Interviews with clinicians about an ambient artificial intelligence documentation platform.

Interviews with clinicians about an ambient artificial intelligence documentation platform - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice J Am Med Inform Assoc . 2026 Feb 10;33(4):821–826. doi: 10.1093/jamia/ocag021 Search in PMC Search in PubMed View in NLM Catalog Add to search Interviews with clinicians about an ambient artificial intelligence documentation platform Cheryl D Stults Cheryl D Stults , PhD 1 Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States Conceptualization, Formal analysis, Investigation, Methodology, Supervision, Validation, Writing - original draft, Writing - review & editing Find articles by Cheryl D Stults 1, ✉ , Meghan C Martinez Meghan C Martinez , MPH 2 Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States Data curation, Formal analysis, Project administration, Validation, Writing - review & editing Find articles by Meghan C Martinez 2 , Nina K Szwerinski Nina K Szwerinski , MS 3 Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States Data curation, Formal analysis, Project administration, Validation, Writing - review & editing Find articles by Nina K Szwerinski 3 , Naveed Rabbani Naveed Rabbani , MD 4 Clinical Informatics, Digital Division, Sutter Health, Sacramento, CA 95833, United States Investigation, Validation, Writing - review & editing Find articles by Naveed Rabbani 4 , Veena G Jones Veena G Jones , MD 5 Clinical Informatics, Digital Division, Sutter Health, Sacramento, CA 95833, United States Funding acquisition, Investigation, Validation, Writing - review & editing Find articles by Veena G Jones 5 Author information Article notes Copyright and License information 1 Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States 2 Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States 3 Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States 4 Clinical Informatics, Digital Division, Sutter Health, Sacramento, CA 95833, United States 5 Clinical Informatics, Digital Division, Sutter Health, Sacramento, CA 95833, United States ✉ Corresponding author: Cheryl D. Stults, PhD, Center for Health Systems Research, Sutter Health, 795 El Camino Real, Ames Building, Palo Alto, CA 94301, United States ( cheryl.stults@sutterhealth.org ) Roles Cheryl D Stults : PhD , Conceptualization, Formal analysis, Investigation, Methodology, Supervision, Validation, Writing - original draft, Writing - review & editing Meghan C Martinez : MPH , Data curation, Formal analysis, Project administration, Validation, Writing - review & editing Nina K Szwerinski : MS , Data curation, Formal analysis, Project administration, Validation, Writing - review & editing Naveed Rabbani : MD , Investigation, Validation, Writing - review & editing Veena G Jones : MD , Funding acquisition, Investigation, Validation, Writing - review & editing Received 2025 Aug 8; Revised 2026 Jan 21; Accepted 2026 Jan 25; Collection date 2026 Apr. © The Author(s) 2026. Published by Oxford University Press on behalf of the American Medical Informatics Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial License ( https://creativecommons.org/licenses/by-nc/4.0/ ), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact reprints@oup.com for reprints and translation rights for reprints. All other permissions can be obtained through our RightsLink service via the Permissions link on the article page on our site—for further information please contact journals.permissions@oup.com. PMC Copyright notice PMCID: PMC13089541  PMID: 41666081 Abstract Objective Understand the qualitative impact of an ambient artificial intelligence (AI) documentation platform on clinicians’ experiences and workflows. Materials and Methods A quality improvement (QI) qualitative study using semi-structured interviews after pilot implementation of an ambient AI documentation platform at a large healthcare organization in Northern and Central California. Pragmatic thematic analysis was used to code and analyze the interviews. Results 100 clinicians were invited and 42 (42%) participated in an interview. 23 (54.8%) were males and 28 (66.7%) in primary care. Many respondents noted that ambient AI had decreased their cognitive burden by eliminating the need to remember as many specific visit details and saved time for other tasks. They also liked how ambient AI generated transcripts in multiple languages and then created an English-language progress note. However, clinicians also reported challenges, particularly missed or inaccurate information that required them to review the transcript/audio and edit the note. Additionally, many clinicians, particularly specialists, disliked the note formatting and the inability to customize the note template as this resulted in additional manual editing. Discussion Results from these QI qualitative interviews suggest that ambient AI improved clinicians’ overall experience at work. Conclusion While there are many similarities, there may be key differences in clinician experience between ambient AI documentation platforms depending on unique features of each. Future research is needed to understand the potential range of experiences based on type of ambient AI platform and if these findings continue with longer use and broader expansion of this new and evolving technology. Keywords: qualitative evaluation, generative artificial intelligence, physician, electronic health records Background Ambient artificial intelligence (ambient AI) documentation programs, which use generative AI to capture clinical visits and generate progress notes for clinician review, 1–3 are increasingly being adopted by health systems to improve clinicians’ efficiency and overall well-being. 4–6 Ambient AI records patient-clinician conversations, generates a transcript, and drafts portions of the clinical note that clinicians can edit and finalize. Several studies have reported benefits of ambient AI based on survey data, 2 , 3 , 7–9 while findings from electronic health record (EHR) data have been mixed regarding time savings and satisfaction, 9–12 with differences between vendors. 13 A few analyses of open-ended survey comments of clinician experience with ambient AI have highlighted benefits such as time savings, 9 decreased cognitive load, 9 , 14 and improved patient connection during visits, 9 , 14 but they have also noted some challenges regarding note quality, with some reporting it to be “overly error prone,” 9 and difficulties with specialty-specific content, particularly in the physical exam section. 14 Only a few have provided additional details about the qualitative clinician experience through in-depth conversations and in these, clinicians reported favorably on ambient AI’s ability to help summarize long appointments, reduce cognitive burden, and improve patient interactions. 11 , 15 However, given the limited qualitative research to date and that there are several different ambient AI platforms, there has been a call for more in-depth interviews to better triangulate and understand quantitative findings 12 and experiences across ambient AI platforms. Our study addresses this gap through analysis of numerous semi-structured clinician interviews to examine their experiences using ambient AI in a large healthcare system and inform future development of these tools to better support clinical work. Methods Sutter Health is a large, not-for-profit health care organization serving approximately 3 million patients across Northern and Central California. Sutter launched an initial pilot of the ambient AI platform, Abridge, on April 10, 2024 with 100 pilot clinicians who were clinical leaders or medical informatics champions. These participants represented both primary and specialty care across Sutter’s geographic regions. After at least one month of use, pilot clinicians were invited to complete a post-implementation survey which included the option to provide contact information if they were interested in participating in a follow-up interview. Additional details about the survey and administration processes have been previously published. 14 Clinician demographics were collected from the EHR and human resources. Sutter Health’s Institutional Review Board reviewed and approved all study materials and determined this evaluation to be quality improvement (QI). We followed Standards for Reporting Qualitative Research (SRQR) reporting guidelines. We reached out via email to all clinicians who provided contact information on the survey. A total of 47 clinicians initially expressed interest. Of these, 42 clinicians engaged with the research team and participated in phone/video interviews conducted by 2 research team members. The remaining 5 clinicians were contacted through 3 outreach attempts but were not available. Prior to the start of each interview, clinicians were informed that participation was voluntary, only de-identified information from the interview would be shared, and verbal consent was obtained for both the interview and audio recording using digital recorders. Our qualitative evaluation team was multi-disciplinary: 3 health services researchers (C.D.S., M.C.M., N.K.S.) and 2 physicians and clinical informaticists (N.R. and V.G.J.) (Author initials). Three qualitative researchers (C.D.S., M.C.M., N.K.S.) developed a semi-structured interview guide in collaboration with the ambient AI clinical implementation lead (V.G.J.) to ensure that the interviews captured important aspects of the ambient AI experience. Because this was a QI pilot, we also wanted to capture any recommendations clinicians might have for further expansion of the technology across the system (see Supplementary File S1 for Interview Guide). Interviews ranged from 15 to 60 minutes (average 30 minutes) and were conducted primarily by M.C.M. and N.K.S., with C.D.S. only doing a few. C.D.S. was previously familiar with 2 clinicians included in the study; however, she did not participate in recruitment, and no interviewer conducted an interview with a clinician they personally knew. Data collection occurred between May 2024 and August 2024. Audio recordings were transferred from the digital recorders to secure folders on the Sutter system. They were then sent via secure transfer to a HIPAA-compliant agency for transcription with the resulting transcripts returned through the same secure process. The research team also stored the transcripts on secure Sutter network folders, de-identified them, and imported them into an online qualitative software analysis program, Dedoose (version 9.0), for coding. Given our multidisciplinary team of health services researchers and clinicians, we employed practical thematic analysis, 16 following 3 steps: reading, coding, and theming. The main analysis was done by M.C.M. and N.K.S., who are Project Managers, and C.D.S., one of the principal investigators, and all 3 have extensive expertise in qualitative methods. To minimize any inherent power dynamics within the research team and minimize their impact on coding and interpretation, the 2 project managers conducted most interviews and led the coding process, reaching consensus among themselves with minimal input from the principal investigator or clinician leaders. During analysis, coders were blinded to participant identities to further reduce potential bias from prior familiarity. The initial draft of the codebook was collaboratively developed by M.C.M. and N.K.S. based on insights from previous coding of write-in survey responses (eg, cognitive burden was a common response). All 3 qualitative team members (M.C.M., N.K.S., C.D.S.) then read through the interview transcripts and refined the codebook to reflect the more detailed information provided during interviews. To validate the revised codebook and ensure inter-coder reliability, M.C.M. and N.K.S. independently coded 2 full transcripts and met to compare interpretations, resolve discrepancies, and establish consistency before proceeding to code the full dataset. The final codebook has been included as Supplementary File S2 . The codebook was finalized and no new codes were added after approximately half of the transcripts had been coded. Once coding was completed, C.D.S. reviewed the codes and grouped them into preliminary themes. C.D.S., M.C.M., and N.K.S. met several times and finalized theme construction, identifying accompanying illustrative quotations. After themes had been compared across primary care, medical, and surgical subspecialties, we determined thematic saturation had been reached and no additional interviews were needed. The final themes and quotations were checked and discussed with clinical leaders (N.R. and V.G.J.) to confirm their meaning and validity. Results We conducted 42 interviews. Clinician respondents were primarily female (19, 45.2%), White (29, 69.0%), non-Hispanic (34, 81.0%), and most were primary care providers (28, 66.7%). The mean age was 50.5 years (10.7 SD) ( Table 1 ). These characteristics are not representative of the broader clinician population at Sutter Health as participants were largely clinical leaders or medical informatics champions. Table 1. Characteristics of interviewed clinicians in ambient AI pilot. Characteristics No. (%) Interviews ( n = 42) Sex Female 19 (45.2%) Male 23 (54.8%) Race a Asian 6 (14.3%) White 29 (69.0%) Multiple race 1 (2.4%) Unknown 6 (14.3%) Ethnicity a Non-Hispanic 34 (81.0%) Hispanic 2 (4.8%) Unknown 6 (14.3%) Age b Mean (SD) 50.5 (10.7) Median (IQR) 51 (43, 57) Type Physician 40 (95.2%) Advanced practice clinician (APC) b 2 (4.8%) Specialty Primary care c 28 (66.7%) Medical subspecialty d 7 (16.7%) Surgical subspecialty e 7 (16.7%) Years of service f Mean (SD) 13.3 (7.5) Median (IQR) 14 (16.5-7) Clinical FTE FTE < 0.5 5 (11.9%) 0.5 ≤ FTE < 0.7 15 (35.7%) 0.7 ≤ FTE ≤ 1 18 (42.9%) Unknown 4 (9.5%) Open in a new tab a Race and ethnicity were obtained from the EHR and human resources. b includes NP and PA c Includes internal medicine, family medicine, pediatrics, and urgent care. d Includes neurology, hematology/oncology, cardiology, gastroenterology, rheumatology, dermatology, psychiatry, pulmonology, endocrinology/metabolism. e Includes podiatric medicine/surgery, interventional cardiology, orthopedic surgery, otolaryngology, OBGYN, neurosurgery, anesthesiology, plastic surgery, urology. f 3/42 with missing years of service. Table S3 summarizes key themes from clinician respondents, highlighting variation in experience, particularly between specialists and primary care providers. Five themes were identified in the thematic analysis, and the sections below discuss each in greater detail. Mental and physical strain from documentation Many clinicians reported that ambient AI significantly decreased their mental cognitive burden by removing the need to remember as many specific details, particularly during “complex visits” where multiple issues were discussed. With ambient AI, one clinician described how they were able to remain focused on the patient without needing to stop and take notes: “[the patient] can shoot as many complaints as they want. It doesn’t matter to me because, cognitively, I can do everything. I can give them all the recommendations” (C10, Primary Care). Some clinicians noted that ambient AI helped to overcome the “dread” of charting after particularly “difficult” visits, such as those involving patients struggling with their mental health. However ambient AI did not eliminate all cognitive burden, as some clinicians reported that reviewing and editing generated notes still required mental effort. In addition to the overall cognitive ease provided by ambient AI, clinicians also liked being able to type less during visits, which they felt improved both physical strain and patient connection. One clinician described how they no longer had to “frantically type,” while another noted how the reduction in repetitive strain from typing had helped their carpal tunnel and arthritis. Many clinicians also mentioned that typing less during the visit improved the quality of the experience with patients. Several specifically mentioned that instead of typing, they can now make more eye contact with the patient: So I would type a lot in the room because there’s just kind of no way to not be efficient. To be efficient, you’ve got to type in the room, you’ve got to be doing stuff. So now I really do just find myself just looking at the families and making eye contact, and only going to the computer if I’m, like, placing an order or, you know, throwing up the growth chart or, you know, doing something like that. So the experience that I have in the exam room is nicer. (C44, Primary Care) Another clinician said that with ambient AI, they mainly use the computer to show information to the patient, like lab results. Impact on time Many shared how ambient AI saved time so they could focus on other things, both in their clinical and personal responsibilities. Several clinicians described how ambient AI resulted in them having more time during the day, which they used to complete in-basket work or, in some cases, to see more patients. Others talked about how they now have time and energy to focus on personal well-being, whether in the form of exercise or time with family: For me, one of the reasons I’m the most happy is I want to be able to see more patients, and I can. And my colleagues are worried about me. They’re like, you’re seeing 30 patients. You’re going to burn out. I was like, I’m not. I’m totally happy, and I’m leaving the day full of energy. I’m going to go to the gym and workout. You don’t have to worry about me. I’m worried about you…You’re seeing like 17 patients, but how are you seeing 17 patients without AI? That’s gotta suck. You have to do all this yourself. I’m worried about you. (C10, Primary Care) Note format and sections: history of present illness, after-visit summary, physical exam Many respondents described how ambient AI improved specific sections of the note, particularly the History of Present Illness (HPI) and patient instructions (used to populate the after-visit summary, AVS). Many clinicians, both primary care and specialists, appreciated how well ambient AI captured and summarized the patient history, like organizing the note by topic rather than simple chronological order: I’ve been using the short HPI, and I think it does a good job, like, doing a synopsis. My patients tend to have more than one problem when they come in, so they can talk about many different things, and it [ambient AI] does a good job of synthesizing it [what they say] into coherent, cohesive sections about each individual problem. I like that it’s good at, instead of just repeating what the patient says, it [ambient AI] actually makes language that is… It’s good at adopting medical language and turning what patients say into succinct medical language. I like the assessment and plan in that it [ambient AI], like in the same way, is really good at synthesizing a bunch of different information into cohesive sections that are easy to follow. (C12, Surgical Subspecialty) However, some medical subspecialists mentioned challenges with HPI during new patient visits. Specifically, they reported ambient AI would not separate out key information into separate paragraphs or sections for previous medical history, for example, for neurology: motor functions, non-motor functions, current medications, or previous medications. Despite these concerns, one clinician observed that there had already been improvements with documenting information from family history in new patient visits over the short time between initial implementation and their interview. Several clinicians also described how well ambient AI generated patient instructions. One called ambient AI “amazing” because it summarized the medical jargon into plain language for patients to improve comprehension and overall patient satisfaction: now what Abridge does better than almost anything else, it does an amazing AVS. Because the after visit summary is usually a summary, so it summarizes the visit. It [Ambient AI] also puts it in a level of language that’s understandable to patients. So I think it does an amazing job writing an AVS. (C14, Surgical Subspecialty) Respondents noted how they thought that this contributed to a better patient experience as clinicians felt more comfortable listening and engaging during the visit, knowing that all relevant information would be captured for them and shared after the visit with patients in the AVS. Specialists in particular reported challenges with the physical exam section of ambient AI notes, pointing out that it was not able to capture important clinical details. One surgical subspecialist felt that “hallucinations” tended to happen more frequently with the physical exam and provided an example where the thyroid was examined but not mentioned in the note, and instead a complete head and neck exam was inserted in the note, yet never performed: I would say it [ambient AI] tends to hallucinate a lot of stuff for physical exams. Like, it will make up things I didn’t say about certain areas. Like, it [ambient AI] will know I’m talking about a certain area of the body and then, like, formulate a normal exam for that whole, like, section of information without me having actually said anything about those things. And then it’ll also leave out the one thing I was mentioning. So for instance, like if I examine someone’s thyroid and I say, oh, you have a normal feel, like your thyroid feels normal, it’s a normal size. It [Ambient AI] will hallucinate a head and neck exam that does not include a thyroid exam, but includes everything else in the head and neck and says it’s normal. (C12, Surgical Subspecialty) Due to these inaccuracies, a few specialists noted that instead of ambient AI, they opted to use EHR templates to document the information in the progress note. Since the ambient note was “not very specific to neurology in general,” this neurologist elaborated that they didn’t “really see a utility for me to use AI to navigate and explain and have it all turn out the way I like it documented as it would in a template form from Epic” (C2, Medical Subspecialty). Other clinicians also mentioned feeling uncomfortable verbalizing every component of a physical exam, describing it as “awkward” ( Table S3 ). However, others found unexpected value in this practice. One initially skeptical clinician explained the benefit they found to the patient of speaking aloud during the physical exam: And then I have to put in the physical exam, and that was one of the things I really wasn’t planning on utilizing, was the physical exam in Abridge, because that means that I have to say out loud their exam, and that’s a little awkward. But you know. You know what? I’m going to play along. I’m going to try it. I’m going to say it out loud, and it’s been pretty natural. Okay. Yeah. Your lungs are clear. Oh, yeah. The ears look okay. No infection … And, it’s capturing all that, because I’m doing the exam anyway, and I’m verbalizing it out loud, so the patients are hearing this. And I know that there have been studies that, you know, patients appreciate that. (C1, Primary Care) Functionality with languages other than English Several mentioned how ambient AI was able to accurately translate conversations from other languages, even when more than one language is used during a single visit (eg, when a conversation switched between Hindi and Punjabi). The transcript preserved the original language characters, while the generated note was written in English: It[Ambient AI] doesn’t need a translator. So I can speak a couple languages like Hindi and Punjabi, and,, It [Ambient AI] gets it [translates text into English] … If they’re speaking Mandarin, and there’s a translator translating for them, it [ambient AI] does it [translates text into English] beautifully… it [ambient AI] actually writes the language that’s spoken in the native written way. (C10, Primary Care) Several respondents explained that they liked how ambient AI wrote and summarized the progress note. A few clinicians for whom English was not their first language felt ambient AI was able to compose a better note: “the language is better than what I would use, because English is not my first language. You know, English is my second language, so sometimes I feel like, oh, Abridge is writing an even better narrative than I would do” (C23, Medical Subspecialty). Amount of information In addition to information issues with the physical exam section of the note, many clinicians and nearly all specialists reported instances where ambient AI had “missed” or included inaccurate information, requiring the clinicians to go back and review and edit the transcript/audio. A few clinicians explicitly stated how the omission of some key details required them to go back in and add or modify as necessary for billing. This seemed to be a more frequent complaint among specialists. One medical subspecialist described how ambient AI did get the “gist” of the conversation, but did not include the specific number of actinic keratoses, which is necessary for billing, requiring the clinician to check the transcript later: When I would call out—for our notes, it’s really important if we say three actinic keratoses on the right cheek, four actinic keratoses on the hands, and it [ambient AI] would just summarize that to be actinic keratosis. I actually need the number for billing, and I couldn’t seem to get it to capture that. For me, that wasn’t helpful because if I wanted to keep moving, the reason I want to say that is because I may not remember an hour later. Was it three? Was it four? It [Ambient AI] wouldn’t capture that. I’d have to go back to the transcript and look. (C40, Medical Subspecialty) But, some primary care respondents felt differently, reporting that ambient AI increased accuracy with billing by documenting all topics discussed, even ones they might have previously forgotten, so that they could bill at a higher visit level ( Table S3 ). While many clinicians expressed that ambient AI omitted specific details, some felt that ambient AI provided too much information which made it difficult to find key components: “I think on the flip side of kind of adding more details, sometimes it [Ambient AI] could add too many details and make it harder to look back on previous notes and being able to isolate the information that I need” (C4, Primary Care). Another clinician expressed that so many additional details could be difficult for colleagues treating the patient in the future to find relevant information. Discussion This study is the first to present qualitative findings from a large set of in-depth interviews with clinicians using the Abridge ambient AI platform at a large healthcare organization. Similar to other qualitative and quantitative results across various ambient AI platforms, our respondents indicated that one of the main benefits of ambient AI was decreased cognitive burden. 9 , 14 , 15 However, there may be key differences among various ambient AI platforms as we found novel strengths of Abridge to handle visits in languages other than English and in documentation for the History of Present Illness and Patient Instructions sections. Two of the main challenges reported, especially by specialists, were the mental effort required for reviewing and editing documentation, particularly in relation to specialty-specific billing requirements, and the Physical Exam section. These challenges provide context into our previous quantitative findings 14 where specialists had lower overall satisfaction with their ambient AI experience. Given the rapidly evolving nature of AI, future versions may be able to better accommodate clinician preferences, reducing the need for such adjustments. One challenge mentioned in previous physician interviews 15 about ambient AI was the difficulty in using a translator for those with English as a second language. In contrast, we heard that one of the major benefits of Abridge ambient AI was generating transcripts in multiple languages and then creating an English-language progress note. This finding suggests that despite many similarities, there may be key differences in clinician experience between ambient AI documentation platforms depending on unique features of each, 3 as evidenced by a randomized controlled clinical trial which showed the ambient AI scribe platform, Nabla, was associated with a statistically significant decrease in documentation time, whereas another ambient AI scribe platform, DAX, did not. 13 Thus, further investigation is needed to understand the potential range of experiences based on features of each ambient AI platform. Like Shah, et al., 15 many respondents highlighted inaccuracies such as missed, incorrect, and “hallucinated” information included by the ambient AI program in the progress notes. While many respondents reported that these types of errors led to increased time spent editing the note, some also mentioned larger implications for billing and sharing information with other clinical colleagues. The U.S. Food and Drug Administration (FDA) raised concerns about “hallucinations” as they “can present the difference between summarizing a healthcare professional’s discussion with a patient and providing a new diagnosis that was not raised during the interaction.” 17 Thus, it is imperative that clinicians carefully review the progress note before signing their name to it. Clinicians using ambient AI should be aware of the potential for “automation bias” 18 , 19 where they might rely too much on the AI-generated content. As we heard from many respondents that they still needed to carefully review the ambient AI progress notes, health systems should recognize that the time savings for clinicians may be less than expected. Limitations Our interviews were conducted at a single organization, which may limit generalizability. We recognize that there may be potential bias from the pilot participants as they were leaders, champions, and informaticists, thus their experiences may not be fully representative of all clinicians. Additionally, there may be response bias in that those who were willing to speak with us might have had stronger opinions, both positive and negative, about their experience. There is also concern about social desirability bias, in that participants may have been more inclined to share what they believed the research team wanted to hear. To mitigate this, interviews were conducted by the qualitative researchers rather than clinical operations leaders, who may have been perceived as preferring to hear about successful aspects of implementation. Interviews were voluntary and unpaid, and clinicians participated by providing their email and best time to contact. We also began with open-ended questions to encourage clinicians to share both positive and negative experiences before asking about specific benefits and challenges. Given the large geographic area of our healthcare system, we conducted interviews over telephone and videoconference, and we recognize that the rapport and what they shared may have been different than in-person interviews. 20 However, by choosing to conduct all interviews remotely, rather than a hybrid of in-person and remote, we were able to maintain consistency as well as allowing for participation from all interested respondents and not just those who were geographically close to the interviewers. Conclusion This is one of the first analyses to describe clinicians’ experiences with Abridge ambient AI using a large number of in-depth clinician interviews within a major healthcare organization. Our findings suggest traditional surveys or EHR-based metrics may not fully capture the multitude of ways ambient AI impacts clinicians. Additionally, our results highlight challenges unique to specialists compared to primary care and so future work should further explore these differences. Given the rapidly evolving nature of ambient AI, several concerns mentioned by our respondents have already been addressed since the time of our interviews in summer 2024. One month after we finished these interviews, Abridge became fully integrated into Epic in September 2024, and over the past year Abridge has introduced physical exam templates that address some of the specialty-specific documentation challenges noted by respondents. As these systems continue to develop, it is important to continue to understand the ongoing experience of clinicians with this evolving technology. Supplementary Material ocag021_Supplementary_Data ocag021_supplementary_data.zip (74.4KB, zip) Contributor Information Cheryl D Stults, Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States. Meghan C Martinez, Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States. Nina K Szwerinski, Center for Health Systems Research, Sutter Health, Palo Alto, CA 94301, United States. Naveed Rabbani, Clinical Informatics, Digital Division, Sutter Health, Sacramento, CA 95833, United States. Veena G Jones, Clinical Informatics, Digital Division, Sutter Health, Sacramento, CA 95833, United States. Author contributions Cheryl D. Stults (Conceptualization, Formal analysis, Investigation, Methodology, Supervision, Validation, Writing—original draft, Writing—review & editing), Meghan C. Martinez (Data curation, Formal analysis, Project administration, Validation, Writing—review & editing), Nina K. Szwerinski (Data curation, Formal analysis, Project administration, Validation, Writing—review & editing), Naveed Rabbani (Investigation, Validation, Writing—review & editing), and Veena G. Jones (Funding acquisition, Investigation, Validation, Writing—review & editing) Supplementary material Supplementary material is available at Journal of the American Medical Informatics Association online. Funding Sutter Health funded the interviews, analyses, and manuscript, and is a strategic investor in Abridge (Pittsburgh, PA). The funding organization had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. Conflicts of interest The authors have no conflict of interests to declare. Data availability The data will be shared on reasonable request to the corresponding author and data sharing will need to comply to organizational agreements. References 1. Haberle T, Cleveland C, Snow GL, et al.  The impact of nuance DAX ambient listening AI documentation: a cohort study. J Am Med Inform Assoc. 2024;31:975-979. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Galloway JL, Munroe D, Vohra-Khullar PD, et al.  Impact of an artificial intelligence-based solution on clinicians’ clinical documentation experience: initial findings using ambient listening technology. J Gen Intern Med. 2024;39:2625-2627. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Albrecht M, Shanks D, Shah T, et al.  Enhancing clinical documentation with ambient artificial intelligence: a quality improvement survey assessing clinician perspectives on work burden, burnout, and job satisfaction. JAMIA Open. 2025;8:ooaf013. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Olson KD, Meeker D, Troup M, et al.  Use of ambient AI scribes to reduce administrative burden and professional burnout. JAMA Netw Open. 2025;8:e2534976. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Pearlman K, Wan W, Shah S, Laiteerapong N.  Use of an AI scribe and electronic health record efficiency. JAMA Netw Open. 2025;8:e2537000. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. You JG, Dbouk RH, Landman A, et al.  Ambient documentation technology in clinician experience of documentation burden and burnout. JAMA Netw Open. 2025;8:e2528056. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. MIsurac J, Knake LA, Blum JM.  The effect of ambient artificial intelligence notes on provider burnout. Appl Clin Inform. 2025;16:252-258. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. Shah SJ, Devon-Sand A, Ma SP, et al.  Ambient artificial intelligence scribes: physician burnout and perspectives on usability and documentation burden. J Am Med Inform Assoc. 2025;32:375-380. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Duggan MJ, Gervase J, Schoenbaum A, et al.  Clinician experiences with ambient scribe technology to assist with documentation burden and efficiency. JAMA Netw Open. 2025;8:e2460637. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Ma SP, Liang AS, Shah SJ, et al.  Ambient artificial intelligence scribes: utilization and impact on documentation time. J Am Med Inform Assoc. 2025;32:381-385. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Tierney AA, Gayre G, Hoberman B, et al.  Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catal Innov Care Deliv. 2024;5 (3):1–15. [ Google Scholar ] 12. Liu T-L, Hetherington TC, Dharod A, et al.  Does AI-powered clinical documentation enhance clinician efficiency? a longitudinal study. NEJM AI. 2024;1:AIoa2400659. [ Google Scholar ] 13. Lukac PJ, Turner W, Vangala S, et al.  Ambient AI scribes in clinical practice: a randomized trial. NEJM AI. 2025;2:AIoa2501000. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Stults CD, Deng S, Martinez MC, et al.  Evaluation of an ambient artificial intelligence documentation platform for clinicians. JAMA Netw Open. 2025;8:e258614. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Shah SJ, Crowell T, Jeong Y, et al.  Physician perspectives on ambient AI scribes. JAMA Netw Open. 2025;8:e251904. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Saunders CH, Sierpe A, von Plessen C, et al.  Practical thematic analysis: a guide for multidisciplinary health services research teams engaging in qualitative analysis. BMJ. 2023;381:e074256. [ DOI ] [ PubMed ] [ Google Scholar ] 17. U.S. Food Drug Administration. Executive summary for the Digital Health Advisory Committee meeting: total product lifecycle considerations for generative AI-enabled devices. November 20,2024. Accessed 03/14/2025. https://www.fda.gov/media/182871/download 18. Cohen IG, Ritzman J, Cahill RF.  Ambient listening—legal and ethical issues. JAMA Netw Open. 2025;8:e2460642. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Khera R, Simon MA, Ross JS.  Automation bias and assistive AI: risk of harm from AI-driven clinical decision support. JAMA. 2023;330:2255-2257. [ DOI ] [ PubMed ] [ Google Scholar ] 20. Johnson DR, Scheitle CP, Ecklund EH.  Beyond the in-person interview? How interview quality varies across in-person, telephone, and Skype interviews. Soc Sci Comput Rev. 2021;39:1142-1158. [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials ocag021_Supplementary_Data ocag021_supplementary_data.zip (74.4KB, zip) Data Availability Statement The data will be shared on reasonable request to the corresponding author and data sharing will need to comply to organizational agreements. Articles from Journal of the American Medical Informatics Association: JAMIA are provided here courtesy of Oxford University Press ACTIONS View on publisher site PDF (568.5 KB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Protocol for an EHR-embedded pragmatic randomized control trial of Ambient AI to Reduce Nursing Staff Documentation Time#441664
• Barriers and opportunities of scaling ambient AI scribes for clinical documentation across diverse healthcare settings.#186168
• Human-Centered Evaluation of Ambient Artificial Intelligence Clinical Documentation#983493
• Ambient AI Documentation in Mixed-Language Encounters: A Heuristic Evaluation of Spanish-English and Mandarin-English Conversations#296518
• Clinician Experiences with Ambient AI Scribe Technology in Singapore: A Qualitative Study#18859