81.7% of appealed denials get overturned

Two numbers explain why prior authorization feels the way it does when you are the clinician, the biller, and the appeals department. The American Medical Association's 2026 prior authorization survey, published 13 May 2026, found that 81.7% of appealed denials were fully or partially overturned. KFF's analysis of 2023 ACA marketplace data found that consumers appealed fewer than 1% of denied in-network claims. The system is not primarily rejecting care that fails review. It is rejecting care that nobody has time to argue about.

For a solo occupational therapist or speech-language pathologist, that gap is the whole problem. The AMA's respondents reported roughly 40 prior authorization requests per physician per week, consuming an average of 13 hours of physician and staff time, and 40% said they employ staff who work exclusively on prior authorizations. You do not have that person. You are that person, usually after 6pm.

So the AI question here is not "will AI make prior auth disappear." It is narrower and more useful: which parts of this can software actually take off your evening, and which parts are legally yours no matter what tool you buy. Seven questions, answered with primary sources.

1. Can my insurer use AI to deny my OT or speech therapy prior authorization?

To screen and process it, yes. To be the only thing that decided it, increasingly no, depending on where you practice. California's SB 1120, the Physicians Make Decisions Act, took effect 1 January 2025 and requires that any AI-assisted denial, delay, or modification of care be reviewed and decided by a licensed physician or qualified health professional, and prohibits basing a medical-necessity determination solely on an algorithm. Nebraska LB 77 and Texas SB 815 carry similar prohibitions, and Iowa House File 2635, effective 1 July 2026, is the clearest drafting of the pattern: utilization review organizations may use AI for the initial review of a request, but AI may not be the sole basis for denying, delaying, or downgrading a medical-necessity decision (Sheppard Mullin and Holland & Knight state legislative reviews, 2026).

State AI-denial legislation is moving fast enough that a list is out of date within a quarter, so confirm your own state's current statute with your insurance commissioner before you cite one in an appeal letter. What matters practically is simpler: if your denial arrived in under a minute with generic language and you are in a state with one of these laws, "please confirm this determination was reviewed by a licensed clinician" is a legitimate first line of an appeal.

So what for you: find out today whether your state restricts AI-only denials, because that single sentence costs you nothing to include and changes who has to do work next.

2. Do payers have to tell me why they denied it now?

For Medicare Advantage, Medicaid managed care, CHIP, and plans sold on the federal marketplace, yes, and this is newer than most solo practices realize. The CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F) requires impacted payers to send a specific reason for every prior-authorization denial, and to decide expedited requests within 72 hours and standard requests within seven calendar days. Those provisions started applying in 2026. CMS has been explicit that the notice, turnaround, and public metrics-reporting requirements apply regardless of how the payer received the request, so they do not wait for the payer's FHIR API, which has a compliance date of 1 January 2027 (CMS, CMS-0057-F fact sheet, 2024, and CMS implementation guidance, 2026).

The reason this matters more than it sounds: a specific denial reason is the input an AI appeal draft needs to be worth anything. A denial that says "does not meet medical necessity criteria" gives software nothing to work with. A denial that names the criterion you failed tells you exactly which paragraph of your evaluation to rewrite.

So what for you: if a plan covered by this rule sends you a denial without a specific reason, that is itself a compliance failure worth naming in writing before you spend an hour drafting a clinical rebuttal.

3. Is prior authorization for OT and speech therapy actually going away?

In parts, and not evenly. UnitedHealthcare announced on 5 May 2026 that it will eliminate prior authorization for 30% of the services that currently require it, naming certain outpatient therapies and chiropractic care among them, affecting roughly 50 million members across commercial, Medicare Advantage, and employer plans, with implementation by the end of 2026 (UnitedHealth Group newsroom, May 2026). Cigna has said it intends to remove about a quarter of medical services from prior authorization.

Then there is the other direction. Superior HealthPlan, a Texas Medicaid and CHIP plan, added a new prior authorization requirement for speech therapy evaluations for dates of service on or after 1 August 2026, with claims denied on receipt if no authorization is on file (Superior HealthPlan provider newsroom, 2026). That is six days ago. Requirements are being removed at the national level and added at the plan level in the same quarter.

So what for you: do not build your 2027 admin plan on the assumption that this burden is shrinking. Build it on the assumption that the burden is moving, and that you will find out from a claim denial rather than a letter.

4. Is it worth appealing a denial, or is it a waste of an evening?

On the AMA's numbers, appealing is the single highest-yield hour in your admin week. 81.7% of appealed denials were fully or partially overturned in the 2026 survey. That figure comes from physician practices, not OT and SLT practices, so treat it as directional rather than a rate you can bank on. Even discounted heavily, it sits against the fewer-than-1% appeal rate KFF measured in marketplace data, and against a finding in the same AMA survey that 74% of physicians say denials have increased over the past five years.

Frankly, if you are currently writing off denied OT or SLT authorizations because the appeal takes 45 minutes and the visit is worth less than that, the math runs the other way. The appeal is not just recovering one visit. It is recovering the whole authorized episode of care behind it, and it is the only mechanism that generates a record of the payer being wrong.

So what for you: pick a floor, say any denial covering three or more scheduled visits, and appeal every single one above it for the next quarter. Track the overturn rate. That number is your own business case for whether to spend money automating this.

5. Can I use ChatGPT to write the appeal letter?

Not with patient identifiers in the prompt, unless the vendor has signed a Business Associate Agreement covering the specific product and subscription tier you are using. HIPAA defines 18 identifiers as protected health information: names, dates of service, medical record numbers, and more. Pasting a denial letter with the patient's name and dates into a free consumer chatbot account with no BAA in place is a disclosure to a third party you have no agreement with.

There is a workable version of this. Draft the clinical reasoning from a de-identified summary (diagnosis category, functional deficits, objective measures, criterion cited in the denial), then add identifiers yourself inside your own EMR or word processor. That keeps the drafting benefit and keeps PHI out of the tool. We worked through which general-purpose AI vendors will and will not sign a BAA, and at what price, in ChatGPT or a purpose-built AI scribe: which one your BAA actually covers.

So what for you: write down which AI tools in your practice have a signed BAA and which do not, then set a rule that the second list never receives patient identifiers. That is a fifteen-minute exercise that removes most of the realistic exposure here.

6. Will an AI-drafted appeal get rejected for being written by AI?

No rule requires you to disclose that software helped draft an appeal, and no payer criterion turns on authorship. What payers reject is an appeal that does not address the criterion they cited. The failure mode with AI-drafted appeals is not detection, it is fluency: a well-written letter that argues general medical necessity while ignoring the specific coverage criterion in the denial code will lose to a clumsy letter that quotes the criterion and shows the objective measure that satisfies it.

The clinical content also remains yours. If an AI draft asserts a functional gain your notes do not document, you have signed your name to a claim your record cannot support, which is a bigger problem than a denied visit. Every professional-body AI position published so far, including AOTA's Policy E.19 on ethical AI use and ASHA's guidance for clinicians, puts final responsibility for AI-assisted output on the licensed clinician, not the tool.

So what for you: use AI for the structure and the first draft, then check one thing before sending, that every clinical assertion in the letter traces back to something already in your documentation.

7. What is the cheapest way for a solo practice to cut prior-auth time?

Sequence matters more than software here, and the first two steps cost nothing. Start by building a payer-criteria file: for each plan you bill, save the actual coverage criteria for your top three CPT codes, so you are writing the initial request against the standard rather than guessing at it. Second, template the appeal itself, one structure per denial reason you see repeatedly, because your denials cluster far more than they feel like they do.

Only then does automation earn its keep. Vendors in the appeal-automation category claim drafting time falls from roughly an hour to under five minutes, which are vendor figures and should be treated as marketing until you have measured your own baseline (vendor blog claims, 2026, flagged as vendor-sourced). Your own tracked overturn rate from question 4 is a better basis for a purchase decision than any of them. The billing-side economics for these two specialties, including where automation stops paying for itself at solo volumes, are covered in the $2,480 threshold: what AI billing tools actually fix for solo OT and SLT practices.

So what for you: spend the first month building the criteria file and the templates, and the second month measuring. Buying software before you have either is how practices end up paying a subscription to automate a process that was never defined.

The bottom line

Prior authorization is not primarily an AI problem for a solo OT or SLT practice. It is a capacity problem that AI touches at two points: payers are using it to decide faster than you can respond, and you can use it to lower the cost of responding. The regulatory picture has moved in your favor this year more than most practice owners realize, with specific denial reasons and hard turnaround clocks now required across Medicare Advantage, Medicaid managed care, and marketplace plans, and a growing set of states refusing to let an algorithm be the last word on medical necessity. The professional bodies are pushing the same direction: APTA, AOTA, and ASHA released a joint framework in March 2026, "Care Delayed Is Care Denied: A Therapy Consensus to Reform Prior Authorization," calling for clear denial explanations and published payer metrics.

None of that helps if the denials still go unappealed. Start there, before you buy anything. For the Medicare side of AI-driven review, see the WISeR model: 6 questions about Medicare's AI prior-auth pilot.

Related: The $2,480 threshold: what AI billing tools actually fix for solo OT and SLT practices  ·  ChatGPT or a purpose-built AI scribe: which one your BAA actually covers  ·  The WISeR model: 6 questions about Medicare's AI prior-auth pilot

This article is for informational purposes only and does not constitute legal, billing, or clinical advice. State AI-denial statutes and payer prior-authorization policies change frequently; confirm current requirements with your state insurance commissioner, your payer contracts, and legal counsel before relying on anything above.