Blog
AI in claims review: what can an algorithm decide about my care?
How automated screening differs from clinician review in claims and prior authorization, the rules as of November 2026, and written questions to ask your plan.
Most of what happens to a health claim happens without a person reading it. A claim arrives electronically, software checks it against rules, and the large majority of routine claims are paid or denied by the time a human could have looked. That has been true for decades. What is new is the increasing use of statistical models and machine-learning tools alongside those rules, and the questions patients are asking about what those tools are allowed to decide.
This explainer lays out the layers where automation can show up, the difference between a tool that helps a person decide and one that decides, what rules govern it as of November 2026, and the written questions that put the plan on record. It does not evaluate any company's systems. News reports and lawsuits about particular insurers contain allegations and pending matters, not settled findings, and this page does not rely on them.
Where automation sits in the life of a claim
It helps to separate five stages. A tool at one stage says little about the others.
- Eligibility and intake. Software checks whether you were covered on the date of service and whether the claim form is complete. This is routine and rarely controversial.
- Claim edits. Rules compare the billing codes on the claim against each other and against payment policy. They catch duplicates, services billed together that the plan treats as one, mismatches between a diagnosis and a procedure, and limits such as visits per year. This is the stage behind many denials with a "remark code" and a short reason. It is a rules engine: if X, then deny or pend. A person wrote the rule, and a different person can say why it fired. claims adjudication describes the process.
- Prior authorization intake. When a doctor requests approval (prior authorization), software may match the request against the plan's criteria. If the request meets them, many plans approve it automatically. If it does not, it is sent to a person.
- Clinical review. A nurse or physician reviewer reads the records and applies the plan's criteria (medical necessity). Whether a software tool scored the case first, and what weight the score carried, is exactly what patients and regulators are asking about.
- Appeals. An appeal is typically reviewed by someone who was not involved in the first decision, and by an independent reviewer if it goes to external review.
Automation is common in the first three stages and contested in the fourth. Notice that the three kinds of decision are different in kind:
| Kind of decision | Typical role of software | Why it matters to you |
|---|---|---|
| Approve | Often automatic when criteria are met | Faster answer; rarely disputed |
| Deny for a rule (duplicate, limit, wrong code) | Often automatic | A correction by the provider may fix it |
| Deny because care is not medically necessary | Software may score or recommend; rules and plan policies vary on who may decide | This is the one that most needs a clinician's judgment about you |
Screening is not deciding
Two sentences can describe the same system and mean very different things. "The tool flags requests for review" means software sorts cases, and a person reads the flagged ones. "The tool denies requests" means software issues the denial with no clinician involved. Everything in between matters: a reviewer who sees a recommendation and approves it in seconds is technically a human decision and may practically be the tool's.
Common uses of software in this area include:
- Auto-approval of routine requests, which supporters say speeds up care and clears queues.
- Triage, which places complex requests in front of specialists.
- Predicting typical recovery or length of stay after surgery, a stroke or a hospitalization, used to estimate when coverage for rehab or home care should end. Critics argue that an average does not describe an individual patient.
- Pattern detection for suspected billing errors or fraud.
- Drafting letters and summaries for the plan's staff.
Supporters generally say these tools reduce administrative cost, cut delays and catch errors, and that clinicians remain responsible for denials. Critics, including some patient and clinician groups, say opaque tools can apply uniform criteria to individuals, that few denied patients appeal, and that review time per case can be too short for real judgment. Both descriptions are about how the tool is used, which is why the questions below ask about process rather than technology.
What the rules say, and what they do not, as of November 2026
This area changes quickly, and the sources checked for this site on November 16, 2026 did not turn up a verified federal rule specific to artificial intelligence in claims. What the sources do record is a general rule on prior authorization decisions:
- The federal Interoperability and Prior Authorization final rule (CMS-0057-F) set requirements that start January 1, 2026 for Medicare Advantage plans, state Medicaid and CHIP fee-for-service programs, Medicaid and CHIP managed care plans and issuers of qualified health plans on the federal exchanges. Those plans must give a specific reason when they deny a prior authorization, and, except for the federal-exchange issuers, decide within 72 hours for expedited requests and 7 calendar days for standard ones. Drug prior authorization is excluded.
- Electronic interfaces (APIs) for prior authorization and other data exchange are due by January 1, 2027.
- It does not say who or what may make the decision. It speaks to speed, reasons and data exchange.
Beyond that, requirements about whether a licensed clinician must review a medical-necessity denial, and whether a plan must disclose its use of automated tools, come mostly from state law, plan contracts and program rules, and they vary. Some states have enacted or proposed laws on human review or disclosure, and the details and effective dates differ. They also generally apply to plans the state regulates; a self-funded employer plan under federal benefits law is typically outside state insurance rules. Your state insurance department (how to file a complaint with your state insurance department) is the authority on what applies to yours, and Medicare Advantage and Medicaid plans have their own program rules that the plan's notice and CMS or the state Medicaid agency can explain.
Two practical consequences follow:
- Do not assume a rule exists, or that it does not. Ask.
- The strongest protections are the old ones: a written reason for the denial, the right to appeal to someone who was not involved, the right to your claim file, and the right to external review. All of those apply however the first decision was made. how prior authorization works and what to do when it is denied and how to appeal a denied health insurance claim, step by step walk through them.
What to check for updates: your state insurance department's website for any new law on automated decisions, the plan's member handbook or notices for changes, and the Centers for Medicare & Medicaid Services' Medicare Advantage and Medicaid pages for rule updates. Do not rely on a news summary for the details of a law; read the agency's own page.
Questions to put to the plan in writing
If you have a denial and wonder what part software played, ask, and ask in writing so that the answer is a record. A plan may answer some questions and decline others; a refusal or a vague reply is itself information. Questions that are reasonable to ask:
- Who made this decision? Ask for the name or the title and credentials of the person who made the denial, and whether the reviewer was a physician, a nurse or a non-clinical staff member.
- Was a clinician involved in this denial, and did the clinician review my medical records? Ask for the date the review took place.
- Which criteria or guideline did the decision use, with its name, version and the specific section? Ask for a copy. Plans generally must share the criteria they relied on when you appeal.
- Did an automated tool, scoring system or algorithm contribute to this decision or to the estimate of how long care should last? If yes, what was its output and how was it used?
- What documents were considered? Ask for a list of the records the reviewer had. If your doctor's recent notes are missing, send them.
- What exactly in my case did not meet the criteria? A specific reason, not a template phrase, is what you need to answer.
- How can I get a review by a physician in the same specialty? For some appeals, you can ask that a reviewer with relevant expertise decide it.
- What is my deadline to appeal, and is an expedited appeal available? Put this next to your calendar; the prior authorization and claim tracker can hold it.
- May I have my complete claim file, including internal notes, free of charge?
A short script for the first message might read: "I am writing about the denial dated the date on the notice for claim or reference number. Please tell me who made the decision and their credentials, whether any automated tool was used, the criteria applied and the records reviewed, and send me my claim file. I am also appealing; please confirm the deadline." how to ask your insurer for your full claim file and the criteria it used goes deeper on the file request.
What to do with the answers
- If a clinician reviewed your records and gave a specific reason, you have a clear target. Your doctor's letter should answer the stated reason, point by point, and cite clinical evidence. Ask for a peer-to-peer call (peer-to-peer review) with the plan's physician, if available.
- If the answer is vague or contradictory, include that in your appeal. Saying "the plan could not identify who reviewed my case" is a fact the reviewer on appeal can weigh, and a regulator can act on.
- If the plan says a tool was used, ask how a person assessed your specific situation. You do not need to argue about the technology; you need the decision to rest on your records and your doctor's judgment.
- If time is short, appeal first and ask the questions in parallel. Do not wait for the answers before filing.
- If you are not satisfied, a complaint to the regulator for your plan type can ask the same questions with an authority behind them.
new prior authorization rules: what they mean for my next request explains the faster-decision rule in question-and-answer form, which pairs with this post.
What this does not tell you
No page can say whether any specific denial was made by a person, a program or both. It also cannot tell you that automation is better or worse than manual review: automated approvals speed care, and the same technology used badly can delay it. The reasonable stance for a patient is practical. Assume your case should be judged on your records. Ask who judged it. Keep the paper trail. And remember that every appeal right you have applies regardless of how the first decision was reached.
Key takeaways
- Software already touches most claims, mainly at intake, in rule-based edits and in routine prior authorization approvals; the contested use is in medical-necessity denials.
- As of November 2026, the federal prior authorization rule covers reasons, speed and data exchange for certain plans; rules on who must review a medical-necessity denial come largely from state law, program rules and plan contracts and vary.
- Ask in writing who decided, with what credentials, using which criteria and records, and whether an automated tool contributed.
- Appeal on the clinical record regardless; your rights to a written reason, your claim file, an independent appeal reviewer and external review do not depend on how the first decision was made.
- Check your state insurance department and the plan's notices for updates, because this area is changing.