Fusion and SPRY both draft occupational therapy and speech therapy session notes with AI. Medicare's plan of care rule still needs three numbers neither tool is shown to set.

If you are a solo occupational therapist or speech-language pathologist weighing an AI documentation add-on built for your discipline rather than a general-purpose scribe, two names come up first. Ensora Health launched Ensai Session Assistant across Fusion, its rehab therapy platform, in February 2026, marketed to physical therapists, occupational therapists, and speech-language pathologists as a tool that "listens to your sessions from any device and instantly writes structured, high-quality notes" (Ensora Health, ensorahealth.com, product page accessed September 2026). SPRY, an AI-native EMR for PT, OT, and SLP clinics priced from $79 per provider per month, describes its AI-Powered Documentation as converting "voice notes and session observations into compliant SOAP notes in seconds" (SPRY Therapeutics, sprypt.com, accessed September 2026). Both are strong at the job they were built for: turning what happened in a session into a readable note fast.

Neither vendor's marketing copy mentions the document that actually decides whether Medicare pays for that session in the first place. The plan of care is a separate, specific piece of documentation, and it is checked against a named list of required content, not against how well the note reads.

What Medicare's plan of care actually requires

CMS's own fact sheet on outpatient rehabilitation therapy documentation states the plan of care "must have, at minimum," five elements: diagnoses, long-term treatment goals, the rehabilitation therapy service type, and three numbers described as the therapy amount (sessions per day), therapy frequency (sessions per week), and therapy duration (total weeks or treatment sessions) (Centers for Medicare and Medicaid Services, MLN905365, "Complying with Outpatient Rehabilitation Therapy Documentation Requirements," September 2025). ASHA's own Medicare documentation guidance for speech-language pathologists describes the identical three-number structure in near-identical language: "the amount of treatment refers to the number of times in a day the type of treatment will be provided," frequency is "the number of times in a week," and duration is "the number of weeks, or the number of treatment sessions" (American Speech-Language-Hearing Association, "Overview of Documentation for Medicare Outpatient Therapy Services," asha.org, accessed September 2026). Two independent sources, one federal payer and one professional association, describe the same checklist. This is not a gray area of interpretation.

So what this means for you: a plan built around goals, interventions, and clinical narrative, which is exactly what an AI documentation tool is designed to produce well, covers roughly half of what Medicare actually requires. Diagnoses and long-term goals are the clinical half. Amount, frequency, and duration are a scheduling decision, and nothing about a session transcript tells an AI tool how many visits a week you intend to bill for.

What CMS's own audit program actually flags

This is not a hypothetical gap. CMS's Comprehensive Error Rate Testing (CERT) Program, which reviews a random sample of Medicare fee-for-service claims for correct payment, names "missing or incomplete POC" as one of its recurring error categories among outpatient rehabilitation therapy claims. The fact sheet's own error-prevention guidance is direct: "Create a complete POC that includes diagnoses, long-term goals, and service type, amount, frequency, and duration" (CMS, MLN905365, September 2025). That sentence is effectively a list of the exact fields CERT reviewers are trained to check are present. A plan of care missing any of the three numbers falls squarely into a documentation category CMS already audits for at scale, separate from whatever a physician's or NPP's certification requirement covers.

So what this means for you: this is not a theoretical documentation nicety you might get away with skipping. It is a specific, named audit category, and CMS has already told providers, in writing, what a reviewer will be looking for.

What today's AI tools for OT and SLT actually draft, and what they do not

Ensora Health's own product page for Ensai Session Assistant is more precise than most vendor marketing about the boundary. It lists "intelligent goal suggestions," described as "discipline-aware prompts aligned with your clients' goals and treatment plans," and an "optional compliance suggestions" feature that helps "highlight missing elements that may be important for insurance or defensibility," Ensora's own words (Ensora Health, ensorahealth.com/product/fusion-rehab-therapy/features/ensai-session-assistant, accessed September 2026). Asked directly in its own FAQ whether the tool captures functional goals and outcome measures, Ensora answers that Ensai "transcribes and summarizes verbal content, including discussions of goals, progress, and functional changes," but adds that "standardized test scores and measurements are best added or verified during your review." That is a vendor telling you, in its own documentation, that the numbers a reviewer checks are not something the AI is confirmed to generate on its own.

SPRY's marketing describes its documentation feature converting session content into SOAP notes, and separately claims, in its own content, that "AI algorithms analyze client data to create individualized treatment plans." That is a vaguer claim, made in a SPRY-published comparison article that also ranks SPRY first among six competing platforms, a vendor's own self-comparison rather than independent evidence, and it should be read with that conflict in mind (SPRY Therapeutics, sprypt.com, comparison content accessed September 2026, vendor-published and self-ranked; flagged accordingly). Nothing in SPRY's public materials describes the AI setting a specific visit amount, frequency, or duration for Medicare purposes either.

Both tools sit well ahead of the generic AI scribes evaluated in our earlier look at OT's five-stage documentation structure, where an independent reviewer flagged that a scribe built solely on a general-purpose language model, rather than one trained specifically on discipline-specific clinical documentation, is itself a red flag. Fusion and SPRY are purpose-built for PT, OT, and SLP terminology, which is a real advantage. Purpose-built for your terminology is a different claim from purpose-built to complete Medicare's plan of care checklist, and the marketing language for both tools blurs that line.

So what this means for you: an AI tool marketed as documentation software for your discipline is very likely producing goal language and session narrative well. Nothing reviewed for this piece, from either vendor's own materials, describes the AI deciding how many sessions a week a patient needs or how many weeks the plan should run. That remains a clinical and administrative decision you make, every time, regardless of which tool drafts the note underneath it.

A checklist that takes less time than the AI draft did

Before filing any AI-assisted plan of care for a Medicare patient, or a commercial payer using comparable medical necessity standards, confirm the diagnosis is stated, not just implied by the goals listed. Confirm the long-term treatment goals are written as measurable outcomes, not just a list of interventions the AI has captured from the session. Confirm the plan states which discipline's services are being provided, occupational therapy or speech-language pathology, when that is not already obvious from the template. Then confirm the three numbers by name: the amount of treatment as sessions per day, the frequency as sessions per week, and the duration as a total number of weeks or sessions. None of that is a task an AI documentation tool has been shown to do for you, and all four checks together take a few minutes against a note the AI has already mostly finished.

This is a distinct check from the plan of care signature workflow covered in our earlier piece on Medicare's 30-day referral shortcut, which is about whose signature legally certifies the plan. This is about what the plan actually has to say before anyone signs it.

The call

If your practice is already using or evaluating Ensai Session Assistant, SPRY's AI documentation, or a comparable discipline-specific AI tool, keep it. The documentation time savings on session notes and goal drafting are real, and both vendors are explicit, in their own product pages, that a clinician has to review and finalize every draft. Add one step to your workflow: before any Medicare plan of care is filed, run it against CMS's own five-element checklist, specifically the amount, frequency, and duration triad, since that is the exact category CMS's CERT program names as a recurring finding and the one no AI tool reviewed for this piece is documented to complete for you.

If you are not sure whether your current documentation would survive that kind of check, that is precisely the sort of gap the AI Opportunity and Growth Assessment is built to find against your own caseload and payer mix, not a generic vendor demo. Start with a free 20-minute discovery call.

See also: our look at how HIPAA's six-year retention clock differs from your state's own record-keeping rules and Medicare's 30-day plan of care signature shortcut for OT and SLT referrals.

Related: OT's paperwork has 5 stages, not 1 · 30 days: Medicare's new referral shortcut · HIPAA's 6 years isn't your state's clock