Charting Tomorrow: A Practical Review of Artificial Intelligence in Medical Documentation Authors: Ferris Zeitouni, Nada E. Botros, Alisha S. Khosla, Marion McKinney, Jasmine N. Craig, Luis Alamo-Irizarry, Venkat K. Rao Published: 2026 ABSTRACT Summary:Rising physician burnout, driven by increased administrative load and documentation burden, has led to the widespread adoption of artificial intelligence (AI)–powered transcription services in medical practices. These tools generate clinical notes from patient encounters and have shown the ability to significantly reduce documentation burden and streamline workflows. This article provides clinicians with a primer on the capabilities, limitations, and practical considerations for implementing AI transcription services in their practice. Across platforms, core features include encounter transcription, automated note generation, multilingual capabilities, custom templates, and Health Insurance Portability and Accountability Act compliance. Platforms with deep integration into electronic health records offer greater workflow automation, such as summarizing prior notes, drafting referral letters and after-visit summaries, and preparing orders, but require significant training and onboarding. More user-friendly options provide rapid note creation with minimal setup, though with limited electronic health record integration. Industry partnerships have facilitated the current development of novel features, including point-of-care prior authorization, a virtual assistant to schedule visits, answer patient questions related to appointments, and suggest relevant screenings. Physicians should evaluate the advantages and logistical challenges of implementing AI transcription tools while also considering liability. Each platform offers unique benefits and limitations, and although these tools may enhance physician workflow, they also present risks for both patients and providers. Takeaways Question: What are the benefits, limitations, and key considerations for implementing artificial intelligence (AI)–assisted transcription services in clinical practice? Findings: This article provides an overview of AI-assisted transcription platforms. These tools can reduce documentation time and improve physician efficiency, but differ in features, data storage, cost, and implementation. Key risks include hallucinations, privacy concerns, and unresolved liability. Meaning: AI transcription tools may improve clinical documentation workflows; however, surgeons should adopt them cautiously, considering vendor capabilities, institutional policies, and a careful review of AI-generated notes. INTRODUCTION Clinical documentation informs patient encounters, ensures continuity of care, provides reference for malpractice claims, and sits at the crux of optimal billing and reimbursement. However, the growing documentation burden has contributed to widespread clinician burnout and inefficiencies in care delivery.1–3 Studies show that more than 60% of physicians report at least 1 symptom of burnout, with clerical burden commonly cited as a leading contributor.3,4 Physicians are estimated to spend between 34% and 55% of their workday completing clinical documentation, with an equivalent opportunity cost of $90–$140 billion annually in the United States.5 In response, healthcare systems have increasingly turned to technological solutions to streamline documentation workflows and reduce clinician load. One promising area of innovation is the use of artificial intelligence (AI) in medical transcription. AI-powered “digital scribes” can transcribe, summarize, and structure clinical conversations in real time, improving efficiency and accuracy in charting, freeing up time for direct patient care, and even increasing usable data.1,5,6 This article provides clinicians with a primer on AI-powered medical transcription tools, outlining their capabilities, limitations, and practical considerations for implementation. It aims to guide physicians in evaluating whether these tools are appropriate for their practice and which option may best meet their clinical and operational needs. EVOLUTION OF AI TRANSCRIPTION IN MEDICINE Medical transcription has evolved significantly from manual documentation to sophisticated AI-powered systems. The introduction of speech recognition technology in the 1980s marked the first major shift toward automation. As computing capabilities advanced, the integration of speech recognition with electronic health records (EHRs) laid the groundwork for improving efficiency. Key technological progress has centered around large language models (LLMs), which combine automated speech recognition with natural language processing to summarize conversations.7 In medical documentation, LLMs function most effectively when integrated directly with EHRs, allowing for automated population of documentation fields and streamlined clinical workflows. Current medical transcription options range from fully manual documentation to advanced AI-powered systems, each with unique benefits and limitations (Table 1). Manual documentation provides complete control and accuracy, but is time-consuming, can detract from patient engagement, and increases administrative burden.8 Both in-person and virtual scribes alleviate documentation burden, allowing physicians to focus on patient interactions. However, medical scribes often have limited medical training and require coaching and oversight.9 Non-AI technology-driven solutions include voice-to-text modalities, which convert spoken words into written text for EHR entry, and real-time medical dictation tools, which transcribe directly into structured note templates while adapting to physician preferences. These methods streamline documentation, but their accuracy may be compromised by accents, background noise, and complex terminology.7 Table 1.Comparison of Commonly Used Documentation ModalitiesModalityStrengthsLimitationsApproximate Annual Cost*Manual documentationProvides complete control and accuracy over clinical notesTime-consuming, detracts from patient engagement, and increases administrative burden$40,371Voice-to-text servicesQuickly converts spoken words into written text and can adapt to structured note templatesAccuracy decreases with accents, background noise, or complex terminology$600–$1,200In-person medical scribeReduces physician documentation burden and allows greater patient focusScribes often have limited medical training and require coaching and quality oversight$40,000Virtual medical scribeProvides similar benefits to in-person scribes by reducing documentation workload and improving patient interactionRequires coaching and oversight; limited medical training may impact quality$38,000AI-assisted servicesPassively capture encounters, generate comprehensive notes in real time, and significantly reduce clerical burdenRaise privacy concerns and depend on high-quality audio for accuracy$700–$2500*See Supplemental Digital Content 1 (https://links.lww.com/PRSGO/E966). Comparison of Commonly Used Documentation Modalities See Supplemental Digital Content 1 (https://links.lww.com/PRSGO/E966). AI-powered tools, also known as ambient documentation technologies (ADTs), can further enhance documentation efficiency. ADTs passively capture patient encounters and generate comprehensive clinical notes in real time.3 Hybrid record-and-review systems record patient visits, then use AI to generate transcripts and draft notes for later review and approval. ADTs process historical or batch audio files into structured notes.10 BENEFITS OF AI-DRIVEN CLINICAL DOCUMENTATION The emergence of ADTs has quickly demonstrated their ability to reduce administrative burden and improve workflow, with recent studies reporting marked, measurable (up to 51%) reductions in documentation time during and outside of business hours.2,11 The increased efficiency allows clinicians to dedicate more time to patient care, enhancing healthcare delivery. For example, 39% of surveyed patients reported spending more time speaking directly with their physician, and the majority of patients reported a positive impact on the quality of their visit.11 Similarly, surgeons piloting ADTs report greater clinic efficiency, patient satisfaction, and quality of life, and reduced burnout.12–15 AI integration within EHR systems and partnerships with third-party vendors have further allowed the ability to generate referral letters, patient education documents, consent forms, insurance denial appeals, and orders based on the clinical encounter.16–18 AI can improve the readability of medical documentation, including clinical summaries and discharge recommendations, thereby enhancing patient communication and satisfaction.19 Many ADT platforms also produce current procedural terminology codes from encounter notes.20,21 A 2024 systematic review demonstrated that AI tools can potentially lead to fewer denied claims by using language that aligns with coding, billing, and reimbursement requirements.5 Furthermore, AI can improve the quality of care by highlighting inconsistencies in patient notes, alerting clinicians to incomplete follow-up recommendations, and flagging actionable test results and incomplete screenings.19 In the context of plastic surgery, high volume of consults, procedure-heavy clinics, and clinic documentation offer potential utility for ADT. Addressing economic concerns, the average annual medical scribe salary is around $40,000. 22–24 In contrast, current ADTs reportedly cost between $90 and $600 per user each month. 6,22,25 This highlights the economic advantage of adopting AI scribes, with particular relevance for physicians and surgeons in smaller groups or private practices, where overhead expenses are borne directly by clinicians and where throughput and efficiency are critical to solvency. (See appendix, Supplemental Digital Content 1, which displays the cost calculations, https://links.lww.com/PRSGO/E966.) Collectively, ADTs hold the potential to improve clinical workflows and patient and physician experience in clinical interactions, while reducing administrative costs. However, these benefits must be weighed against potential risks. RISKS AND CHALLENGES The use of ADT in healthcare introduces potential systemic vulnerabilities, including inaccuracies in documentation, data privacy and security, and clinician overreliance. A recent study evaluating the accuracy of LLMs on clinical note generation demonstrated a common hazard of AI: generation of factually incorrect information, or “hallucinations.” This study observed that of the 12,999 sentences produced by the AI platform, 191 (1.4%) contained hallucinations, 84 of which had potential impacts on patient diagnosis and management.26 Another study evaluating clinical notes from 2 ambient digital scribes reported that 70% of notes included errors, with an average of 3 errors per note.27 The accuracy and consistency of AI stem from its training on large amounts of data; therefore, it is important to understand that accuracy declines with novel and complex concepts.6 Errors in AI-generated notes may be due to inherent limitations regarding the ability to acknowledge and understand subtle nonverbal cues. Similarly, speech recognition software used by AI scribes has reduced accuracy when transcribing patients with accents or different dialects, resulting in higher error rates.28 Errors can be embedded in fluently written notes, making them difficult to detect. Yet, these inaccuracies can have clinical effects.6 Therefore, physicians must carefully check AI-generated clinical notes to mitigate liability and promote patient safety. The privacy and security of stored patient information is another important risk. Large data repositories increase the risk of unauthorized access and breaches. Furthermore, recent increases in cyberattacks have made comprehensive safeguards essential to protect patient health information.5,19 Finally, there is potential for physicians to overrely on such technologies. Reliance on AI scribes may amplify documentation errors and increase the risk of adverse events, similar to the issues caused by the expanded use of copy-paste and copy-forward functions.29 Currently, it is unclear whether AI-based charting errors would result in liability for AI companies, practitioners, or both. These concerns emphasize that without rigorous oversight and transparent use boundaries, these tools may inadvertently compromise the accuracy, security, and efficacy of clinical practice. Risks must be carefully weighed against the potential benefits, underscoring the need for cautious and deliberate implementation. Still, as LLMs continue to improve, their innovation offers promise in improving healthcare delivery. REVIEW OF POPULAR AI TRANSCRIPTION SERVICES The increased popularity of ADTs has resulted in the rapid emergence of new platforms. However, the integration and applicability of these tools vary in terms of compatibility, cost, and workflow integration. The following section outlines 4 platforms commercially available in the United States, detailing their key benefits, limitations, and overall fit for clinical use (Table 2). Additionally, an overview of Epic’s new AI tools, which are said to be released in early 2026, is included. All information regarding the functionality of specific ADT platforms was sourced from the platforms’ respective websites or related press releases.16–18,30,31 All 4 ADT platforms share common features, including the use of ambient AI to transcribe patient–clinician conversations and automatically generate a clinical note. A comparison of other common features is shown in Table 3. Table 2.Comparison of Commonly Used AI Transcription ServicesSystemCompanyStrengthsLimitationsTranscripts/Audio StorageCostAbridgeAbridgeDeep EHR integration, up to date partnership, partnerships to develop prior-authorization technologyRequires training and installation; more expensive comparative to Heidi and FreedAll recordings are deleted after 30 dConsultation requiredDragon CopilotMicrosoftEnables real-time speech prompting and task automation, deep EHR integrationRequires training and installation time; more expensive compared with Heidi and FreedAudio stored for up to 90 d; transcripts retained for 180 dPer-user or flexible plans available; price upon consultationHeidiHeidiEasy to use with a simple user interface, does not require significant training, free plan option with unlimited note generationFree plan limits customizable features (10 template actions per month); advanced features require a paid planAudio is deleted immediately following patient encounter; physician specifies duration for transcript storageFree basic tier; Pro plan costs $99/mo or $700 annuallyFreedFreedEasy to use with a simple user interface, does not require significant training, free plan option with unlimited note generationDoes not automatically generate billing codes and EHR integration requires additional feesAudio is automatically deleted after note generation, and quality checks are complete. Transcripts can be manually deleted at any time or set to be deleted after 30 dIndividual plans cost $90/mo with group pricing for 10+ practices; discounted rates for students and residentsEpic AI toolsEpicWill have the ability to assist with patient scheduling, provide ambient documentation, and aid revenue cycle management and codingStill in development, not expected until early 2026Unknown at this timePricing not yet available Comparison of Commonly Used AI Transcription Services Table 3.Comparison of Common Features Provided by ADT ServicesAbridgeMicrosoft Dragon CopilotFreed Base PlanFreed Upgraded PlanHeidi Free PlanHeidi Paid PlanAvailable through web browser, desktop, or mobile applicationYesYesYesYesNoYesMultilanguage translationYesYesYesYesYesYesEHR linkingYesYesNoYesNoYesEHR integrationYesYesNoNoNoNoGenerates billing codesYesYesNoNoYesYesCustomizable templatesYesYesYesYesLimitedYesHIPAA compliantYesYesYesYesYesYesEHR linking refers to the ability to automatically send notes to or populate notes in the patient’s EHR following a completed clinical encounter; EHR integration describes embedded functionality such as task automation and note summaries. Comparison of Common Features Provided by ADT Services EHR linking refers to the ability to automatically send notes to or populate notes in the patient’s EHR following a completed clinical encounter; EHR integration describes embedded functionality such as task automation and note summaries. Abridge Abridge emerged as one of the first major ADT platforms and is one of the most widely used in the United States. Transcripts are stored via US-based, Health Insurance Portability and Accountability Act (HIPAA)-compliant servers, and recordings are deleted after 30 days.32 Notes can be populated directly in Epic for review and editing and include the generation of billing codes. Abridge includes a linked evidence feature, which enables clinicians to verify the origin of information in the note by referencing the original transcript and, in some cases, reviewing direct audio from the encounter. Abridge is best suited for enterprise-level adoption in large hospital systems. Pricing and cost estimates require direct consultation. In 2023, the platform was added to the Epic Workshop (formerly known as Partners and Pals), which is reserved for third-party developers within Epic.33 Additionally, more recent features include a partnership with Wolters Kluwer Health to integrate UpToDate into the Abridge system.34 Abridge has also partnered with Highmark Health to develop prior-authorization technology, aiming to provide approvals at the point of conversation.35 Pilot studies report significant improvements in documentation efficiency, cognitive task load, and time spent focused on patients.36,37 Microsoft Dragon Copilot Microsoft Dragon Copilot is a more recent ADT. Released in May 2025, Dragon Copilot combines the speech-to-text services of Dragon Medical One and the AI ambient scribe technology of Dragon Ambient Experiences. A unique feature of this technology is the ability to prompt the ADT through speech, allowing it to understand and respond in real time to voice commands. Furthermore, it can query existing notes through speech prompting and generate billing codes. This software also provides task automation prompted by clinician–patient encounters, including creation of note summaries, referral letters, after-visit summaries, and order preparation. All data are stored in US-based, HIPAA-compliant servers. Audio recordings are stored for up to 90 days, and note transcripts are retained for 180 days.38 Access to Microsoft Dragon Copilot is obtained through a Microsoft licensing agreement. The Dragon Copilot per-user license offers the full Dragon Copilot capabilities. The flex plan offers fewer features, and generative AI capabilities can be added as a pay-as-you-go option on an as-needed basis.39 Pricing and cost estimates require consultation with the Microsoft Licensing team. Plastic surgery residents using Dragon Copilot reported significantly reduced documentation time and more time for wellness, case preparation, in-service examination preparation, and learning.13 Heidi Heidi AI, founded in 2019, is an ADT that can be used in the inpatient, outpatient, and telehealth settings. Following note generation, Heidi can source related billing codes based on the clinical details documented. Notes can be reviewed, edited, and manually copied into EHR systems. Heidi’s consult recordings are transcribed in real time, and audio is deleted immediately following the patient encounter, never stored, and cannot be accessed later. Transcripts for note generation may be retained for a specified period or deleted, as determined by the physician. Transcript data are stored for auditing in secure servers based in the region where the data originates, and physicians can specify the duration for which they will be stored before deletion. Basic EHR linking is available for both enterprise and group plans, allowing notes to be sent directly to the EHR. The main benefit of this platform is that the free plan includes unlimited note generation; however, it offers limited features, such as custom and document templates. On the free plan, physicians are allowed 10 actions per month, where they can customize or use premade templates. Documents include referral letters, patient explainers, or consent forms. The pro (paid) version unlocks unlimited custom and document templates and is priced at $99 per month or $700 annually per user. In pilot studies simulating orthopedic ward rounds, Heidi significantly reduced documentation time for progress notes and discharge summaries.40 Freed Freed AI, founded in 2022, is another ADT. The platform can summarize patient information from previous notes, and following an encounter, clinical note generation typically takes between 1 and 5 minutes. Freed stores its encrypted data within Microsoft Azure’s secure centers. Patient audio recordings are temporarily saved for quality checks before being automatically deleted. Transcriptions and note summaries can be deleted by physicians at any time or set to be deleted after 30 days. Additionally, Freed enables recording sessions using an external device and uploading the saved audio file for processing into a clinical note. Freed shines for its ease of use, with a simple user interface that decreases onboarding time. However, it lacks some more advanced features, such as automatic generation of billing codes. Freed is a paid service with a flat rate fee of 90$ per month for individuals and solo practices, with group pricing customized for 10+ physician practices. Additional fees of $49 per month are required for EHR linking, which enables a 1-click transfer of notes into the patient’s EHR. Discounted pricing is offered to students and residents. This platform is recommended for individual physicians, telehealth practices, and residents or students. Epic’s Emmie, Art, and Penny In August 2025, Epic announced that it was developing its own AI-powered clinical documentation tools. The first tool, “Emmie,” will serve as a virtual assistant to help patients schedule visits, answer appointment-related questions, and suggest relevant screenings for discussion with their doctor. The “Art” assistant will act as the clinician’s digital colleague, anticipating what is needed for the visit by sourcing the information from the patient’s medical records and creating tailored summaries for the physician. The Art assistant will incorporate ambient AI technology to generate medical documentation from patient encounters, orders, clinical notes, and referrals, to be reviewed and signed by the clinician at the end of an encounter. Epic is partnering with Microsoft to further facilitate the transcription services, using Microsoft’s ambient AI technology to develop its platform. This new technology is expected to be available in early 2026. Finally, an AI assistant called “Penny” is being designed to assist with revenue cycle management and other administrative tasks, such as generating insurance appeal letters and medical coding. PRACTICAL CONSIDERATIONS FOR PHYSICIANS When adopting ADT, physicians should consider the benefits and logistical challenges in the context of their own practice, accounting for practice size, liability, and specialty-specific needs. Independent or group practices, primarily in outpatient or clinic settings, may benefit from user-friendly tools such as Heidi and Freed, which offer intuitive interfaces and the ability to quickly generate notes, but lack deep EHR integration. These tools are simple and easy to use and provide utility for physicians experiencing documentation fatigue or lacking access to scribes. Physicians employed in larger hospital systems may benefit more from services such as Abridge or Microsoft Dragon Copilot, which integrate deeply into EHR systems and offer broader functionality. Another important consideration is liability, as physicians remain legally responsible for the content of AI-generated notes. To mitigate liability, physicians should procure a Business Associate Agreement with vendors, explicitly outlining that it is the vendor’s responsibility to mitigate any data breach, notify involved parties, and answer to the US Office of Civil Rights.41 For most ADT platforms, transcripts and recordings are typically uploaded to a secure cloud-based platform and may be stored for a period of 30–180 days.32,38 Saved transcripts can be referenced in malpractice lawsuits and may influence future insurance payouts.6 As good practice, written informed consent should be obtained before using any AI scribe. Freed AI includes a recommended script for explaining its purpose and obtaining patient consent.6 Hospital-employed physicians should clarify institutional protocols and determine whether the use of AI-based tools is supported by their institution. Although our review is limited to US products, ADTs are rapidly becoming available in other countries, and rules and legislation regarding the use of AI tools in healthcare vary globally. Physicians should also consider the specialty-specific limitations of ADTs. Despite the evolution of ADTs with specialty-specific tools, these platforms still struggle to understand certain terminology and context.6,42 In plastic surgery, flap terminology (free versus pedicled; deep inferior epigastric perforator versus transverse rectus abdominis myocutaneous flap), complex anatomy, and operative sequencing may be unfamiliar to LLMs, increasing error risk. Errors involving procedure laterality or flap viability can have detrimental consequences. Furthermore, due to the complexity of operative notes, LLMs show higher omission rates and lower clinical acceptability compared with operative notes written by physicians.43 Selecting among available solutions involves a careful balance of practice-specific needs and vendor capabilities, including practice volume, cost structures, and specialty-specific vocabulary. FUTURE OF AI IN DOCUMENTATION The integration of AI in medicine has accelerated over the last decade, with applications primarily focused on medical documentation and automating administrative tasks. The race to control the AI industry has driven strategic partnerships and product development aimed at maintaining relevance and novelty in AI-driven documentation.44–46 Microsoft and Abridge have partnered with EHR systems, such as Epic, enabling codevelopment of new technologies.44–46 This trend toward EHR-integrated AI tools puts pressure on smaller startups that lack such integration. However, the rapid development of novel functions has led to new features in development, including AI-assisted prior-authorization technology.35 Prior authorization often causes delays in care or coverage due to issues with clinical documentation.35 AI scribes can streamline this process by synthesizing the required information that insurers need and generating the necessary forms following the encounter. Additional emerging trends include the use of AI in preoperative planning and surgical education.47 With increasing EHR integration, future functionality will include analyzing vast amounts of patient data for quality improvement, clinical research, and preoperative risk stratification.44 As ADT is increasingly implemented into clinical practice, there is a growing need to educate physicians, residents, and medical students on the responsible use of these tools. Future ADT development should incorporate specialty-specific training focused on high-risk terminology, error-prone contexts, and clinically consequential phrasing. Finally, we encourage the plastic surgery community to take on an active role in piloting and evaluating these technologies, as representation is critical when training these models. CONCLUSIONS Increased documentation burden and high rates of physician burnout have led to the widespread adoption of AI transcription services, which hold promise for enhancing clinical documentation and improving efficiency in administrative tasks. Although these tools may enhance physician workflow, these benefits must also be weighed against the risks for both patients and providers. The emergence of novel technologies underscores the importance of specialty-specific training and standardized guidelines to ensure physicians can continue to deliver safe, effective, and comprehensive care. DISCLOSURE The authors have no financial interests to declare in relation to the content of this article. Supplementary Material