Can I appeal a health insurance denial using AI? Yes, and the useful answer is narrower than the marketing around it. AI is genuinely good at the two jobs that make appeals expensive: turning a denial reason code into a structured argument, and assembling the evidence packet around it. It is unreliable at the two jobs that decide the outcome: knowing what is actually in the chart, and knowing what the plan document actually says. Use it for the first pair and never for the second.
For a dental office the question has two halves. There is the dental plan denial, which follows the carrier contract. Then there is the true medical carrier denial on cross coded claims: the impacted third molars, the bone grafts, the sleep apnea appliance, the trauma case. Different rules, different deadlines, different letterhead. An AI draft that does not know which one it is writing produces a confident letter aimed at the wrong reader.
Name the denial before you draft anything
Most failed appeals fail here, not at the writing. The remittance tells you which family the denial belongs to, and the family decides whether an appeal is even the right instrument. A claim adjustment reason code of 16 is a corrected claim, and appealing it burns days off your clock for nothing.
| CARC | Denial family | The argument that can win | Evidence that must already exist |
|---|---|---|---|
| 16 | Missing or invalid information | None. File a corrected claim instead | The corrected field, tooth number, surface, or provider identifier |
| 50 | Not a medical necessity per the payer | Clinical necessity measured against the plan's own criteria | Dated chart notes, radiographs, probing depths, failed prior treatment |
| 96 | Non covered charge | Only wins if the exclusion was misapplied to this code or date | Plan language, and your dated verification reference number |
| 151 | Frequency or quantity exceeded | The service history the payer used is wrong, or an exception applies | Actual prior service dates, and a narrative for the exception |
| 197 | Prior authorization absent | Authorization was obtained, or the case met an urgent exception | Authorization number, call reference, or documented urgency |
| 29 | Timely filing expired | The claim was submitted on time and the payer lost it | Clearinghouse acceptance report showing the original date |
Two of those six are not appeals at all. A team that routes everything to an appeal queue spends its month writing letters a corrected claim would have fixed in four minutes. Our breakdown of the top reasons for dental insurance claim denials covers sorting them at intake, and the case for putting AI in earlier, before the denial exists, is in can AI help prevent dental insurance claim denials.
The clock is different for medical and dental
This is where a generic AI appeal tool quietly misleads you: the consumer tools are built around federal health plan rules, and your dental denials frequently sit outside them. As of this writing, non grandfathered group and individual major medical plans follow the federal internal claims and appeals rules. Stand alone dental coverage is often an excepted benefit, so those timelines may not apply and the contract plus state law govern instead. Confirm both with the plan document and your state insurance department.
| Track | Where the appeal goes | Common window to file | Common decision window |
|---|---|---|---|
| Medical plan, internal appeal | The plan or its administrator | At least 180 days from the denial notice | 30 days pre service, 60 days post service |
| Medical plan, urgent care claim | The plan, expedited | Immediately, by phone or portal | As soon as possible, not later than 72 hours |
| Medical plan, external review | State or federal independent reviewer | Within 4 months of the final internal denial | Commonly 45 days standard, 72 hours expedited |
| Stand alone dental plan | The carrier, per the contract | Set by contract, commonly 90 to 180 days | Set by contract and state law |
| State complaint | State department of insurance | Varies by state | Varies by state |
One practical point no AI tool will raise: on a medical plan governed by federal rules, the appeal right belongs to the member. To appeal on the patient's behalf you generally need an assignment of benefits and a signed authorized representative designation on file, and some plans require their own form. Get it signed at the visit, not on day 170. Our guide to how long you have to appeal a dental claim denial covers the dental side of the same clock.
How to successfully appeal a health insurance denial?
The sequence below survives contact with real payers. AI can accelerate steps 4 through 6. It cannot do 1, 2, 3 or 8.
- Get the reason in writing. The remittance code, the plan's notice, and any policy bulletin the reviewer cited. A phone explanation is not an appealable record, though the call reference is worth logging.
- Confirm it is appealable. Corrected claims, coordination of benefits mismatches and wrong payer routing are fixes, not appeals.
- Verify the evidence exists. Open the chart before the letter template. If the radiograph, the periodontal chart or the dated note is missing, the appeal is not ready and good writing will not fix it.
- Write one issue per letter. A letter arguing necessity, then timely filing, then a fee dispute gets denied on the weakest of the three.
- Answer the stated reason directly. If the plan said the film was not received, your first line says what film, taken what date, enclosed at what tab.
- Assemble a packet a stranger can read. Cover sheet listing enclosures, claim number, date of service, subscriber identifier and provider on every page, images labeled with tooth numbers and dates.
- Send by the channel the plan requires. Some payers accept only their own form or portal. A perfect letter faxed to a decommissioned number is a write off.
- Log the send and set the follow up. Date, method, confirmation number, and a calendar entry inside the decision window.
Orthodontic denials are their own case, turning on scored severity indexes and age limits rather than on narrative. Our guide to disputing a dental insurance denial for braces covers what that packet needs.
What is the success rate of health insurance appeals?
Honest answer: nobody can quote a number that describes your payers. Published overturn rates span a wide range and move with the denial type, the appeal level, the line of business and the year. Commonly quoted figures show internal appeals overturning a meaningful minority of denials, with external review overturning a further share of what survives. Any tool promising a specific win percentage is selling.
The statistic that changes your revenue is local, and you can compute all of it from your own remittance data.
| Metric | How to compute it | What it tells you |
|---|---|---|
| Appeal rate | Appeals filed divided by appealable denials | Whether the money is lost at the letter or before it |
| Overturn rate by reason | Appeals paid divided by appeals filed, grouped by CARC | Which denial families are worth the labor |
| Overturn rate by payer | Same, grouped by carrier and plan type | Where to spend your best biller's hours |
| Median days to send | Denial posting date to appeal send date | How much of the window you are giving away |
| Dollars recovered per hour | Net recovered divided by staff hours on appeals | The only number that settles the automation argument |
Run those five for a quarter before and after you introduce AI drafting. Most offices find the appeal rate, not the win rate, is where the money went: the $220 claim never got appealed because an hour of work for $220 does not clear the bar. That is the economics AI changes, worked through in how AI speeds up dental insurance appeals.
How to convince health insurance to approve your appeal?
Reviewers are not persuaded by tone. They are persuaded by a letter that makes approval the path of least resistance, which means arguing against the plan's own standard, not against unfairness in general.
Quote the criterion, then satisfy it. If the plan's policy requires documented pocket depths of 5 millimeters or greater with radiographic bone loss before it pays periodontal scaling and root planing, name the policy, state the measured depths by tooth, and reference the dated radiograph. D4341 and D4342 denials turn on exactly this.
Give dates and measurements, not adjectives. "Severe decay" argues nothing. "Distal caries extending to the pulpal floor, confirmed on the bitewing dated March 4, with a failed restoration placed in 2023" gives the reviewer a box to check.
Use the plan's own verification against it. If you were told during verification that the service was covered, the dated reference number for that call belongs in the letter. It is the strongest single fact a dental office can carry into an appeal, and it costs nothing beyond capturing it at the time.
Ask for the peer to peer. When a denial turns on clinical judgment, a conversation between the treating dentist and the plan's reviewer moves cases that letters do not. Request it explicitly and offer two windows of availability. Keep the written argument to one page; attachments can be thick.
Can AI handle insurance claims?
Split the word "handle" into four verbs, because the answer differs for each.
Read: yes. Reason codes, denial letters, benefit responses and clinical notes are exactly the text machines parse reliably. An AI layer can pull every denial from the day's remittances, group them by reason, and put the appealable ones in front of a human with the context attached.
Draft: yes, with review. Letters, narratives, cover sheets and enclosure lists. The draft should reference only findings that exist in the record, and someone has to confirm it did.
Check: yes, and this is underrated. Scrubbing a claim for the attachment, tooth number, narrative and frequency history before submission prevents more revenue loss than appeals recover. The same pattern surfaces revenue you never billed, the subject of using AI to identify missed dental billing opportunities.
Decide: no. Medical necessity is a clinical determination, and an appeal asserting it is a clinical statement made by a licensed clinician. AI carries no license, cannot attest, and cannot sign. The treating dentist owns anything clinical that leaves the building.
Payers run automation on their side too, and several states have moved to restrict who may issue a medical necessity denial and how automated tools may be used in utilization review. Those rules are changing quickly, so as of this writing, confirm your state's current position with your state insurance department.
What to feed the model, and what never to paste
The difference between a generic AI letter and a usable one is entirely in the inputs. A draft produced from "write an appeal for a denied crown" reads like every other one the reviewer saw that week. Supply all nine:
- The exact denial language and the reason code, copied not summarized.
- The policy or bulletin number the reviewer cited, if there is one.
- The procedure code, tooth number, surface and date of service.
- The relevant dated clinical findings, in the clinician's own words.
- Prior treatment on that tooth or site, with dates and outcomes.
- The plan's written criteria for the service, where you have them.
- Your verification record for the service, with the reference number and date.
- The appeal level and the deadline.
- The submission channel and address, so the letter is formatted for it.
And the rule that outranks all nine: none of that goes into an AI tool without a business associate agreement behind it. A denial letter, a clinical note and a radiograph are all protected health information. Confirm the agreement, encryption in transit and at rest, and access restricted to your team before the first paste, not after.
Where AI written appeals fall apart
Each has a straightforward guardrail, and each has cost somebody a case.
- Invented clinical findings. The draft asserts bone loss that is nowhere in the chart. Guardrail: the reviewer checks every clinical sentence against the record before signing.
- Invented citations. A model will happily produce a plausible policy number that does not exist, and sending one damages your credibility on every future appeal. Guardrail: cite only documents you are holding.
- Wrong identifiers. Claim number, subscriber identifier, date of service and rendering provider carried over from the previous case. Guardrail: a four item check before send.
- Ignoring the actual reason. A fluent necessity argument answering a timely filing denial. Guardrail: the first line names the stated reason.
- Over appealing. When drafting is nearly free, teams start appealing correct denials, which wastes hours and dulls your payers' attention. Guardrail: written criteria for what enters the queue.
When the patient brings you an AI letter
This is increasingly common on medical cross coded claims, and the patient is usually right to try. Consumer appeal tools build a competent structure from the denial notice alone, so the letter never saw the chart. It may assert findings you cannot support, or argue necessity for a code you did not bill.
Handle it in one conversation. Offer to supply the clinical documentation as an attachment under their appeal, confirm the assignment of benefits and authorized representative form are on file, and read the letter for any clinical statement the record does not support before it goes out with your name near it. A patient appeal backed by your dated evidence is strong. One that contradicts your chart notes is a problem you answer questions about later.
Putting it in the workflow
The realistic shape of this is not a chatbot on a billing coordinator's second monitor. It is denials arriving sorted by reason code, the appealable ones drafted against the chart, deadlines already on a calendar, and a human doing the only two things that were ever hard: deciding whether to fight, and verifying every clinical sentence is true. Curo works this way, reading remittances into a denial management queue with the reason attached and a draft ready for review, so the decision arrives before the deadline instead of after it.
The part AI cannot shorten
An appeal argues that the evidence in your chart satisfies a standard the plan set. If the evidence was never captured, the argument does not exist, and the best drafting in the world produces a longer way of losing. The offices that win consistently are the ones where the probing depths were recorded, the pre operative radiograph was labeled, the failed restoration has a date, and the verification call has a reference number in the record.
Get that right and the letter takes minutes, machine or no machine. Get it wrong and nothing else here helps.