iScribeHealth

AI now does the documentation work that people once did. The Nashville company started in 2015 inside orthopedic practices with an EHR integrated mobile app for speech to text dictation, from which providers could also select diagnoses, place orders and complete E&M coding. From 2016 to 2022 it ran a Virtual Scribe Service, in which remote scribes turned recorded encounters into finished notes, and that service grew to more than 150 scribes.

The AI platform arrived in 2023, and the company says its entire physician base moved from human scribes to AI without losing a provider. Its current products all run on that engine. iScribe AI produces ambient notes and pushes them into the EHR, iScribe Flex serves practices without an integration, and iScribe E&M Assist recommends evaluation and management codes. Speech to text dictation remains in the app.

The generative layer was added to a dictation and scribing business rather than founding it. The company describes the scribes' accumulated knowledge of good documentation as what shaped the AI.

The clinician signs every note. During the visit, iScribe AI records the provider's macros and surfaces clinically relevant data so errors can be caught and corrected on the spot. The full encounter is processed in seconds, and the structured note reaches the EHR for the provider to review and sign. The company describes that review as taking seconds, done before the provider leaves after the last patient, and suggests signing on the way to the next one.

Dictation is driven by the clinician, who speaks into a named note section rather than leaving placement to the system. On the ambient path the system decides what goes where, and no acceptance rate, edit burden figure or confidence threshold is published for it.

Coding follows the same pattern. E&M Assist links each recommended code to the supporting documentation with a justification, nothing bills without the provider's sign off, and administrators can flag individual notes for review from a Coding Insights dashboard.

The engine combines prompt engineering, provider specific customization and several large language models alongside natural language processing. The company says it uses multiple frontier models with consistently strong performance rather than a single domain specific model. It describes the AI it introduced in 2023 as trained on nearly a decade of institutional scribe knowledge, with long tenured scribes engineering the prompts and early adopter physicians testing each version.

Two accuracy figures are published. Speech recognition is put at greater than 90% transcription accuracy, rising above 95% when tuned to a provider's specialty, audio environment and dictation style. E&M Assist recommends a code with 95% audited accuracy, backed by in house certified professional coders and analysis of AMA medical decision making guidelines.

The models themselves are not named, and no model card, version history or evaluation method behind either figure is published.

Encounter content reaches more than one external model. iScribe AI combines multiple large language models with natural language processing, choosing frontier models on demonstrated performance rather than relying on a single domain specific model. The models in use can change as performance changes, and none is named. No hosting arrangement or subprocessor list is published either.

People sit in the chain as well. Employees may review recordings internally to support provider specific customization, troubleshoot issues and improve the service, so their access reaches the audio and not only the generated text. The company says this information is never shared externally. Its former scribes engineered the prompts, and a 2025 profile reported that its most expert scribes were kept on to review AI generated notes.

Those reviewers are employees rather than contractors. Where they work, whether their access is logged and time bound, whether review covers every note or a sample, and whether a practice can decline it are not published.

The evidence is customer outcomes reported by the company. Its Florida Orthopaedic Institute case study describes 92 providers trained and live in a single day while the practice launched on athenahealth, $120K in annual savings from moving off manual transcription, and 75% fewer issues reported at go live than in a typical EHR transition. The company also cites savings of $15,000 to $25,000 per provider per year in reduced transcription spend, testimony from an operations executive at a large orthopedic group, and a 92% rate of trial providers converting to full adoption.

Adoption at scale supports the picture. iScribeHealth supports more than 230 athenahealth practices across 30+ medical specialties, and athenahealth named it a Preferred Partner in July 2026, citing its performance, customer value, adoption and integration.

These figures measure cost, speed and uptake rather than documentation quality or clinical benefit. No study, controlled evaluation or outside assessment of the notes is published.

Customer data is never used to train public large language models, in the company's words. The word public carries weight, since the statement does not reach the company's own models or uses such as evaluation and quality improvement. The company describes the AI it launched in 2023 as trained on nearly a decade of institutional scribe knowledge.

Retention follows each practice's needs rather than a published schedule. Audio files are stored under SOC 2 compliance standards, providers can reach their own audio in the Enterprise Management Console, and downloads are off by default for every provider and customer unless enabled on request. Recordings made without a connection stay on the iPhone or iPad until they can be submitted, and Flex autosaves audio locally in the browser.

People are part of the data path. The virtual scribe service grew to more than 150 remote scribes, and a 2025 profile put the workforce at around a hundred and twenty when the company moved to AI, keeping the most expert to review AI generated notes against the recording and each physician's style. Employees may review recordings internally for customization, troubleshooting and service improvement, and the company says this information is never shared externally. That makes reviewer access to audio an access surface alongside the systems.

Reviewer location, access logging, whether identifiable content is visible, whether every note or a sample is reviewed, and whether a practice can opt out are not published. Neither is any de identification practice, guardrail, hallucination control or process for handling a safety event.

The company states that it is HIPAA compliant. Practices sign a master service agreement with it directly, and no business associate agreement, or description of what one covers, is published.

Two features of the product put weight on that scope. The first is the review workforce. Employees may listen to recordings for customization, troubleshooting and service improvement, and nothing published says how those people are trained, bound to confidentiality, located or monitored.

The second is the contracting chain. Many customer practices run athenahealth, which named the company a Preferred Partner in July 2026, and whether athenahealth plays any part in the agreement covering iScribe's processing is not said. Practices that used the earlier human scribe service may also hold agreements written for that service, whose data flows differ from those of the AI product.

Security rests on a SOC 2 program and HIPAA compliance. The company says it maintains compliance with all SOC 2 requirements and actively maintains its SOC 2 certification, covering security, availability and privacy controls, and audio files are stored under SOC 2 compliance standards. No report type, audit period or auditor is named, and there is no trust center where a report could be requested.

People with access to customer content make those controls matter more. Employees may review recordings for quality assurance, customization and troubleshooting, and how reviewer access is granted, logged, limited to assigned work and removed is not described publicly. The company ran a remote scribe service of more than 150 people before moving to AI in 2023. athenahealth named it a Preferred Partner in July 2026, citing performance, adoption and integration, and whether that designation involved a security review is not said.

Ambient documentation that the clinician reviews and signs needs no FDA clearance, and none is claimed. The product now does more than document. Since its first version the app has let providers select diagnoses, place orders and complete E&M coding from their phone, and E&M Assist now recommends an evaluation and management code from the note, the patient's history and prior claims. Each code links to its supporting documentation, and nothing bills without the provider's sign off. The company says real time denial prevention tools are in development.

Coding and ordering touch reimbursement and clinical action rather than documentation alone, though neither makes the product a device. E&M Assist is limited for now to orthopedic and urology practices on athenahealth or NextGen.

Human review adds a second consideration. A 2025 profile reported that expert scribes review AI generated notes against the recording and the physician's style. Where a note has been checked before it arrives, a clinician may read it less closely, while the regulatory position still rests on the clinician's own review. The company does not say whether a note shows what, if anything, a person has checked.

Language support is a setting. With Multilingual Support turned on, iScribe AI captures complete multi speaker conversations in English, Spanish, French, Portuguese, Japanese, Italian, German, Dutch, Hindi and Russian. When an interpreter joins the visit, the model captures the conversation in English and recognizes both the provider and the interpreter.

Accuracy is reported as a single figure for speech recognition, above 90% and above 95% when tuned to a provider's specialty, audio environment and dictation style. No fairness statement, subgroup analysis or performance figure by language, accent or dialect is published.

The company has people positioned to see uneven performance. Former scribes engineered the prompts, employees review recordings, and a 2025 profile reported expert reviewers checking AI notes against the recording. Whether their corrections are logged and examined for patterns by speaker is not said. A review step that quietly fixes the same kind of error again and again keeps the output looking right while the model keeps failing for the same speakers.

Quality control rests on people and process more than on contract terms. Long tenured scribes engineered the prompts behind iScribe AI, early adopter physicians tested each version, and a 2025 Healthcare IT Today profile reported that the company kept its most expert scribes to review AI generated notes for accuracy against the recording and for the physician's style. Employees may also review recordings for quality assurance, customization and troubleshooting.

The company's own pages describe the result as a fully automated documentation experience, with a processed note in about 30 seconds. They do not say whether a person checks a given note before the clinician sees it, whether review covers every note or a sample, whether a practice can opt out, or how reviewer access is logged and bounded.

Published accuracy figures cover speech recognition, above 90% and above 95% when tuned, and E&M coding at 95% audited accuracy. No warranty, indemnity or remediation commitment for an inaccurate note is published, and scope and pricing are set in a signed master service agreement whose terms are not public.

Notes go straight into the chart. When the provider ends a recording, iScribe AI pushes a signature ready note with discrete data into the EHR, with no copy and paste. Dictated content can be placed into specific note sections, including chief complaint, history of present illness, review of systems, physical exam and assessment and plan. The company describes the flow between app and record as running both ways, and the app shows the provider's real EHR schedule and a live view of the chart, with previous visits, medications, vitals and demographics.

athenahealth is the deepest integration. athenahealth named the company a Preferred Partner for ambient clinical documentation in July 2026, and it supports more than 230 athenahealth practices across 30+ specialties. iScribe AI also integrates with NextGen, Veradigm (Allscripts Pro), Altera TouchWorks, Greenway PrimeSuite and Epic, though how deep each of those goes is not described. E&M Assist runs only on athenahealth and NextGen.

Practices on other systems can use iScribe Flex, which runs on the same engine without an integration. Providers record from a phone or computer and paste the note into the chart through a browser extension for Chrome and Edge.

Delivery is a cloud service reached through apps. iScribe AI runs on iPhones and iPads on the three latest versions of iOS, and iScribe Flex records from any browser, with a Chrome and Edge extension that floats beside the EHR. Recording works without a connection, since audio is saved on the device or in the browser and submitted later, but transcribing and sending a note needs a stable connection. EHR integration is central to the product, from the athenahealth Preferred Partner link to the five other systems it connects to.

Onboarding is hands on. A kick off call and virtual or on site training precede go live, and the company's Florida Orthopaedic Institute case study describes 92 providers trained and live in a single day. Live support runs Monday to Friday, 7am to 7pm Eastern.

No hosting region, residency option or subprocessor detail is published, and the model services that generate the notes are not named. Configuration is held per provider. Administrators can manage AI preferences on a provider's behalf, and notes and preferred configurations carry over when a practice moves from Flex to iScribe AI.

Residency has a human side too. Employees review recordings, and where those people work is a data location question of its own. It goes unanswered alongside the server question.

Adoption starts with a 14 day complimentary trial that includes a virtual demo built around the practice's workflow and specialty, macros prepared before the trial begins, and onboarding support. Enterprise partners also get a live, in clinic mock encounter, and the company says 92% of trial providers convert to full adoption.

Prices are not published. The company says it does not disclose iScribe AI pricing because customization runs to the provider and practice level, and pricing is set in a signed master service agreement, usually during the trial. Flex pricing is also withheld while the product is in limited early access, and E&M Assist is sold as an add on to an iScribe AI license.

Orthopedics and urology are where the product goes deepest. The company says iScribe AI serves over 55 specialties and subspecialties, tuned for high volume, procedure heavy practices, with custom macros, EHR integration and E&M Assist coding fullest in those two. Named users include the Tennessee Orthopaedic Alliance, Campbell Clinic Orthopaedics, Florida Orthopaedic Institute and Urology Associates in Nashville. The other specialties it names include OB/GYN, ENT, cardiology, neurology, gastroenterology, pain management, podiatry, pulmonology, general surgery, behavioral health and family medicine, internal medicine and urgent care.

Across athenahealth alone, it supports more than 230 practices in 30+ specialties.

The setting is ambulatory and mobile, built for clinicians moving between exam rooms, and telehealth visits are covered. iScribe AI runs on iPhone and iPad, while Flex works from any browser for practices without an integration. Multilingual Support captures conversations in ten languages, from Spanish and French to Hindi and Japanese.