2019: the year Medicare's outcome mandate died

Medicare stopped requiring you to report functional outcomes on therapy claims on January 1, 2019. If your EHR still nags you to pick a severity modifier before you can submit, or a vendor is pitching an AI feature that promises to "automate your outcome-reporting compliance," that's habit and marketing, not law. Outcome monitoring itself hasn't gone anywhere: it's still how you prove an intervention worked, to a patient, to a referral source, or to yourself. What's changed is that AI vendors have started attaching the word "outcomes" to nearly every documentation and analytics tool sold into solo OT and SLT practices, and it's worth knowing exactly which claims hold up. Here are the six questions clinicians are actually asking.

1. Does Medicare require outcome reporting for occupational therapy in 2026?

No, and it hasn't since 2019. From 2013 through 2018, Medicare required PT, OT, and SLP providers to report nonpayable G-codes and severity modifiers on claims, documenting a patient's functional status at the start, during, and at the end of an episode of care. CMS's own Functional Reporting page confirms the requirement "have been discontinued, effective for dates of service on and after January 1, 2019," a change set out in the CY2013 Physician Fee Schedule final rule (CMS.gov, confirmed direct fetch September 2026). No claims-based functional outcome reporting mandate has replaced it for therapy services since.

So what for you: if a vendor demo leads with "AI-powered functional reporting compliance," ask what specific current requirement it's solving for, because the federal one it's most likely describing hasn't existed for six years.

2. Can AI accurately predict patient outcomes in therapy?

The honest answer is that almost none of it has been independently checked. A systematic analysis published in PLOS Digital Health on August 19, 2026 (Abulibdeh, Cajas Ordóñez, Celi, and colleagues) reviewed every FDA-cleared AI and machine-learning medical device through December 5, 2025, 1,357 devices in total, and found that only 3, 0.2%, had been evaluated for a patient-centered outcome such as mortality, morbidity, or readmission. Only 34 devices, 2.5%, even had a registered prospective trial; just 12 posted results, and 12 reached peer-reviewed publication. That figure spans every AI medical device category, not therapy specifically, and it comes with an important second fact: most of the AI outcome-tracking or predictive-analytics features bundled into OT and SLT documentation platforms are not FDA-regulated devices at all, since software that only tracks, summarizes, or charts data, without making a diagnosis or treatment recommendation, generally falls outside the agency's device definition. That means those features carry no independent review either way, positive or negative. One vendor blog promoting AI tools for occupational therapy claims a "40% improvement in treatment success rates" and "40% faster goal achievement" (Sprypt, December 2, 2025); no study, sample size, or methodology is cited anywhere in the post, which makes it a marketing claim, not evidence, per this site's standing rule against citing vendor-only figures without flagging the source.

The same PLOS Digital Health analysis found that radiology accounts for 78% of all FDA-cleared AI devices, yet even radiology, the specialty with by far the largest and most mature AI device market, had prospective trials for fewer than 1% of its own cleared tools. Outcome-tracking and predictive-analytics software for OT and SLT sits well behind radiology in both device count and scrutiny, which is another way of saying the validation gap in therapy-specific AI outcome tools is not an exception to the pattern; it's the pattern, playing out in a category with even less regulatory attention on it than radiology gets.

So what for you: treat any AI vendor's "outcomes improved by X%" claim as a sales pitch until they can show you the actual study behind it. Most can't, because for outcome prediction specifically, the study usually doesn't exist yet.

3. What outcome measure should occupational therapists use with AI documentation tools?

AOTA doesn't name one, and that's deliberate. Its guidance on intervention outcomes instructs practitioners to "use outcome measures and assessment tools at the beginning of the occupational therapy process to guide decision-making" and to "assess outcomes frequently to determine if there is a need to modify the intervention plan" (AOTA, "Domain and Process: Intervention Outcomes," aota.org, confirmed direct fetch September 2026), but it stops short of prescribing a single standardized instrument, leaving that choice to the clinician's judgment based on caseload and setting. That matters for AI tools specifically: no documentation platform can correctly "auto-select the right outcome measure" for you, because AOTA itself never defined a single right answer to select.

So what for you: pick the standardized measure that actually fits your caseload first, a goal-directed instrument for adult rehab, a validated pediatric measure for a school-based caseload, then ask whether an AI tool can score and track that specific instrument. Don't let a vendor's default measure define your outcome framework for you.

4. Is ASHA's NOMS outcome tracking system required for speech therapy?

No. ASHA's National Outcomes Measurement System is, in the organization's own words, "a voluntary data collection registry that illustrates the value of audiology and speech-language pathology services and enables clinicians to improve the quality of those services" (ASHA, asha.org/noms, confirmed direct fetch September 2026), combining patient-reported outcomes with clinician-administered tests. Enrollment is entirely optional, and no payer or regulator currently ties reimbursement or licensure to participation in NOMS or any equivalent registry.

So what for you: enrolling in NOMS is a legitimate way to benchmark your outcomes against a national dataset if that's useful to you, but it's a choice, not a compliance requirement, so don't pay extra for an AI feature sold specifically as "NOMS compliance."

5. Can I use ChatGPT to write occupational therapy goals?

Yes, for drafting language once you've already set the clinical target yourself, and this is the cheapest, lowest-risk AI use case on this list. Clinicians have been documenting this workflow since at least 2023, using a general-purpose chatbot to turn a rough clinical target into properly graded, measurable goal language, faster than writing each line from a blank template (OT Latina, "Using Chat GPT to Develop Occupational Therapy Goals," July 2023). The tool is drafting the sentence structure, not deciding what the patient needs to achieve, and that division matters for exactly the same reason it matters in the outcome-measure question above: the clinical judgment stays yours.

So what for you: strip identifying details before you paste anything into a consumer chatbot, then let it help phrase the goal. A free or $20-a-month tool used this way saves real drafting time without needing a signed vendor contract at all, see our documentation and consent questions piece for exactly where that line sits.

6. Does AI outcome-tracking software need a HIPAA business associate agreement?

Yes, if it stores or processes identifiable patient data, and that's true regardless of what the feature is marketed as doing. Any AI-powered outcome-tracking, analytics, or prediction tool that touches protected health information needs a signed BAA, the same requirement that applies to an AI scribe, a billing tool, or any other vendor handling PHI. Being labeled an "outcomes" feature doesn't create an exemption; it's the data the tool touches that triggers the requirement, not the feature name on the pricing page.

So what for you: before you adopt an AI outcome dashboard, ask the vendor for their BAA in writing before you ask about their algorithm. If they can't produce one, how accurate their predictions are is the least of your problems.

The bottom line

Run the questions together and the outcome-monitoring landscape for OT and SLT practices in 2026 looks a lot less regulated than most marketing implies: no federal reporting mandate has applied since 2019, no national registry is required, AOTA leaves the choice of instrument to you, and the AI industry's own validation record on outcome prediction specifically is thin enough that a single peer-reviewed study, three devices out of 1,357, stands out as news. The one place AI reliably earns its cost with the least risk attached is goal drafting, not outcome prediction. If a vendor pitches you an AI outcome-tracking feature this quarter, the useful order of questions is: show me the BAA, then show me the study, then we'll talk about the dashboard. For how the same documentation-versus-clinical-judgment line plays out in plan-of-care requirements specifically, see Medicare's POC needs 3 numbers no AI drafts.

Related: HIPAA's 6 years isn't your state's clock.  ·  Medicare's POC needs 3 numbers no AI drafts  ·  $656-$1,319: AI audit for OT/SLT practices

This article is for informational purposes only and does not constitute legal or clinical advice. Federal, state, and association rules change frequently; confirm current requirements with legal counsel and your professional association before relying on anything above.