Was This Denial Automated? A Human-Review Request Letter
A letter that asks a payer to confirm whether an algorithm produced a behavioral-health denial, who reviewed it, and when — with the paragraphs that make the question answerable.
On this page: Direct answer
Direct answer
AI insurance denial request human review: what operators need to know
A letter that asks a payer to confirm whether an algorithm produced a behavioral-health denial, who reviewed it, and when — with the paragraphs that make the question answerable. Ask what was used, who reviewed it, and when — three separate questions. Vague questions get vague answers.
Denials that arrive within minutes, in batches, with identical wording across unrelated patients, raise an obvious question that almost nobody asks in writing: did a person actually review this? The question matters because the answer changes the appeal. A determination involving medical judgement carries requirements about who makes it and what they must disclose, and a determination produced by a tool a clinician never meaningfully examined is vulnerable on procedure rather than on clinical merit.
This guide is deliberately narrow. It does not claim your payer uses automation, and it does not tell you that automated review is unlawful — in many contexts it is not. It gives you a letter that asks the question precisely enough to require an answer, and explains what the answer is worth in each plan type.
Key takeaways
The short version
- Ask what was used, who reviewed it, and when — three separate questions. Vague questions get vague answers.
- For ERISA plans the algorithmic output is reachable as a document generated in the course of the determination under 29 CFR 2560.503-1(m)(8).
- CMS-0057-F requires impacted payers to give a specific denial reason, which makes boilerplate wording a documentable problem in itself.
- Several states enacted 2026 requirements around automated denials and clinician review; check your state rather than assuming a federal rule.
- The timestamp comparison is the strongest single fact: when the output was produced against when a clinician signed off.
- Patterns beat instances. One fast denial proves nothing; forty with identical wording is a compliance conversation.
Take the template with you
Free to copy · no email required
Automated-determination disclosure request
A factual letter asking what tool was involved, who reviewed the determination with what credentials, and the timestamp of each step, with plan-type citation blocks to choose from.
Saves as automated-denial-disclosure-request.txt
Determination timing and reason-text log
Observable facts only — submission and decision timestamps, elapsed time, verbatim reason text, reviewer, and whether a specific criterion was named. No clinical detail, so the log stays safe to analyse.
request_ref,payer,plan_type,level_of_care_requested,submitted_at,determination_at,elapsed_hours,outcome,reason_text_verbatim,specific_criterion_named,criteria_set_and_edition,reviewer_name_stated,reviewer_credentials_stated,peer_to_peer_offered_before_decision,disclosure_request_sent,disclosure_response_received,automation_disclosed,notes REQ-0001,[payer],ERISA group / MA / Medicaid MC / FFE QHP,PHP,2026-10-06T09:12,2026-10-06T09:31,0.3,denied,[paste the reason sentence exactly as written],no,none stated,yes,no,no,2026-10-07,,,identical wording to REQ-0004 and REQ-0011
Saves as determination-timing-log.csv
1. What the question is actually for
There is no general federal prohibition on payers using software in utilization management, and plenty of legitimate automation exists — eligibility checks, completeness screening, routing, flagging. The issue is narrower: whether a determination that required medical judgement received it, and whether the basis disclosed to the claimant reflects the basis actually used.
That is why the letter asks three distinct things. What tool or model was involved. Who the human reviewer was, with credentials. And when each happened. A plan can answer the first two in a way that sounds reassuring while the timestamps show a clinician attached their name after the decision had already been generated and communicated.
- Automation that screens for completeness and routes work is ordinary and not the target of this request.
- Automation that produces a medical-necessity conclusion is the target, because that conclusion requires judgement.
- A reviewer's name on a notice is not evidence of review; the sequence of timestamps is closer to evidence.
- Identical denial language across clinically unrelated patients is the pattern worth documenting.
- Decisions issued faster than a record could plausibly be read are worth noting with the actual elapsed time.
- None of this replaces the clinical appeal. It runs alongside it.
2. What gives the question footing, by plan type
The basis for asking differs by plan type, and using the wrong one weakens the request. For employer-sponsored coverage the claims procedure regulation is the lever, because its definition of relevant documents reaches material generated while deciding. For Medicare Advantage, Medicaid managed care, CHIP, and federally facilitated exchange plans, CMS-0057-F added a duty to give a specific reason for denial, which bears directly on boilerplate.
| Plan type | The lever | What it supports asking for |
|---|---|---|
| ERISA group health | 29 CFR 2560.503-1(h)(2)(iii) and (m)(8) | The algorithmic output itself, internal rules relied on, reviewer identity, and notes generated while deciding |
| ERISA group health | 29 CFR 2560.503-1 full and fair review standard | An argument that a basis the claimant cannot examine cannot be meaningfully contested |
| Medicare Advantage | CMS-0057-F specific-denial-reason duty; 42 CFR part 422 determination rules | A reason specific enough to identify the criterion, rather than a template sentence |
| Medicaid managed care | 42 CFR 438.210(c) notice requirements | Notice content sufficient to support an appeal |
| Marketplace QHP on a federal exchange | 45 CFR 156.223 and CMS-0057-F | The same specific-reason expectation as other impacted payers |
| Any plan, where state law applies | State utilization-review and 2026 automated-decision statutes | Clinician review of medical-necessity denials, where the state requires it |
3. The three questions, and what answers look like
Record every answer, including the non-answers, in the same structured place you log the denial itself. A refusal to say whether a tool was used is itself a fact, and it reads very differently in aggregate than it does in one case.
| Question | A weak answer | A useful answer |
|---|---|---|
| What was used? | "Our determinations comply with applicable law." | The named tool or model, its role, and whether its output expressed a medical-necessity conclusion |
| Who reviewed it? | A name with no credentials | Name, licensure, specialty, and whether that person was the decision-maker or a signatory |
| When did each step occur? | The determination date only | Timestamps for request receipt, record review, any automated output, clinician review, and determination |
| What record did the reviewer see? | "All submitted documentation." | An inventory of what was in the file at decision time |
| Was an internal rule used? | Silence | A yes or no, with the rule provided if yes |
| How was the specific reason derived? | The template sentence restated | The criterion, the finding against it, and the evidence weighed |
4. Build the pattern, not the anecdote
A single fast denial is not evidence of anything. Forty denials across unrelated patients with byte-identical reason text, median decision time measured in minutes, and the same reviewer name on all of them is a different object entirely — it is the kind of thing a state regulator, a plan sponsor, or a contracting conversation can act on.
The log below is deliberately small. It records only what you can observe from the notice and your own submission timestamps, which means it stays accurate and stays free of clinical detail.
- Capture submission timestamp and determination timestamp for every request, then compute elapsed time.
- Store the verbatim reason text so identical wording is detectable by comparison rather than memory.
- Record the reviewer name and whether credentials were stated.
- Note whether a peer-to-peer was offered before the determination.
- Flag whether the notice named a specific criterion or used a template sentence.
- Review the log monthly by payer. The outlier payer is usually obvious within one quarter.
5. Where this sits in the appeal sequence
- 01
Do not let it delay the appeal
Calendar and file the clinical appeal on its own deadline. This request runs in parallel and never replaces it.
- 02
Combine it with the claim file request
For ERISA plans, the automation questions belong inside the claim file request rather than in a separate letter. One document, one response obligation.
- 03
Use the answer to choose the lead argument
If timestamps show no meaningful clinical review, lead the appeal on procedure. If a person reviewed a complete record, the clinical argument is what you have.
- 04
Escalate patterns, not cases
Take aggregate findings to the plan sponsor, the regulator, or contracting. Individual cases stay in the appeal process.
- 05
Keep your own house in order
If you use software in your appeal or documentation workflow, hold it to the standard you are asking of the payer: a named human approves anything that goes out.
Common questions
Answers before you build.
Is it legal for a payer to use AI to deny claims?+
There is no single answer, because it depends on the plan type, the state, and what the tool actually did. Using software to screen, route, or flag is ordinary. Using it to reach a medical-necessity conclusion without meaningful clinician review is where federal procedural requirements and a growing set of state rules bite. This guide helps you establish which of those happened rather than assert a conclusion.
Will a payer actually answer these questions?+
Sometimes fully, sometimes partially, sometimes not at all. All three outcomes are useful. For ERISA plans the document entitlement gives the request real weight; elsewhere the value is in creating a documented record of what was asked and what came back.
Should we send this for every denial?+
No. Reserve it for denials where the pattern suggests it: implausibly fast decisions, template reasoning, identical language across unrelated patients, or a notice with no named criterion. Sending it indiscriminately dilutes the signal and slows your own desk.
Does this apply to Medicaid and Medicare Advantage?+
The questions do, but the footing differs. Use the CMS-0057-F specific-denial-reason duty and the relevant part 422 or part 438 notice rules rather than the ERISA citations, and expect state Medicaid contract terms to matter as well.
Can we mention state AI laws in the letter?+
Only after confirming the current text applies to that plan and that line of business, with counsel. Several states acted in 2026 and the provisions differ materially; a miscited statute undermines an otherwise strong letter, and state insurance law generally does not reach self-funded ERISA plans.
Practical closeout
Use this operator checklist.
- Ask what was used, who reviewed it, and when — three separate questions. Vague questions get vague answers.
- For ERISA plans the algorithmic output is reachable as a document generated in the course of the determination under 29 CFR 2560.503-1(m)(8).
- CMS-0057-F requires impacted payers to give a specific denial reason, which makes boilerplate wording a documentable problem in itself.
- Several states enacted 2026 requirements around automated denials and clinician review; check your state rather than assuming a federal rule.
- The timestamp comparison is the strongest single fact: when the output was produced against when a clinician signed off.
- Patterns beat instances. One fast denial proves nothing; forty with identical wording is a compliance conversation.
Continue through the cluster
Verified customer case studies are added only with customer permission and supporting evidence; none is implied by these operational examples.
Sources & methodology
Trace the operational claims.
Marsa Health Editorial reviewed the primary and research sources below on October 18, 2026. We translate them into workflow controls, distinguish proposals from final rules, and flag where plan, program, state, contract, or clinical requirements vary.
- 0129 CFR 2560.503-1: Claims procedure Electronic Code of Federal RegulationsParagraph (h)(3)(i) gives group health claimants 180 days to appeal; (h)(2)(iii) requires reasonable access to and free copies of all documents relevant to the claim; (m)(8) defines what counts as relevant.Accessed or rechecked October 18, 2026
- 02CMS Interoperability and Prior Authorization Final Rule CMS-0057-F Centers for Medicare & Medicaid ServicesThe fact sheet listing which payers are impacted, the four required APIs, the denial-reason duty, and the compliance dates.Accessed or rechecked October 18, 2026
- 0342 CFR 438.210: Coverage and authorization of services Electronic Code of Federal RegulationsParagraph (d) sets 7-calendar-day standard and 72-hour expedited decision limits for rating periods beginning on or after January 1, 2026; (c) governs notice of adverse benefit determination.Accessed or rechecked October 18, 2026
- 0445 CFR 156.223: Prior authorization requirements Electronic Code of Federal RegulationsApplies the prior authorization API and public-metrics requirements to qualified health plan issuers on the federally facilitated exchanges.Accessed or rechecked October 18, 2026
- 05Appealing health plan decisions Centers for Medicare & Medicaid ServicesCMS guidance on internal claims and appeals rights, notice contents, and the relationship between internal appeals and external review.Accessed or rechecked October 18, 2026
- 06External Appeals Centers for Medicare & Medicaid ServicesFederal external-review process and consumer protections.Accessed or rechecked October 18, 2026
Organizational author. Editorial review covers source accuracy, search intent, workflow boundaries, and human-oversight requirements. This material is educational and does not provide clinical, legal, coding, or coverage advice.
No named clinical or legal expert reviewer is attributed to this version. Marsa Health does not invent reviewer credentials.
Read our editorial methodRevision history
What changed and when
October 18, 2026
Initial publication, source review, and operational editing.