Prompt Health

An EMR and billing platform that has sold since 2019 carries the AI as add on modules. Sidekick's models turn a therapy visit into a structured SOAP note, suggest goals and CPT codes and answer documentation questions, and Insight's models read every note to score it against Medicare guidelines. Neither module exists without them.

Most of what a clinic buys, though, is scheduling, the clearinghouse, billing, patient engagement, intake, compensation plans and an optional billing service, all of which run without a model. Sidekick is priced separately from the core EMR, and the platform sold for six years before the PredictionHealth merger brought the scribe in. Take the AI away and Prompt is still a complete rehab therapy EMR, which makes the models a feature of the platform rather than its core.

Review sits at each point where AI output enters the record. Sidekick drafts the SOAP note for the clinician to add, edit or delete section by section; Structured Note Generation fills a structured field only after the clinician approves each suggestion; CPT suggestions are accepted or rejected one at a time or all at once; and goal and treatment suggestions are offers, with treatments drawn only from the clinic's own library. Prompt states that its AI assists and that final decisions stay with clinicians and staff, and real time compliance flags appear before sign off.

The missing half is what happens at the edges: no threshold at which Sidekick declines to draft, no description of how an error found after signing is traced or corrected, and no account of what AskAI will refuse to answer. The pitch calls finished notes audit ready, a claim the review steps rather than the drafting have to carry.

The AI is described by what it does, ambient listening, smart dictation, structured note generation, real time CPT checks, a rehab specific chatbot and AI case summaries, without naming what produces it. No model, model provider, training data or version is identified.

Product update posts announce features, such as the Treatment Suggestions beta of October 2026 and structured field drafting, without saying when the underlying models change. The merger FAQ says Prompt now owns PredictionHealth's proprietary AI tools, technology infrastructure and clinical informatics team, which places development in house, while the AI pages concede that outside AI services process data in some form. A practice cannot tell which parts are Prompt's own models and which are licensed.

The chain is described only in outline. Patient data is said never to be processed through external AI services in a way that would compromise protected health information, which confirms that outside AI services exist without naming any of them, the model provider included. The merger brought PredictionHealth's tools and infrastructure in house, so part of the chain is Prompt's own.

No subprocessor list covers the clinical service. The website privacy policy names marketing and analytics tools such as Google Analytics and reCAPTCHA, and the website terms send chatbot conversations to unnamed third party licensors. A practice cannot tell which companies receive visit audio or note text, or what retention and training terms apply once they do.

Outcome figures are the vendor's own, drawn from what it calls analytical data across more than 6,000 practices: 75 percent less documentation time, a 20 percent rise in functional CPT code use, compliance scores up 30 percent in one month, 33 percent less clinician churn, and coding accuracy up 8 points from the CPT checker. For Insight the claims are 37 percent better compliance among the clinicians with the most documentation risk, a 5 to 10 times return through reimbursement and half the time to sign off.

None of these comes with a method, baseline, period or denominator. Named clinics stand behind the product, among them Dynamix Physical Therapy, whose owner says notes that took 45 minutes to an hour now take 5 or 10, and HIDEF Physical Therapy on compliance and coding. No study measures drafted notes or code suggestions against what happened in the visit.

Assurances are general. Every AI feature is said to operate entirely within Prompt's secure ecosystem, and patient data is never processed through external AI services in a way that would compromise protected health information. That wording implies outside AI services are used, without saying which ones or on what terms.

Nothing addresses whether recordings or notes train models, how long ambient listening audio is kept, or whether data is de identified before any outside processing. The legal pages cover the website, where the terms bar visitors from using site content to train AI and route chatbot conversations to third party licensors. How Sidekick guards against an invented or wrong detail in a draft, beyond the clinician's review, is not described, and no safety engineering is set out for AskAI's answers.

HIPAA compliance is claimed across the product pages and FAQ, for the EMR and for every AI feature, Sidekick's ambient listening and AskAI included. Business associate status follows from the privacy policy, which says customer protected health information is governed by Prompt Health's business associate agreement with each customer and by the customer's own notice of privacy practices.

That agreement is not available to read, and neither its scope nor any term on subcontractors is stated. The privacy policy itself covers the website and marketing rather than the service holding patient records, and the website terms tell visitors not to share protected health information there or with the site chatbot. No document describing how the clinical service handles patient data is open before contracting. Recording consent is left to each clinic under its own state's law, with guidance in the support center.

No security certification is claimed: no SOC 2 report, HITRUST or ISO 27001, no trust center, and no security page among the legal and product pages. Security appears only as assurance, a secure, HIPAA compliant ecosystem and patient data protected across all workflows, roles and integrations, with no outside audit, penetration test or control description behind it.

The website privacy policy covers marketing data rather than the clinical service. For an EMR that holds the full chart and billing record, plus visit audio captured by ambient listening, there is no attestation for a practice to read before contracting, and nothing to show how the PredictionHealth systems were brought under the same controls after the merger.

No regulatory position is stated for any AI feature. Most of the AI is documentation and billing work, drafting notes, checking CPT codes and scoring compliance, where that silence is unremarkable.

Two features sit closer to clinical support. AskAI answers clinical as well as administrative questions for clinicians inside the note, and the Treatment Suggestions beta ranks stretches, exercises and education for a new patient, drawing only from the clinic's own treatment library with its usual sets, reps and codes. Prompt's statement that AI assists and final decisions remain with clinicians assigns responsibility rather than stating a regulatory position. Nothing says what AskAI's clinical answers or the treatment rankings are intended for, in what setting, or what they must not be used for.

Sidekick, Insight and AskAI come with no account of how they are tested: no responsible AI policy, no evaluation results, no bias testing and no material written for a practice's own review of the models. Accuracy across disciplines, conditions, accents and patient groups goes unaddressed.

Specialty instructions and the orthopedic limit on Structured Note Generation show the product is tuned by discipline, yet nothing compares performance in speech therapy, pediatrics or chiropractic with orthopedic physical therapy, where most of the customer stories sit. Insight's provider scorecards rate clinicians against Medicare guidelines, and how that scoring model is itself checked for accuracy or consistency is not stated either.

Review before commit is the mechanism behind every AI output: drafts, structured field suggestions, CPT codes and treatment suggestions all wait for the clinician, and real time compliance flags surface before sign off. Insight then scores every note after the fact against Medicare guidelines, which gives a practice a second check and a trail of where documentation fell short.

Nothing stands behind the output itself. No error rate for drafted notes or code suggestions is stated, no warranty or indemnity attaches to the AI, and the only terms open to read are the website terms, which disclaim any warranty for the accuracy of the site chatbot. The CPT checker's 8 point coding accuracy gain comes without method or baseline, so it does not work as an error rate.

Prompt is itself the record system, and Sidekick and Insight live natively inside it, so the scribe writes straight into Prompt's own SOAP structure and structured fields. Exchange with systems outside Prompt is thin: an integrated clearinghouse for claims and remittances, telehealth through integrations that go unnamed, partner connections such as Weave and the former PT Wired app, and migration from a previous EMR by importing reports the clinic pulls.

No FHIR or HL7 interface, health information exchange or certification is named. Under PredictionHealth, Sidekick also worked alongside Clinicient, Raintree and WebPT, and Prompt's 2025 giveaway rules still listed those EMRs. Existing customers can keep using that platform, but the current product pages describe Sidekick only inside Prompt.

Prompt runs as a cloud platform sold only in the United States, through the web and a Sidekick Mobile app that records visits on a phone and syncs notes to the EMR. No hosting provider, data center region or tenancy model is stated, and residency is implied by the US only market rather than committed.

All AI processing is said to stay inside Prompt's secure ecosystem even though some outside AI services are involved, so the path visit audio and note text take during drafting is not described. The PredictionHealth platform still runs for its existing customers on its own infrastructure, with single sign on between the two planned as part of the integration.

The platform is sold at a flat rate per provider, with reminders, digital intake, home exercise programs, claim scrubbing and billing tools included rather than charged as extras, but no figure appears. Sidekick is a separate add on to the core EMR with a free trial, priced per practice through a demo, and Insight is quoted the same way.

The only number on Prompt's own site sits in 2025 giveaway rules, which valued a one year, single clinician Sidekick license at about $1,260. Contract terms are set with sales, a buyout program covers clinics locked into another EMR's contract, and the optional revenue cycle service is priced separately. A practice can learn how it would be charged and what is bundled before a demo, and not what it would pay.

Coverage is narrow by design: outpatient physical therapy, occupational therapy, speech language pathology, pediatric therapy and chiropractic, for solo clinics through enterprises with hundreds of locations, university programs included. Prompt says plainly that it is not built for hospitals, acute care or general medical practices and is sold only in the United States.

Sidekick takes specialty instructions so drafts fit each discipline, and Structured Note Generation, which fills Prompt's structured fields from approved suggestions, is limited for now to providers treating standard orthopedic conditions. Customer stories come from physical therapy groups such as Marino Physiotherapy, Evolve Physical Therapy, Inspired Physiotherapy and Onondaga Physical Therapy. They back the orthopedic physical therapy core more than the speech, pediatric and chiropractic settings the product also names.