There is a person in your clinic, almost certainly the lowest-paid person involved in the patient's care, who is about to spend forty minutes on the phone discovering that your health system's referral list is largely fictional.
She has the printout. She starts at the top. The first number rings out. The second is a fax line. The third is answered by a receptionist who explains that the doctor left the practice in 2023. The fourth is a real psychiatrist who is not taking new patients and has not been for two years. The fifth does not take that insurance despite being listed under it. The sixth has a voicemail box that is full.
On the seventh or eighth call, if she is lucky and persistent, she finds someone. The wait is eleven weeks.
She writes the name on a sticky note and puts it in her desk drawer, because she has learned that the drawer is more reliable than the directory.
That sticky note is one of the most important artifacts in American medicine. It is a private, hand-maintained, entirely undocumented expertise graph, built by clinic staff out of necessity because the official one does not work. Every busy practice in the country has some version of it. None of them are shared, none survive staff turnover, and none of them are in any database anywhere.
This article is about why the official version fails, why fifteen years of regulation has not fixed it, and what the failure reveals about the deeper problem: medicine's only public map of who does what is not merely inaccurate, it is measuring the wrong thing entirely.
The number that got a Senate hearing
In 2023, the US Senate Finance Committee did something refreshingly direct. Rather than asking insurers to report on their own directory accuracy, investigators became patients.
Staff conducted a secret-shopper study, calling listed in-network mental health providers across multiple Medicare Advantage plans in several states, and simply attempting to book an appointment the way any patient would.
They made 120 calls.
They were able to obtain an appointment 18 percent of the time.
More than 80 percent of the listed providers were, in the committee's own word, "ghosts." They were unreachable, not actually in network, or not accepting new patients. Roughly ten calls were needed to reach two bookable providers.
The phrase "ghost network" entered the policy vocabulary at that moment, and it was a genuinely useful piece of naming. It converted a diffuse frustration that every clinician had experienced into a specific, countable, investigable phenomenon.
The peer-reviewed version is worse, because it is unarguable
Secret-shopper studies are powerful but can be dismissed as anecdotal by anyone motivated to dismiss them. So the more important evidence is the claims-based analysis, because claims data cannot be talked around.
Researchers Zhu, Charlesworth, Polsky, and McConnell published an analysis in Health Affairs examining Oregon Medicaid provider directories against actual claims. Their method was elegantly simple: for every provider listed in the directory, did that provider see even a single Medicaid patient during the measured year?
The findings:
- 58.2 percent of behavioral health directory listings were "phantom" providers who saw zero Medicaid enrollees that year.
- Among prescribers specifically, the phantom rate was 67.4 percent.
- Among non-prescribers it was 59.0 percent.
- Even in primary care, generally assumed to be far more reliable, 54.0 percent were phantoms.
Read the primary care number again. More than half of listed primary care providers saw no patients from that program at all.
The authors' conclusion was carefully worded and quietly devastating: significant discrepancies between listed providers and those enrollees can actually access "suggest that provider network monitoring and enforcement may fall short if based on directory information."
In plain language: the regulatory apparatus that determines whether an insurance network is adequate is built on data that describes a network that does not exist.
It is not a behavioral health problem. It is every specialty.
The natural instinct is to file this under mental health access, which is known to be uniquely broken. That instinct is wrong, and the study that proves it is one of the most under-discussed papers in health services research.
Published in JAMA in 2023, researchers examined 449,282 physicians and compared their directory entries across five major national insurers.
Only 19.4 percent had consistent entries on address and specialty across all five.
Which means 81 percent of American physicians are listed inconsistently depending on which insurer you ask. The same doctor, the same week, described differently by five organizations that all claim to have verified the information.
A follow-up analysis in BMC Health Services Research in 2024 examined address consistency specifically and found it ranged from 16.5 to 27.9 percent across insurers, and from 13 percent in Minnesota to 47 percent in the District of Columbia by state.
Notice what that narrow range across insurers tells you. If one insurer were far better than the others, this would be a vendor quality problem with a vendor solution. The fact that all five perform similarly badly means the failure is structural, not operational. Nobody is doing this well because the incentives do not permit anyone to do it well.
CMS reviews have found similar patterns in Medicare Advantage, with roughly half of directory locations containing at least one inaccuracy, and 40 percent of flagged errors still uncorrected 500 days later. An HHS Office of Inspector General examination in Arizona found 55 percent of listed Medicare Advantage behavioral health providers were inactive. In 2026, the New York Attorney General reached a $2.5 million settlement with EmblemHealth after finding that 82 percent of contacted directory providers were unavailable.
Different regulators. Different states. Different methods. Same answer.
The deeper error: "accepting new patients" is not a field
Here is the part that everyone working on this problem gets wrong, and it explains why fifteen years of increasingly strict regulation has produced so little.
Every fix so far has treated this as a data accuracy problem. The No Surprises Act requires verification of directory information every 90 days. CMS audits Medicare Advantage plans. Vendors including HiLabs, Ribbon, and Madaket sell data-scrubbing services to payers. State regulators impose penalties. All of it aims at the same target: make the fields correct.
But consider what a perfectly accurate directory would actually give a referring clinician. Correct name, correct address, correct phone number, correct specialty, correct network participation status.
And it would still be nearly useless, because none of those fields answers the only questions that matter:
- Will this person see my patient, this month?
- Do they actually do the specific thing my patient needs, right now?
- If I send the patient, will I ever hear what happened?
"Accepting new patients" is not a static attribute of a person. It is a state that changes weekly, depends on the referring relationship, varies by patient complexity and payer, and often differs from what the front desk tells a cold caller. A specialist with a closed panel will frequently see a patient for a colleague who calls personally. That same specialist is a "ghost" to a directory.
This is the conceptual error at the center of the whole problem. A directory is a static artifact describing a dynamic relationship. You can make the artifact perfectly accurate and it will still fail, because the thing it describes is not a fact about a person. It is a state between two people.
This is why the sticky note in the desk drawer outperforms a multi-million dollar directory infrastructure. The sticky note does not record who exists. It records who called back last time.
Follow the incentives
Once you look at who produces this data and who bears the cost of it being wrong, the persistence of the problem stops being mysterious.
Directories are compiled from contracting data. An insurer's directory reflects who signed a network agreement. It is fundamentally a legal roster, not a clinical availability system. A provider who signed a contract in 2019, moved states in 2022, and stopped taking that plan in 2023 may remain a valid contract record long after becoming a clinical fiction.
Directory breadth is a regulatory asset. Insurers must demonstrate network adequacy to regulators. A broad directory helps with that filing. An operationally accurate directory that revealed how few listed providers are genuinely available would create adequacy problems. This is not a claim about intent; it is a description of what the incentive structure rewards.
The people who bear the cost have no input. The referring clinician and their staff absorb the entire cost of the failure in unbillable labor and delayed care. They cannot correct the record. They have no channel to report that an entry is dead. The most informed party in the system, the person who called yesterday and found out the truth, is structurally silenced.
Enforcement is weak relative to the effort of compliance. A $2.5 million settlement is a meaningful signal but a rounding error against the cost of continuously verifying hundreds of thousands of entries, and it arrives years after the harm.
Put those together and you get a market where nobody who could fix it wants to, and everybody who wants to cannot.
What it costs, conservatively
The labor cost alone is substantial and almost entirely invisible because it never appears as a line item.
Research in the International Journal of Medical Informatics found referring practices spend roughly 10 percent of clinic time on referral management. If even a fifth of that is chasing bad directory data, across roughly 250,000 US primary care clinicians, at a loaded staff cost of $100 an hour over 1,800 annual hours, the arithmetic lands near $900 million a year in wasted labor. Every assumption there is debatable and the order of magnitude is not.
Then the downstream harms:
- Coverage failures. An AMA survey found 52 percent of physicians reported patients encountering coverage problems stemming from directory errors at least monthly. Out-of-network bills follow.
- Abandoned referrals. MGMA reported in 2025 that 38 percent of referrals stall entirely, most often because nobody follows up when the process gets stuck.
- Untreated illness. Research in Health Affairs found that about two-thirds of primary care physicians could not obtain outpatient mental health services for their patients. Separate work in PRiMER found roughly 70 percent of primary care physicians in one rural region had no psychiatrist to readily refer to, with half of studied counties having no psychiatrist at all.
That last set of findings deserves a note of intellectual honesty. Some of what looks like a directory failure is genuine absence. When half the counties in a region have no psychiatrist, no directory can conjure one. The distinction matters enormously for policy: a routing failure and an expertise desert require completely different responses, and the current data cannot separate them.
That is itself an argument for better measurement. Right now we cannot tell how much of the access crisis is that nobody exists versus that nobody can be found.
What would actually work
If accuracy is the wrong target, what is the right one?
Reachability, verified by the person who tried.
The single most valuable piece of information in this entire domain is not held by insurers, vendors, or regulators. It is held by the medical assistant who made eight calls yesterday. She knows exactly which entries are dead, which numbers work, who called back, and who actually accepted the patient.
That knowledge currently ends its life on a sticky note in a drawer and is destroyed when she changes jobs.
A functional system would have three properties that no existing system has:
One: the specialist declares availability to verified peers, not to the public. A closed-panel specialist will often see a colleague's patient. Making that state visible only to verified referring clinicians, rather than publicly, solves the flooding problem that makes specialists reluctant to advertise availability at all.
Two: referrers report back. Whether the referral connected, how long it took, whether the patient was actually seen. This is the feedback loop that no directory has ever had, and it is the only mechanism that keeps a listing honest, because it is generated by the person with no incentive to inflate it.
Three: the record belongs to the clinicians, not to a payer. The reason directories rot is that their owner's interest is adequacy filings. A record maintained by the people who use it has a different failure mode entirely.
Notice that all three properties are social rather than technical. The technology to build a directory is trivial and has existed for decades. What has never existed is a body of verified clinicians with a norm of telling each other the truth about who actually answers.
What you can do now
If you refer patients
Formalize the drawer. Your practice already has informal knowledge about who actually accepts referrals. Write it down properly, with dates. Name, subspecialty focus, last confirmed acceptance date, typical wait, and who in your office has the relationship. Ten entries beats a thousand-name directory.
Record the call outcome, always. When your staff burns forty minutes finding out that six entries are dead, that discovery is valuable information that currently evaporates. Two lines in a shared document preserves it.
Ask the specialist directly, once a quarter. "Are you taking new referrals from us right now, and what should we send you?" is a five-minute call that outperforms any directory query, and it converts a listing into a relationship.
Report errors upward anyway. Under the No Surprises Act, plans have verification obligations and error-reporting pathways. Compliance is imperfect, but a documented pattern of reports is exactly what state regulators and attorneys general have used to build the enforcement actions that are now producing settlements.
If you are a specialist
Tell your referrers your real state. Most specialists have never explicitly told their referring physicians what they currently want to see and whether they have capacity. Your directory entry is doing that job badly on your behalf right now.
Check what the insurers say about you. Given that 81 percent of physicians have inconsistent entries across five insurers, the odds that at least one is describing you wrongly are high. Patients are being misdirected to or away from you based on it.
If you run a system or a plan
Measure reachability, not accuracy. Run your own secret-shopper study on your own network, using the Senate Finance methodology, which is published and replicable. It takes a few staff days. You will learn more about your network than any adequacy filing has ever told you.
Separate absence from unreachability. For each specialty and county, distinguish how many listed providers are unreachable from how many genuinely do not exist. These require entirely different interventions and are currently reported as one number.
Create a correction channel for referrers. The clinicians calling your directory are your best data source and you almost certainly have no way for them to tell you what they learned.
Frequently asked questions
What is a ghost network? A ghost network is an insurance provider directory containing large numbers of listings that cannot deliver care: providers who are unreachable, no longer in network, not accepting new patients, or not practicing in the listed specialty or location. A 2023 US Senate Finance Committee secret-shopper study of Medicare Advantage mental health listings obtained an appointment in only 18 percent of 120 calls, with more than 80 percent of listings classed as ghosts.
How inaccurate are provider directories? Substantially, and consistently across insurers. A JAMA analysis of 449,282 physicians across five national insurers found only 19.4 percent had consistent address and specialty entries. A Health Affairs analysis of Oregon Medicaid found 58.2 percent of behavioral health listings were providers who saw zero enrollees, and even primary care was 54.0 percent. CMS reviews have found roughly half of Medicare Advantage directory locations contain at least one inaccuracy.
Why are provider directories so wrong? Because they are built from contracting data rather than clinical availability data. A directory records who signed a network agreement, not who is currently seeing patients. Broad directories also help insurers demonstrate network adequacy to regulators, while the cost of inaccuracy falls entirely on referring clinicians and patients, who have no ability to correct the record.
Do the No Surprises Act directory rules fix this? They have improved the legal framework by requiring verification every 90 days and creating liability protections for patients misled by inaccurate directories, and they have supported enforcement actions such as the New York Attorney General's $2.5 million EmblemHealth settlement after finding 82 percent of contacted providers unavailable. They have not solved the underlying problem, because verifying that a contract exists is different from establishing that a clinician will see a patient this month.
How do I find a psychiatrist who is actually taking new patients? Practically, the highest-yield approach today is relational rather than directory-based: ask colleagues who have successfully placed a patient recently, maintain your own dated list of confirmed acceptances, and call the specialist's office directly to ask what they currently want to see. Directory-first approaches fail most of the time in behavioral health specifically, where phantom rates among prescribers have been measured at 67.4 percent.
Is this only a behavioral health problem? No. Behavioral health is the most extreme case and the best documented, but the JAMA analysis covering 449,282 physicians across all specialties found 81 percent with inconsistent listings, and the Oregon analysis found 54 percent phantom rates even in primary care. The mechanism is identical across specialties.
The bottom line
The provider directory is the only public map of who does what in medicine. It is the artifact every referral, every network adequacy filing, and every patient search depends on.
It is wrong for four out of five physicians. In behavioral health, most of it describes people who saw nobody at all.
And the most important thing to understand is that fixing the accuracy would not fix the problem. Even a flawless directory would tell you who exists, when the question every referring clinician is actually asking is who will answer.
That information exists. It is generated fresh every single day by clinic staff making phone calls, and it is thrown away every single day when they hang up.
The sticky note in the drawer is right and the database is wrong, and the reason is not technology. It is that the sticky note records the only thing that matters, which is what happened when a human being actually tried.
Part of a series on the missing professional infrastructure of healthcare. Previously: The 92% Problem: Medicine's Strongest Referral Lever Is Sitting Unused
Evidence note: primary sources include Zhu, Charlesworth, Polsky and McConnell in Health Affairs (2022) on Oregon Medicaid phantom providers; the JAMA (2023) analysis of 449,282 physicians across five insurers; BMC Health Services Research (2024) on address consistency; the US Senate Finance Committee 2023 secret-shopper report; HHS Office of Inspector General findings in Arizona; CMS Medicare Advantage directory reviews; ProPublica reporting on the New York Attorney General's EmblemHealth settlement; and Health Affairs and PRiMER research on primary care access to mental health services. The $900 million labor estimate is the author's arithmetic from published clinic-time research and is explicitly an estimate, not a measured figure.
