Medical Billing · Denials & AI
AI Medical Claim Denials in 2026: why claims die — and how to fight back
If it feels like more of your clean claims are bouncing back — and faster — you are not imagining it. Payers now run much of their first-pass adjudication through automated systems, and the initial denial rate has kept climbing. This guide, written for US practice owners, office managers and billers, explains what is actually happening behind the denials, why claims die, what the new AI laws do and do not change, and a practical playbook to recover dollars you have already earned.
Denials are not a rare accident anymore; they are a routine cost of doing business that most practices under-manage. The good news is that the same structure that makes automated denials feel overwhelming — standardized reason codes, predictable deadlines, repeatable root causes — is exactly what makes a disciplined recovery process work. Let's start with the numbers.
What's happening: automated adjudication, and denials went up
Two things are true at once in 2026. First, payers are leaning harder on algorithms and AI-assisted utilization review to screen claims and prior-authorization requests at scale. Second, the initial denial rate has risen for several years running. Kodiak Solutions, drawing on data from more than 2,100 hospitals and 300,000 physicians, reported that payers initially denied about 11.8% of claims in 2024 — up from 11.5% in 2023 and roughly 15.7% higher than in 2020 (Kodiak Solutions / BusinessWire).
The pattern underneath that number is shifting. Kodiak found that denials tied to prior authorization actually fell by about 7.7% in 2024, while denials citing medical necessity rose roughly 5% and requests for more information rose about 5.4% (Becker's Payer Issues). In other words, the fight is moving from "did you get approval?" toward "can you document that this was necessary?"
On the health-plan side, the volume is stark. Analyzing 2024 HealthCare.gov data, KFF found insurers denied about 19% of in-network claims on ACA marketplace plans — and that fewer than 1% of denied claims were appealed (KFF). That gap between what gets denied and what gets challenged is where practice revenue quietly disappears.
It is worth being fair about the technology. Automation is not inherently the villain — the same systems that flag a claim for denial can approve a clean one in seconds. The problem is asymmetry: automated denials arrive at high volume and speed, while appeals still take human time that a busy front office rarely has. When denials scale and appeals do not, the math favors the payer.
Why claims die
Most dead claims are not mysterious. They fall into a handful of repeatable buckets, and the remittance itself usually tells you which one you are looking at.
Reason codes (CARC/RARC). Every adjustment or denial on an electronic remittance carries standardized codes — Claim Adjustment Reason Codes and Remittance Advice Remark Codes. These are your triage map. Sorting denials by code separates a quick registration fix from a claim that needs clinical documentation, and separates true denials from underpayments. KFF's marketplace analysis found the most-cited denial reasons were "other" (about 36%) and administrative issues (about 25%), with only around 5% for medical necessity (KFF) — a reminder that a large share of denials are fixable process errors, not clinical disputes.
Timely filing. Every payer sets a window to submit the original claim and, separately, a window to appeal a denial. Both can be short. Miss either and an otherwise-payable claim becomes a permanent write-off with no recourse — the single most avoidable way a claim dies.
Medical necessity and documentation. As the Kodiak data shows, necessity denials are rising. These are winnable, but only with claim-specific clinical documentation attached to a payer-specific appeal, which is also the most expensive type of denial to rework.
Underpayments. The most overlooked category. The claim was "paid," so it never lands in a denial queue — but it was paid below the contracted rate. Without comparing remittances against your fee schedule, these losses are invisible and rarely recovered.
The cost of ignoring all this compounds. Widely cited MGMA benchmarks put the administrative cost to rework a single denied claim at roughly $25 for a straightforward case, climbing much higher for complex medical-necessity appeals (industry analysis of MGMA data). Yet analyses commonly find roughly two-thirds of denials are ultimately recoverable — and a large share are simply never reworked. The revenue exists; it just needs someone to chase it before the clock runs out.
What the new rules mean
2025 and 2026 brought real limits on AI-only denials, plus a federal push to speed prior authorization. These help, but none of them make denials disappear — so treat them as guardrails, not a cure.
California SB 1120 (the Physicians Make Decisions Act). Signed in September 2024 and effective January 1, 2025, it prohibits California health plans from using AI as the sole basis to deny, delay or modify care on medical-necessity grounds. The ultimate determination must be made by a licensed physician or qualified clinician competent in the relevant clinical issues (CA Senate). It constrains how denials are made in California; it does not stop them.
Illinois. Illinois HB 5395 (2024) amended its managed-care and utilization-review rules to address algorithmic and automated review and adverse determinations, with provisions phasing in on January 1, 2025 and January 1, 2026 (KFF, state AI tracker). Several other states — including Texas, Arizona and Maryland — have moved on AI in utilization review as well, but the specifics vary and are still settling, so confirm the current rule with each payer's provider manual rather than assuming a uniform standard.
CMS-0057-F (federal Interoperability and Prior Authorization final rule). Finalized in early 2024, it requires impacted payers — Medicare Advantage, Medicaid and CHIP programs, and federally facilitated marketplace plans — to streamline prior authorization, send specific denial reasons, and stand up FHIR-based APIs, with core provisions phasing in through 2026 and API requirements by January 1, 2027. CMS estimates roughly $15 billion in savings over ten years, mostly from reduced prior-authorization friction (CMS). For practices, the near-term takeaway is faster decisions and clearer denial reasons from affected payers — not the end of denials.
The honest read: these rules improve transparency and force a human into the loop for certain medical-necessity denials, but enforcement, scope and timing differ by payer and state. Your recovery process still has to do the work.
What to do: a practical playbook
You cannot control the payer's algorithm, but you can control three things: knowing which denials are recoverable, appealing them before the deadline, and stopping the ones that repeat. Run these in order.
Audit denials by reason
Pull your remittances and sort every denial and underpayment by CARC/RARC code and by payer. Rank by recoverable dollars and by deadline urgency so the biggest, most time-sensitive claims surface first.
Appeal the recoverable ones — before deadlines
Work the ranked list against each payer's appeal window. Send corrected claims for administrative errors and documented, payer-specific appeals for necessity denials. Track every deadline so nothing ages out.
Fix the root causes
Group recurring codes back to their source — eligibility, coding, missing authorization, documentation — and fix the front-end step so the same denial stops recurring. Prevention beats rework every time.
A few principles make this stick. Prioritize by dollars and deadlines, not by whichever claim is on top of the pile. Treat underpayments as denials — compare every payment to your contracted rate. And measure your denial rate by payer and by reason over time, so you can see whether prevention is actually working. If a denial in California (or another state with AI limits) appears to rest on an automated medical-necessity decision without clinician review, that is worth flagging in the appeal.
Where ClaimRecover fits
If the playbook above sounds right but you do not have the staff hours to run it every week, that is the gap ClaimRecover is built for. It is a hosted web app: you upload your remittance and claim files, and it parses X12 835/837 and practice-management CSV exports, classifies denials by CARC/RARC code, ranks recoverable dollars and deadline urgency, drafts payer-specific appeal letters and corrected-claim diffs, and runs a pre-submission scrub to prevent repeats. It does the audit-and-prioritize work so your team spends its time on the appeals most likely to pay — with the same honest caveat that it surfaces and pursues recoverable dollars rather than promising payment. For the deeper walkthrough, see our practical guide to recovering dead medical bills.
Frequently asked questions
Are insurers really using AI to deny claims?
Yes, in the sense that payers increasingly use algorithms and automated utilization-review tools to screen and auto-adjudicate claims at high speed. That is not automatically improper — automation can approve clean claims faster too. The concern regulators have addressed is AI being the sole basis for a medical-necessity denial without a qualified clinician's review, which California's SB 1120 and Illinois' 2024 changes now restrict.
Does the new California law mean my denials will stop?
No. SB 1120 (effective January 1, 2025) prohibits California health plans from using AI as the sole basis to deny, delay or modify care on medical-necessity grounds; a licensed clinician must make that call. It limits how denials are made in California — it does not eliminate them, and it does not apply to every payer or every state. You still need to audit and appeal.
How many denied claims are actually recoverable?
Industry analyses commonly cite that roughly two-thirds of denials are ultimately recoverable, yet a large share are never reworked because appeals take staff time. On ACA marketplace plans, KFF found fewer than 1% of denied in-network claims were appealed in 2024. No specific claim is guaranteed to be paid — but leaving denials unworked guarantees they will not be.
What is the deadline to appeal a denied claim?
It varies by payer and plan, and can be as short as 30 to 90 days from the remittance date for a first-level appeal, with separate, shorter timely-filing windows for corrected claims. Check the payer's provider manual and the date on the ERA/EOB. Missing the window usually turns a recoverable denial into a permanent write-off.
What is a CARC/RARC code and why does it matter?
CARC (Claim Adjustment Reason Codes) and RARC (Remittance Advice Remark Codes) are the standardized codes payers put on the remittance to explain an adjustment or denial. Sorting denials by these codes tells you which are quick administrative fixes, which need clinical documentation, and which are underpayments versus true denials — so you can chase the recoverable dollars first.
See what your practice is still owed
ClaimRecover audits the claims and remittances you upload, ranks recoverable dollars, and drafts appeals before deadlines close — then helps you prevent the next round of denials.
Educational content, not legal, medical, or billing-compliance advice. No tool can guarantee payment of a denied claim.