Medicare Now Uses AI to Review Your Claims. Seven Months In, Here Is What WISeR Is Actually Doing.

Most of the conversation about AI in healthcare is about tools you choose to buy. This one you don't get to choose.

Since January 1, 2026, if you practice in Arizona, New Jersey, Ohio, Oklahoma, Texas, or Washington and you perform any of about a dozen outpatient procedures on a traditional Medicare patient, a private technology vendor running AI-assisted review decides whether Medicare will pay for it. The program is called WISeR — the Wasteful and Inappropriate Service Reduction Model — and it is the first CMS Innovation Center model in which technology companies, not medical practices, are the sole participants.

Seven months in, the picture is clearer than it was in January. Gold-carding exemptions started going out in July. A Congressional attempt to kill the model failed by four votes. A digital rights nonprofit is suing CMS to find out how the algorithms were tested. And the earliest affirmation figures to leak out of Texas are not reassuring.

This post covers what WISeR actually requires, what the early data shows, the oversight fight happening around it, and — the part most coverage skips — why your AI scribe may be writing beautiful notes that fail this review anyway.

The one-paragraph version

WISeR applies prior authorization to roughly a dozen outpatient service categories in six states, runs six years, and pays its six vendor participants a share of the savings from care that doesn't get approved. Standard decisions come in three days, expedited in two, and an affirmation is good for 120 days. CMS says a licensed clinician must sign off on every non-payment recommendation and that AI alone cannot deny care. A non-affirmation is not directly appealable — appeal rights attach only after you perform the service and the claim is denied. As of July 2026, clinicians who hit a 90% affirmation rate can be exempted entirely, which makes your first several submissions unusually consequential. And the documentation the reviewers want — dated conservative-therapy history, imaging that correlates with the clinical picture, explicit LCD criteria — is precisely the kind of structured detail ambient scribes are worst at capturing on their own.

What WISeR actually is

WISeR is a six-year mandatory model from the CMS Innovation Center, running January 1, 2026 through December 31, 2031 in six states. It was announced in June 2025 and finalized through a Federal Register notice on July 1, 2025.

The design is unusual in a way worth sitting with. Every prior CMMI model has recruited providers or health systems as participants. WISeR recruited vendors. CMS selected six technology companies, each assigned a single state:

• Cohere Health — Texas
• Genzeon Corporation — New Jersey
• Humata Health — Oklahoma
• Innovaccer — Ohio
• Virtix Health — Washington
• Zyter — Arizona

Each vendor conducts medical necessity review for its state, using what CMS describes as enhanced technology including AI and machine learning. The relevant Medicare Administrative Contractor still runs claims and appeals — in Texas, for example, Cohere makes the determination and Novitas Solutions handles any subsequent appeal.

How the vendors get paid

This is the structural fact that drives nearly all of the criticism. Model participants receive an incentive payment tied to a percentage of the savings from reduced spending attributable to WISeR. Reporting from Crowell & Moring puts the vendor share at up to 20% of program savings on denied care.

KFF's analysis of the model was direct about the implication, describing "financial incentives inherent in the WISeR model, which rewards vendors, in large part, based on the volume of care that they deny, creating financial incentives to maximize denials." The American Hospital Association raised the same objection in October 2025, saying the payment structure "incentivizes denials at the expense of physician medical judgment."

CMS's position is that contractors are not incentivized to deny but to get the determination right, and that guardrails prevent the obvious failure mode. Whether those guardrails hold is the empirical question the next five years will answer.

How many patients this touches

Per KFF, roughly 1.1 million traditional Medicare beneficiaries nationwide received at least one WISeR-covered service in 2024. About 207,500 of them — 19.7% — live in the six model states.

That is a real but bounded footprint. WISeR is not prior authorization for all of traditional Medicare. It is a targeted pilot, and the reason it matters far beyond those six states is that it is explicitly a test of whether this approach should be expanded.

The services under review

WISeR covers service categories CMS identifies as historically associated with fraud, waste, abuse, or low value. The categories include:

• Skin and tissue substitutes for wound care
• Epidural steroid injections for pain management
• Cervical fusion
• Lumbar decompression
• Percutaneous vertebral augmentation
• Arthroscopic knee procedures
• Electrical nerve stimulators and induced nerve lesions
• Sacral nerve stimulation for urinary incontinence
• Phrenic nerve stimulators
• Hypoglossal nerve stimulation for obstructive sleep apnea
• Incontinence control devices
• Services for the diagnosis and treatment of impotence

Review applies in four places of service: physician office (POS 11), home (12), hospital outpatient department (19 and 22), and ambulatory surgery center (24). Inpatient stays and emergency care are not included.

The exact HCPCS and CPT codes live in the appendices of the CMS WISeR Provider and Supplier Operational Guide, which has already been revised multiple times. Verify the current code list against your MAC before scheduling, not after. The University of Texas Health billing compliance group puts it plainly: best practice is to confirm a service's WISeR status before it's on the calendar.

The process, and the trap inside it

You have two paths for a WISeR-covered service, and they are not equivalent.

Path one: submit a prior authorization request. Decisions come back within three days for standard review and two days for expedited review, which requires justification that the standard timeline would jeopardize the patient's life or health. An affirmation is valid for 120 calendar days, counted from the decision date rather than the submission date. In Texas, the Texas Medical Association reports that portal submissions through Cohere return results in "seconds to minutes," while email, phone, and fax are meaningfully slower.

Path two: do nothing and bill the claim. The MAC then suspends the claim and routes it to the WISeR participant for pre-payment medical review. You have not avoided the review — you have moved it to after you already delivered the service and absorbed the cost.

Here is the part that catches practices: a non-affirmation is not directly appealable. Noridian, the MAC for Arizona and Washington, states it flatly. When a request is non-affirmed, your options are:

• Request a peer-to-peer review (in Texas, with a Cohere physician in the same specialty)
• Resubmit with additional documentation
• Decline to perform the service

Formal Medicare appeal rights attach only if you perform the service anyway, submit the claim, and receive an unfavorable determination or pre-payment denial. In other words, exercising your appeal rights requires taking financial risk first. That asymmetry is a meaningful change from how traditional Medicare has worked, and it is the specific feature that triggered the Congressional response below.

What CMS says about the AI

Two commitments matter. CMS asserts that "all recommendations for non-payment are determined by appropriately licensed clinicians" applying standardized procedures — the AI flags and prepares, a human decides. And under the 2024 final rule governing AI in coverage decisions, artificial intelligence may not serve as the sole basis for denying coverage in traditional Medicare, Medicare Advantage, or Medicaid.

Those are the right commitments on paper. The dispute is over verification, which is exactly what the litigation below is about. This is the same tension we covered in FDA's 2026 clinical decision support guidance: a stated human-in-the-loop requirement is only as good as the evidence that the human is actually deciding rather than rubber-stamping.

Gold carding: the exemption, and why your first ten requests matter

The most practically important development of 2026 arrived in July. CMS began granting exemptions — gold carding — that remove a clinician from prior authorization and pre-payment review entirely for the services where they've demonstrated compliance.

The mechanics, per reporting in the National Law Review:

• Launched July 6, 2026 in Washington state, with quarterly rollouts to the other five states following
• Requires a 90% affirmation rate on a minimum of 10 prior authorization requests during the assessment period
• CMS has capped the bar: no participant may set a threshold higher than 90%
• Exemptions are granted at the individual NPI level — not the facility, not the group
• They last at least one year, are re-evaluated quarterly, and can be revoked if your rate falls below the threshold

Read that structure carefully, because it creates a specific incentive. If you need 10 requests at 90% affirmation, then a single non-affirmation inside your first ten submissions costs you the exemption for that assessment period. The rational strategy is to over-document early rather than to learn the reviewer's standards by trial and error. That is the opposite of how most practices approach a new prior auth program.

It also means gold carding is earned per-clinician. A group where two partners are exempt and three aren't is a normal outcome, and worth planning scheduling around.

We wrote about gold carding as an industry trend in our prior authorization and CMS-0057-F guide. WISeR is the first place it shows up as a live federal mechanism with published criteria.

The early numbers, with appropriate caution

The most-cited early data point comes from the same National Law Review reporting: WISeR participants in Texas were affirming roughly 62% of prior authorization requests at initial review, against a 92% national affirmation average under traditional Medicare.

If that holds, it is a large gap — roughly a fivefold increase in the initial non-affirmation rate. But treat it carefully, and here is why:

• It is a single state, early in the program, during the period when both providers and vendors are learning a brand-new documentation standard
• "At initial review" is doing real work in that sentence. It excludes peer-to-peer reversals and resubmissions with better documentation, which are the designed remedies for a non-affirmation. The final rate after those channels is almost certainly higher.
• It comes from trade reporting rather than a published CMS dataset. CMS has not released model-wide affirmation statistics.
• The 92% comparison figure is a national all-services baseline, not a matched comparison against these specific high-scrutiny procedures, which had elevated denial rates before WISeR existed.

So: a genuine signal that early non-affirmations are running well above historical Medicare norms, not a proven denial rate. The absence of official data is itself part of the story, which brings us to the fight over transparency.

The oversight fight

GAO determination — May 12, 2026. The Government Accountability Office concluded that WISeR meets the Administrative Procedure Act's definition of a rule, because it prescribes new prior authorization requirements for traditional Medicare providers. That finding made the model eligible for repeal under the Congressional Review Act.

CRA resolution — introduced May 20, 2026. Senator Ron Wyden (D-OR) in the Senate, with Representatives Suzan DelBene (D-WA) and Greg Landsman (D-OH) in the House, introduced a resolution of disapproval to terminate the model.

Senate vote — failed 46–50, along party lines. WISeR survives.

Appropriations push — ongoing. DelBene, joined by Reps. Ami Bera, MD, Lloyd Doggett, Landsman, Rick Larsen, Alexandria Ocasio-Cortez, and Kim Schrier, MD, led a letter with roughly 28 additional colleagues urging House Appropriations leadership to repeal WISeR through the FY27 funding process.

EFF v. CMS — filed March 25, 2026. The Electronic Frontier Foundation sued CMS over an unanswered FOIA request. EFF sought three things: the agreements with participating software vendors, records of any testing for accuracy, bias, or hallucinations in the vendors' technology, and records of audits, monitoring, or evaluation of the model and its participants. Per EFF, CMS had produced none of it.

That third category is the one to watch. Every reassurance CMS has offered about WISeR — licensed clinicians make the calls, AI doesn't deny alone, contractors are incentivized toward accuracy — is currently unfalsifiable from outside the agency. The lawsuit is an attempt to change that.

Practically, the repeal effort's near-term prospects are poor after the Senate vote. Plan for WISeR to be in place through at least the FY27 appropriations cycle.

The part that connects to your AI scribe

Here is the thing almost no WISeR coverage says out loud: this is an AI system reviewing documentation that, increasingly, another AI system wrote. And the two are optimized for different things.

The reviewers apply Local Coverage Determination criteria. For epidural steroid injections under the relevant LCDs, medical necessity generally requires history, physical exam, and concordant radiological imaging supporting a specific diagnosis — radiculopathy, radicular pain, or neurogenic claudication from disc herniation, osteophyte complexes, or degenerative disease producing foraminal or central stenosis. For most WISeR procedures, coverage also turns on documented failure of conservative therapy.

The UT Health billing compliance guidance states the standard in three parts. Your note must show:

• What conservative treatment was tried
• When it was tried
• Why it failed or was insufficient

And explicitly: a checkbox reading "failed conservative therapy" is not enough.

Now consider what an ambient scribe actually produces. It transcribes and structures a conversation. It is very good at capturing what was discussed in the room. It is structurally bad at capturing what nobody said out loud — and patients do not spontaneously narrate "I completed six weeks of physical therapy beginning in March, then eight weeks of NSAIDs, and my radicular pain persisted at 7 out of 10." That chronology lives in prior records, referral letters, and the clinician's head.

The result is a specific and underappreciated failure mode: a fluent, well-organized, entirely accurate note that still gets non-affirmed, because it documents the visit rather than the coverage criteria. Note quality and note sufficiency are not the same property.

This is a different risk from the one in coding intensity and payer downcoding, where scribes push documentation toward higher-level codes. Here the scribe isn't inflating anything. It's just silent on the facts the reviewer needs.

What to actually do about it

If you practice in a WISeR state and perform covered services:

• Build a pre-procedure documentation checklist keyed to the LCD, not to your note template. Conservative therapy with dates and outcomes, imaging with the correlation stated explicitly, functional impairment described concretely. Verify it before the request goes out.
• Dictate the history the patient won't say. Whatever scribe you use, state the conservative-therapy chronology aloud during the encounter, or dictate it into the note afterward. If it isn't spoken or typed, it isn't in the note.
• Treat the first ten requests as the gold-carding audition. Have someone experienced review each one before submission until the exemption lands.
• Use the portal, not fax. Minutes versus days is a scheduling difference for your patient.
• Track your own affirmation rate. You cannot manage toward 90% without measuring it, and no one is going to hand you the number.
• Use peer-to-peer aggressively. It is faster than resubmission and it is a conversation with a same-specialty physician, not an algorithm.
• Don't default into pre-payment review. Skipping prior auth doesn't skip the review; it just moves your financial exposure to after you've already delivered the care.

For practices evaluating AI prior authorization tools to manage the burden — Myndshft, Develop Health, and Notable all work in this space — the honest framing is that these tools help with submission speed, status tracking, and payer-rule lookup. They do not manufacture a conservative-therapy history that was never documented. The upstream clinical documentation is still the constraint, which is also true of the AI coding and RCM tools like CodaMetrix and Nym Health working the same claim from the other end.

What this means if you're not in a WISeR state

Two things.

First, this is a pilot, and pilots that produce savings get expanded. WISeR runs through 2031 with an explicit mandate to test whether the approach reduces low-value care. Its evaluation results will shape whether prior authorization spreads further into traditional Medicare.

Second, the documentation discipline WISeR demands is not WISeR-specific. Dated conservative-therapy history, imaging correlated to the clinical picture, and explicit LCD criteria are what Medicare Advantage plans and commercial payers already want for these same procedures. Practices in small clinics and family medicine that tighten this now are hedging against a broader trend, not just one pilot.

The larger pattern is worth naming. Payers deployed AI to review claims. Providers deployed AI to write notes and fight denials. Now the federal government has deployed AI on the payer side of that exchange, with a shared-savings incentive attached to the denial. Nobody in that loop is arguing about whether AI belongs in the process anymore. The argument is entirely about whose AI gets to be right, and who bears the cost of being wrong.

Right now, on a non-affirmation you can't appeal until you've already taken the financial risk, the answer to that second question is: you do.

The bottom line

WISeR is live, it survived its repeal attempt, and it will run through at least the next appropriations cycle. If you're in one of the six states, the actionable items are narrow and concrete: confirm which of your services are covered before scheduling, submit through the portal, over-document conservative therapy and imaging correlation against the LCD, protect your first ten requests to earn the gold-carding exemption, and use peer-to-peer rather than absorbing a non-affirmation.

If you're not in one of the six states, the useful takeaway isn't about WISeR at all. It's that "the AI wrote a great note" and "the note satisfies coverage criteria" are two different claims, and only one of them gets you paid.

---

This article is informational only and is not legal, medical, or billing advice. WISeR requirements, covered code lists, and gold-carding criteria have been revised repeatedly since launch — verify current requirements against the CMS WISeR Provider and Supplier Operational Guide and your Medicare Administrative Contractor before making operational decisions. Vendor claims and pricing should be confirmed directly with the vendor.