Prior Authorizations & Insurance Appeal Letters

Prior authorizations and appeals are the administrative tax pharmacists, techs, and prescribers pay every day. A typical retail pharmacist touches dozens of PA workflows a week; specialty pharmacists may touch hundreds. AI cuts the drafting time on each one by 60–80% and, critically, writes cleaner letters that insurance reviewers are more likely to approve.

What You'll Learn

• How to turn a denial reason code into a ready-to-submit appeal in 10 minutes
• A reusable letter-of-medical-necessity template you can adapt per drug
• How to strip PHI safely before using AI for PA documents
• How to track outcomes so you can improve your PA success rate

Why AI Wins Here

PAs and appeals are structured documents. They have predictable sections: patient clinical history, failed therapies with dates, evidence citations, formulary exception rationale, and a closing request. Structured tasks are exactly where large language models excel.

The win is even larger because PBM appeals are often first-read by AI. Optum, CVS Caremark, Express Scripts, and most other PBMs use AI or structured-data extraction to screen appeals before a human reviews them. An appeal that is cleanly structured, with the right keywords and the right clinical evidence, ranks higher in that triage.

De-Identify First, Always

Before any PA workflow in a consumer AI tool:

• Replace the patient's name with "Patient A" or "the member"
• Remove the date of birth; use "75-year-old male"
• Remove the MRN, address, phone, and any insurance ID
• Remove the pharmacy name if it identifies location in a way that combined with age could identify the patient

Or use a HIPAA-compliant enterprise version your employer has licensed (ChatGPT Enterprise, Claude for Work, Copilot with BAA).

The Letter-of-Medical-Necessity Template

Paste this prompt into Claude or ChatGPT, then fill in the brackets:

"Act as a pharmacist drafting a letter of medical necessity for a PBM prior authorization. Structure the letter with these sections: (1) Member information placeholder, (2) Requested medication and dose, (3) Diagnosis with ICD-10 code, (4) Clinical rationale, (5) Prior therapies tried and failed — with reason for failure, dates, and duration, (6) Evidence-based justification referencing the relevant guideline or FDA label, (7) Request for formulary exception or quantity override, (8) Closing with physician signature block.

Drug: [DRUG, DOSE, SIG] Diagnosis: [ICD-10] Prior therapies: [LIST with fail reason and date] Relevant guideline: [e.g., ADA 2025 for insulin analogues] Tone: formal, professional, concise — 1 page max."

Output is a clean, submission-ready letter. Forward to the prescriber for signature, submit to the PBM, log the interaction.

Responding to a Denial Reason Code

PBM denials come back with a short reason: "Step therapy not met," "Quantity limit exceeded," "Non-formulary — try preferred alternative," "Age limit not met." Each has a different rhetorical approach.

A useful prompt pattern:

"The PBM denied a PA for [DRUG] with the denial reason 'step therapy not met — patient must try [PREFERRED ALTERNATIVE] first.' However, the patient has contraindications to the preferred alternative: [LIST]. Draft an appeal letter that: (1) acknowledges the PBM criteria, (2) documents the contraindications with clinical rationale, (3) cites the guideline or label that supports skipping step therapy, (4) requests a medical-necessity override. Formal tone."

The resulting letter is tailored to the actual denial reason rather than a generic appeal — which is exactly the letter that gets approved on the first try.

Faster Forms: Drug-Specific Questions

Most PA forms ask the same core questions: failed therapies, duration of failure, side effects, labs, contraindications. AI can pre-populate answers for you if you paste the form and the de-identified patient summary.

"Here is a PA form for ozempic [paste form text]. Here is the patient summary: 62-year-old with type 2 diabetes, A1c 9.1, BMI 34, tried metformin 1000 mg BID for 6 months with A1c improvement from 10.2 to 9.3 but GI intolerance at max dose, tried sitagliptin 100 mg daily for 4 months with no significant A1c improvement. Also has CKD stage 2. Fill in each PA form question with the correct answer from the patient summary. If a field has no matching data, flag it."

You get a ready-to-type answer sheet. The tech enters it into CoverMyMeds or SureScripts; the pharmacist verifies.

Tracking PA Outcomes to Improve

Log every PA in a simple spreadsheet:

• Drug, diagnosis, PBM, denial reason, appeal drafted with AI (yes/no), outcome, days to approval.

Every month, ask Gemini (with your spreadsheet open) or ChatGPT (with the data pasted, de-identified):

"Analyze my PA outcomes for the last 90 days. Which PBM has the highest denial rate? Which drugs are most frequently denied? Which denial reasons produced the highest appeal-win rate? Suggest three workflow changes to improve approvals."

You now have a data-driven operational improvement plan. This is how a pharmacy of two pharmacists and four techs starts to operate with the efficiency of a team twice its size.

A Specialty Pharmacy Example

A specialty pharmacist is working on a PA for a brand-name biologic (ustekinumab) for ulcerative colitis. The PBM requires documentation of failed anti-TNF therapy.

1. Claude: "Act as a specialty pharmacist. Draft a letter of medical necessity for ustekinumab for a patient with moderate-to-severe UC who failed infliximab (primary non-response after 14 weeks) and then adalimumab (secondary loss of response at month 9 confirmed by endoscopy). Reference the ACG 2024 guideline positioning ustekinumab as a reasonable next-line agent after anti-TNF failure."
2. Review the draft, verify the guideline citation against the actual ACG guideline, edit the clinical narrative for accuracy.
3. Send to the prescriber for signature; submit to the PBM.
4. Log the PA in the tracking sheet.

Two hours of work compressed into forty minutes — and the letter is often better than what the prescriber's office would produce unassisted.

What Not to Do

• Do not let AI invent dates, lab values, or failure criteria to strengthen the letter. Fabricating clinical history in an insurance document is fraud. Every clinical claim must match the chart.
• Do not paste PHI into consumer AI. De-identify or use a BAA-signed enterprise tool.
• Do not submit an AI-drafted letter without a pharmacist-level review of every clinical fact.
• Do not sign the prescriber's name — let AI draft the closing block, but the prescriber signs.

Key Takeaways

• PA letters and appeals are structured documents where AI cuts drafting time by 60–80%
• Always de-identify before pasting into a consumer AI tool; use a BAA-signed enterprise tool for PHI
• Match the appeal rhetoric to the denial reason — step-therapy denials, quantity limits, and non-formulary denials need different arguments
• Track PA outcomes in a spreadsheet and use AI to analyze the data monthly for workflow improvements
• Never fabricate clinical history; every claim must match the chart