Editor’s note: This piece was reviewed by Dr. Heather Signorelli, DO, as physician-reviewed operational guidance. It is not medical, legal, or compliance advice, and NatRevMD does not endorse any specific AI vendor. Verify any workflow against your own HIPAA and payer obligations.
Most of the AI advice aimed at doctors is written by people who have never worked a denial. It is all “reimagine your practice” and zero “here is the exact wording to paste into a box.” We run a billing company, so we live in the box. And the honest truth is that a well-built prompt library does more for a small practice than most of the shiny AI features being sold to you, because it turns your slowest, most repetitive administrative tasks into something a staff member can finish before their coffee gets cold.
So we are going to hand you the prompts. Below is a representative set of the templates we actually use, written out and ready to adapt. The complete set of 22 lives in our AI Kit, but these will get a front desk and a biller moving today. One rule before you start: none of these prompts contain patient identifiers, and neither should yours. We will come back to that.
The one habit that makes every prompt work
Before the templates, the single most important thing. De-identify first, then prompt. Strip names, dates of birth, medical record numbers, member IDs, full dates of service, and addresses before anything touches a consumer AI tool. Replace them with neutral placeholders like “the patient,” “Payer A,” or “CPT 99214.” You lose nothing operationally, because the AI does not need to know who the patient is to draft an appeal or explain a denial code. It needs the structure of the problem.
If you are using a consumer tier of any AI tool without a signed Business Associate Agreement, treat it like a public whiteboard. The prompts below are built to be safe on that assumption. Enterprise or BAA-covered tiers change the calculus, but you have to verify the current terms yourself rather than trust a marketing page.
Denials and appeals
This is where a prompt library pays for itself fastest. Denials are structured problems, and AI is good at structured problems.
1. Decode the denial. Explain what CARC [code] and RARC [code] mean in plain language for a medical biller. List the three most common root causes and the first action to take for each.”
2. Draft an appeal skeleton. “Draft a professional appeal letter for a claim denied for [reason, no PHI]. The service was [CPT], medically appropriate for a patient with [condition, generic]. Leave bracketed placeholders for patient identifiers, dates, and clinical detail that I will fill in manually.”
3. Find the strongest argument. “Given a denial for [reason], list the two or three most persuasive angles for an appeal, ranked, and note what documentation each one requires.”
4. Rewrite for a specific payer tone. “Rewrite this appeal to be more concise and cite the specific policy category it falls under. Keep it under one page.”
5. Triage a denial batch. “Here is a de-identified list of denial reason codes and counts. Group them by likely root cause and tell me which category to work first for fastest cash recovery.”
Eligibility and front desk
The front desk is where revenue is won or lost before a claim is ever filed. These prompts cut the time spent on the tasks that make staff want to quit.
6. Explain a plan in human terms. “In plain language a front-desk staffer can read to a patient, explain the difference between a deductible, copay, coinsurance, and out-of-pocket maximum. Keep it to five sentences.”
7. Build a benefits check script. “Write a phone script for verifying a patient’s benefits with a payer. Include the exact questions to ask about deductible status, prior authorization requirements, and referral rules. No patient data.”
8. Turn a policy into a checklist. “Convert this payer prior-authorization policy [paste public policy text] into a step-by-step checklist my staff can follow, with the required documentation at each step.”
9. Draft the patient cost conversation. “Write a short, warm script for telling a patient they have an outstanding balance and offering a payment plan. Non-judgmental, no medical detail.”
Coding and documentation support
Careful here. AI is a drafting and second-set-of-eyes tool, not a coder of record. A credentialed human signs off on every code.
10. Sanity-check a code selection. “For a de-identified visit described as [generic clinical scenario], list the E/M level considerations and what documentation would support each. Do not assign a final code.”
11. Explain a coding guideline. “Explain the documentation requirements for [service type] in plain language, and give me a three-item checklist to confirm before we bill it.”
12. Draft a documentation query. “Write a neutral, non-leading query to a provider asking them to clarify documentation for a visit where [generic ambiguity]. Do not suggest a specific diagnosis or code.”
Payer research and staff enablement
13. Summarize a payer bulletin. “Summarize this payer policy update [paste public text] into what changes for our billing team, effective when, and one action item.”
14. Compare two policies. “Here are two de-identified payer policies for [service]. Build a side-by-side table of prior-auth rules, documentation requirements, and timely-filing limits.”
15. Write a training one-pager. “Create a one-page reference for a new biller on how to read an EOB, what to check first, and the three most common mistakes to avoid.”
16. Draft an SOP. “Turn this rough process description [paste your notes] into a clean standard operating procedure with numbered steps and a short quality-check at the end.”
Patient communication and collections
17. Rewrite a scary letter. “Rewrite this collections letter to be firm but human. Remove jargon. Keep it compliant in tone and offer a way to resolve the balance. No patient identifiers.”
18. Answer a common billing question. “Draft a clear, friendly response template for a patient asking why their insurance did not cover a visit. Leave the specifics as brackets I will fill in.”
The four we will not fully write out here
The remaining prompts in the Kit cover appeal escalation, aging-report triage, contract-rate spot checks, and a front-desk daily huddle template. We keep the fully built versions in the Kit because they are the ones we tune most often as payer behavior shifts, and a stale prompt is worse than no prompt.
How to actually use these
Do not paste a prompt once and judge it. Run it, read the output like an editor, and tell the AI what to fix. “Too long, cut it in half.” “You invented a policy number, remove it.” That back-and-forth is the whole skill, and it takes about a week of daily use before your team stops treating the AI like a search engine and starts treating it like a fast, tireless junior analyst who needs supervision.
One more time, because it is the thing that gets practices in trouble: no PHI in consumer tools without a BAA. De-identify, then prompt. Every template above is built to respect that line, and so should every prompt you write next.
We built the full library because our own team needed it, and because the practices we work with kept asking for the exact wording. It is the same set our billers use on real denials, minus anything that could identify a patient. If you want all 22, plus the versions we keep tuned as payers change the rules, they are in the Kit.
The same operational thinking behind these prompts runs through our revenue cycle management services and our use of AI in medical billing every day.
Get the AI Kit → https://eligibility.natrevmd.com/natrevmd-ai-kit-tool
