In network when you checked, out of network at the visit

Checking a directory tells you what was true when the page was built, not what is true on the day of care. We watched 41 Medicaid provider directories for 20 days. A change turned up on 43.7% of checks. Federal law only requires verification every 90 days.

The short version

You look up a doctor. The directory says in network. You book. Six weeks later you go in, and the claim comes back out of network.

Nothing about that story requires anyone to have lied to you.

A directory is a snapshot. It tells you what the plan believed when the page was last built. The day of care is a different day, and the gap between those two days is where the money goes wrong.

We measure that gap, so here are our numbers.

Two questions that sound like one

"Is this doctor in network" is really two questions.

Was this doctor in network when the page was published? And will this doctor be in network when I walk in?

Only the first one has an answer you can look up. The second one is a prediction. Most people asking the first question think they are getting an answer to the second.

What the law actually requires

This part surprises people, so here is the exact standard.

A plan must verify and update its provider directory not less frequently than once every 90 days (42 U.S.C. § 300gg-115(a)(2)(A)).

Once every 90 days is the floor. A directory that was verified 89 days ago is fully compliant.

Two other timeframes matter, and they are shorter.

RequirementDeadlineWhere
Verify and update the directoryAt least every 90 days§ 300gg-115(a)(2)(A)
Update after a provider sends new informationWithin 2 business days§ 300gg-115(a)(2)(C)
Answer a phone question on network status, in writingNo later than 1 business day§ 300gg-115(a)(3)(A)
Keep that written answer in your fileAt least 2 years§ 300gg-115(a)(3)(B)

Hold on to that last row. It matters later.

We watched 41 directories for 20 days

We archive state and plan Medicaid provider directories and record whether the published content changed between captures. Here is the window from 1 August to 20 August 2026.

MeasureValue
Directories watched41
Successful checks710
Checks that found a change310
Share of checks that found a change43.7%
Directories that changed at least once29 of 41
Directories that never changed12 of 41

Under half of our checks found something different from the last one. In under three weeks.

Most directories are frozen or churning, not in between

The average hides the real shape. When we split the 41 by how often they changed, they land at the two ends and almost nowhere in the middle.

Changes in the windowDirectories
012
1 to 311
4 to 91
10 or more17

One directory sat in the middle. One.

Sixteen of the 41 changed on 90% or more of our checks. Four of them changed on 17 of 18 checks, which is 94.4%: the Texas, Arizona and Illinois Molina Medicaid directories, and the Pennsylvania state provider directory.

For those four, a lookup is close to a one-day answer. For the 12 that never moved, the date you checked did not matter at all. Same market, opposite behaviour.

The same 1.9 million rows, measured twice

Change counts tell you a page moved. They do not tell you whether it got more accurate.

So we ran a second measurement. We took the same 1,909,677 provider directory rows and compared them against the federal provider registry on two dates, two weeks apart.

Measure1 Aug 202615 Aug 2026Change
Directory rows checked1,909,6771,909,6770
Listed but deactivated in the federal registry5,3375,630+293
Name does not match the registry30,75331,426+673
Entity type does not match7,1747,169−5
NPI not found in the registry6362−1

In 14 days, 293 more listed providers showed up as deactivated in the federal registry. That is about 21 a day, on rows nobody removed.

The directories did not get worse because anyone was careless. They got worse because the world moved and the page did not.

What "changed" does and does not mean

We want to be exact about what we measured, because it is easy to oversell.

Our change flag means the published content is different from the previous capture. It does not mean a doctor joined or left. A directory that reprints today's date on every page load will show as changed every single day, and some of these do.

So read the 43.7% as an upper bound on meaningful change, not as a count of network moves. The two-week registry comparison in the previous section is the stronger evidence, because it compares provider records rather than page bytes.

If the directory was wrong, you are protected

This is the part worth knowing before you need it.

If you relied on incorrect provider directory information and got care from an out-of-network provider, the plan must limit your cost sharing to in-network terms. It must also apply the amount to your in-network deductible and out-of-pocket maximum (CMS).

The provider may not bill you more than in-network cost sharing. If they did bill you more and you paid it, they must refund the excess with interest (CMS).

That protection turns on the word "relied." Which brings us back to that last table row.

Get the answer in writing

Ask your plan by phone whether the provider is in network. They must answer in writing, no later than one business day. They must keep that answer in your file for at least two years (42 U.S.C. § 300gg-115(a)(3)).

A screenshot of a directory page is weak evidence. A written answer from the plan, dated, that the plan is itself required to retain, is strong evidence.

It costs you one phone call.

What to do before the day of care

Check close to the appointment, not once at enrollment. A February lookup says very little about an April visit.

Check again if anything moves the date. A rescheduled procedure is a new lookup.

Confirm with the plan and the office both. The office knows whether the contract is ending. The plan is the one bound by the one-business-day rule.

Keep the written answer. You may never need it, and it is the difference between a claim you can fight and one you cannot.

Where these numbers come from

Source data, part one: our own Medicaid directory archive. 41 state and plan Medicaid provider directories, captured on a fixed schedule, with a content hash recorded per capture. Window: 1 August 2026 to 20 August 2026, up to 18 successful captures per source. Source data, part two: our provider directory table, 1,909,677 rows, compared against the federal NPPES provider registry on 1 August 2026 and again on 15 August 2026. Queried: 20 August 2026.

Four honesty notes. The change flag detects content differences, not confirmed network moves, and we say so above rather than burying it. The 41 directories are the set we archive, not every directory in the country. The two registry comparisons are two points, not a trend, and we will not call a direction until we have more. And a deactivated NPI in the federal registry is strong evidence a listing is stale, but it is not by itself proof the provider stopped seeing patients.

Our other live data work is on the apps page, and the rest of the writing is on the blog.

Questions people ask

How old can a provider directory legally be?

The plan must verify and update it at least once every 90 days. A directory verified 89 days ago meets the federal standard, even if the network changed the day after.

How often do provider directories actually change?

In our 20-day window across 41 Medicaid directories, 310 of 710 checks found changed content, or 43.7%. Sixteen of the 41 changed on 90% or more of checks.

What happens if I get billed out of network after the directory said otherwise?

Your cost sharing is limited to in-network terms and counts toward your in-network deductible and out-of-pocket maximum. If you already paid more, the provider must refund the excess with interest.

Does a screenshot of the directory protect me?

It helps, but a written answer from your plan is stronger. The plan must respond to a phone question within one business day in writing and keep that response for two years.

How far in advance should I check?

As close to the appointment as you reasonably can, and again if the date moves. Checking once at enrollment tells you very little about a visit months later.

Are some directories more reliable than others?

Sharply so. Twelve of the 41 we watch did not change once in 20 days, while 17 changed 10 or more times. The variation between directories is larger than the variation over time.