How Payers Use AI to Deny Claims Faster — And How to Fight Back

In 2023, UnitedHealth Group's subsidiary nH Predict came under federal scrutiny after a lawsuit alleged the AI system was denying post-acute care claims at a rate of 90% — overriding physician recommendations at scale, in milliseconds, without reading a single patient chart.

That lawsuit was not an anomaly. It was a glimpse into how prior authorization works in 2026.

Insurance companies are not staffing up teams of physicians to review your prior auth submissions. They are deploying AI systems that scan submissions against Clinical Policy Bulletins, flag missing documentation elements, and generate denial letters — faster than any human review process ever could.

Your billing coordinator is writing appeal letters by hand against an algorithm.

This article explains how payer AI actually works, what it is looking for, and what independent practices can do to compete on equal terms.

How Payer AI Processes a Prior Authorization Request

When your practice submits a prior authorization — whether through a payer portal, fax, or electronic transaction — it enters an automated triage system before a human ever sees it.

The process looks like this:

• Step 1: Intake and parsing. The submission is received and parsed. AI extracts the diagnosis codes, procedure codes, medication requested, prescribing physician NPI, and patient demographics. Structured fields are checked for completeness. Missing or malformed data triggers an immediate administrative denial — often before clinical review begins.
• Step 2: Criteria matching. The parsed submission is compared against the payer's Clinical Policy Bulletin for the specific medication or procedure. The AI checks for required elements: documented diagnosis, prior treatment history, step therapy completion, specialist attestation, supporting lab values. Each missing element is flagged.
• Step 3: Risk scoring. Many payer AI systems assign a risk score to each submission — a probability that the claim represents appropriate utilization. High-risk scores trigger additional review requirements or automatic denial. Low-risk scores may result in immediate approval.
• Step 4: Denial generation. If criteria gaps are identified, a denial letter is automatically generated. The language is standardized, intentionally vague, and designed to describe the gap in the least specific terms possible. "Medical necessity not established" covers dozens of different actual gaps without specifying which one applies.
• Step 5: The appeal window opens. The denial letter is issued. The clock starts. The payer's bet: your practice does not have the bandwidth to file a complete, criteria-specific appeal before the window closes.

The Scale of Payer AI Investment

The investment payers have made in denial automation is not incidental. It is strategic.

Payer	AI Denial Tool	Known Capabilities
UnitedHealthcare	nH Predict / Optum	Post-acute care authorization, inpatient days, outpatient procedures
Cigna / Evernorth	CARA	Prior auth criteria matching, auto-denial generation
Aetna / CVS	Various proprietary systems	Pharmacy PA, specialty drug authorization
Humana	CenterWell-integrated AI	Medicare Advantage prior auth automation

EviCore Healthcare — a subsidiary that manages prior authorization for multiple major payers including Cigna, Aetna, and others — employs over 400 physicians to review appeals and escalations. That number sounds significant until you consider that EviCore processes millions of prior auth requests annually. The initial denial decisions are made by AI, not physicians. The physicians exist to handle the small percentage of cases that escalate.

The math is deliberate. If 65% of denials are never appealed — and MGMA research shows they are not — the AI denial system generates revenue for the payer on every unappealed case.

What Payer AI Is Actually Looking For

Understanding what payer AI checks allows you to build submissions that pass those checks on first submission — dramatically reducing your denial rate before any appeal is necessary.

For biologic medications (rheumatology, dermatology, gastroenterology)

Payer AI systems for biologics are checking for:

• Step therapy documentation. Evidence that the patient tried and failed one or more conventional therapies before the biologic was requested. The specific medications required, the duration of trial, and the documented reason for discontinuation must all be present.
• Diagnosis specificity. The ICD-10 code submitted must match an approved indication for the requested medication. A code that is one level too broad or one level too specific can trigger denial.
• Disease severity scores. Many payer CPBs require documented severity scores — DAS28 for rheumatoid arthritis, PASI for psoriasis, Harvey-Bradshaw for Crohn's disease. Submissions without these scores fail the criteria check automatically.
• Prescriber specialty. Some payer CPBs require that the prescribing physician be a specialist in the relevant field. A primary care physician prescribing a biologic without specialist attestation may trigger denial regardless of clinical appropriateness.

For procedures and imaging

Payer AI for procedure authorization checks:

• Conservative treatment documentation. Most procedural prior auths require documented failure of conservative treatment — physical therapy, medication management, behavioral interventions — before an invasive procedure is approved.
• Functional limitation documentation. The submission must describe specific functional limitations resulting from the condition, not just the diagnosis itself.
• Recency of supporting documentation. Some payer systems flag supporting clinical notes that are more than 30, 60, or 90 days old as insufficient for current authorization.

The Denial Letter Is Designed to Be Unhelpful

This is not cynicism. It is documented regulatory reality.

Prior to 2024 CMS rules, payers were permitted to issue denial letters that described the denial reason in the broadest possible terms. "Medical necessity not established" was acceptable even when the actual gap was something as specific as a missing DAS28 score.

The 2024 CMS prior authorization rules — implemented under the Consolidated Appropriations Act and extended by CMS-0057-F — now require payers to provide specific denial reasons. In practice, the implementation of this requirement has been uneven. Many denial letters still use generic language that obscures the actual documentation gap.

How to Build Submissions That Beat Payer AI

The most effective strategy against payer AI denial systems is not to fight denials. It is to submit documentation that the AI cannot deny on first pass.

This requires knowing — in advance — what each payer's AI is checking for each specific medication or procedure.

The pre-submission checklist approach

Before any prior auth submission leaves your practice, verify:

• The ICD-10 code matches an approved indication in the payer's current CPB
• All required step therapy medications are documented with trial duration and reason for discontinuation
• Required disease severity scores are present and recent
• Prescriber specialty matches CPB requirements
• Supporting clinical notes are within the payer's accepted recency window
• All required attachments are included in the submission package

This process adds 10 to 15 minutes per submission. It reduces denials dramatically — and every prevented denial eliminates the 2 to 3 hours that would otherwise be spent on the appeal process.

Payer-specific language matters

Aetna CPBs use different terminology than United Coverage Determination Guidelines. A submission written in Aetna's language submitted to United may pass the clinical criteria check but fail because the specific phrases the AI is scanning for are absent.

Top-performing practices maintain payer-specific submission templates for their most common procedures and medications. The templates are updated whenever a payer updates their CPB. The billing coordinator selects the right template, fills in the patient-specific clinical details, and submits — rather than writing from scratch each time.

How to Fight Back When Denials Do Happen

Even well-constructed submissions get denied. Payer AI systems have error rates. CPB criteria change. Documentation that was sufficient six months ago may not satisfy an updated policy.

When a denial arrives, the response strategy depends on the denial type:

Administrative denials (wrong code, missing field, incorrect NPI)

These are the fastest to resolve and the most frustrating to receive. The submission was clinically appropriate — an administrative error caused the denial.

The response: correct the error and resubmit immediately. Do not file a clinical appeal for an administrative denial. Identify the specific error, fix it, and resubmit as a corrected claim within the payer's correction window.

Many administrative denials can be identified before they happen with a pre-submission validation check. A system that catches a mismatched ICD-10 code or missing modifier before the submission leaves the practice eliminates an entire category of preventable denials.

Clinical criteria denials (missing documentation element)

These require a targeted appeal that addresses the specific missing element — not a general clinical argument.

• Identify the specific criterion that was not satisfied
• Pull the clinical documentation that addresses it
• If the documentation does not exist in the record, work with the physician to create it — a supplemental note, a severity score assessment, a treatment history summary
• Write an appeal that cites the payer's CPB criteria by section and document number, then presents the evidence that satisfies each criterion

Step therapy denials (patient has not tried required prior medications)

These are the most common denial type for biologic medications and the most frustrating for physicians, because step therapy requirements often override clinical judgment.

Two response paths exist:

Path 1: Document the step therapy. If the patient did in fact try the required medications, document it. Obtain pharmacy records, prior medical records, or physician attestation. File an appeal demonstrating that step therapy was completed.

Path 2: Request a step therapy exception. Most payers have a step therapy exception process for patients with contraindications to required prior medications or documented clinical reasons why standard step therapy is inappropriate. File for an exception with specific clinical documentation of why the standard pathway is not appropriate for this patient.

The Peer-to-Peer Weapon Most Practices Underuse

When clinical appeals fail, peer-to-peer review is available — and most practices never request it.

Peer-to-peer review is a direct physician-to-physician phone call between your treating physician and the payer's medical director who reviewed the denial. It is not a formal legal process. It is a conversation. And it reverses denials in 30 to 60% of cases when properly requested.

Why peer-to-peer works

The payer's medical director is a physician. When your physician calls and walks them through the specific clinical scenario — the patient's history, the failed alternatives, the clinical reasoning behind the treatment choice — the conversation is between two people who understand medicine. The algorithmic criteria sheet is still there, but the human judgment of the medical director carries weight it does not carry in an automated review.

How to request it

When a clinical appeal is denied, call the payer's provider line and request a peer-to-peer review before the appeal window closes. Most payers are required to offer this. Many do not volunteer it.

Prepare your physician with:

• The specific denial reason and which CPB criterion was cited
• The clinical documentation that supports the treatment decision
• The patient's treatment history and documented failures of prior therapy
• Any relevant clinical literature supporting the treatment approach

The peer-to-peer call typically runs 15 to 30 minutes. The outcome — approval or continued denial — is usually communicated within 24 to 72 hours.

The Competitive Reality for Independent Practices

Large hospital systems and corporate medical groups have dedicated prior authorization teams. Some have contracted with third-party authorization management companies. A few have built or licensed AI tools to generate payer-compliant submissions automatically.

Independent practices are frequently running this process with one or two billing staff members who also handle claims, eligibility verification, patient billing, and a dozen other functions.

The gap is not clinical. Independent physicians are not providing worse care. The gap is administrative infrastructure — and it is being used as a lever to make independent practice financially unsustainable.

Understanding how payer AI works — what it checks, what it flags, how denial letters are constructed to obscure the actual reason — is the first step toward competing on equal terms.

The second step is building systems that generate payer-compliant submissions automatically, surface denials in real time, and construct criteria-specific appeals without requiring your billing team to research CPBs from scratch for every denial.

That infrastructure is no longer exclusive to hospital systems.