we cannot see.

We do not know whether a future version of the directory will carry insurer lists, or how far off that would be.

These are answerable questions. They just need someone to ask the right people.

What we think should happen next

CMS is building this directory in the open and asking people to help shape it. Here is where we think the effort pays off most.

Start with the sixteen companies at zero. They cover 49,036 addresses between them. Fixing that one field for those companies would do more than any other single change, because it moves the largest block of addresses from unusable to usable.

Add a category for health insurers. Right now the directory has no way to file one, so nobody could add an insurer list even if they wanted to.

Treat a web address with no owner as incomplete. If the directory asks for the name up front, this problem does not come back later.

Check our work

We would rather you did not take our word for any of this.

Every number above comes from a file anyone can download.

• The free list: https://ainpi.dev/api/v1/findings/endpoint-org-crosswalk.csv
• How we counted the addresses: https://ainpi.dev/findings/endpoint-org-linkage
• How we counted the insurers: https://ainpi.dev/findings/ndh-payer-endpoint-coverage

The insurer list we tested is public too. It needs one extra setting, and it returns an error in a normal browser, so use this:

curl -H 'Accept: application/fhir+json' \
  'https://providerdirectory-api.capbluecross.com/r4/Practitioner?family=Smith&_count=1'

All of it is public data. Repeating our work costs about three cents.

— Eugene Vestel, FHIR IQ