The SOAP note (Subjective, Objective, Assessment, Plan) has been the structural backbone of clinical documentation for five decades. It was developed by Lawrence Weed in the 1960s as a way to give clinical records logical structure, and despite the shift to electronic health records, the format has remained largely intact. What has changed is the time it takes to produce a SOAP note and who bears that burden.
In 2025, the dominant workflow at most outpatient practices looks something like this: a physician sees 18 to 24 patients in a day, types notes between appointments when possible, finishes the rest during lunch, and completes the remainder after the last patient leaves. On a heavy day, charting extends into the evening. The EHR made many things better - legibility, medication reconciliation, lab integration - but it did not solve the fundamental problem that documenting a 15-minute patient visit takes 5 to 10 minutes of a physician's undivided attention.
Where ambient AI fits
Ambient AI documentation tools record the spoken conversation between a physician and patient during the visit, then use large language models to generate a structured clinical note in near real-time. The physician reviews, edits, and signs. The workflow change is meaningful: instead of typing a note after the visit, the physician reviews a draft during or immediately after it.
This is not voice dictation. Dictation tools - which have existed since the early 2000s - require the physician to verbally describe the note to a microphone, typically after the patient has left. Ambient AI listens to the actual clinical encounter, identifies medically relevant content, and structures it into SOAP or visit-note format without requiring the physician to narrate for the recorder.
The distinction matters because dictation still requires physician attention and time. Ambient generation happens in parallel with care delivery.
What has changed in 2025
Three things have meaningfully shifted the landscape this year.
First, language model performance on clinical text has improved substantially. Earlier ambient tools produced notes that required extensive editing - the subjective section was often accurate, but the assessment and plan sections needed significant physician input to be useful. More recent model generations handle complex multi-problem visits better, capture nuance in the assessment with greater reliability, and produce plan sections that are closer to what an experienced physician would write.
Second, EHR integration has become table-stakes for adoption. Standalone documentation tools that required copy-paste into the EHR never achieved sustained usage in clinical settings. Physicians were willing to try them but not to maintain a parallel workflow long-term. Direct FHIR-based integrations with major platforms have removed that barrier. Notes generated by ambient tools now push directly into the EHR note editor; the physician opens what appears to be a pre-populated note and makes edits rather than writing from scratch.
Third, the patient consent conversation has matured. Early ambient tools struggled with the consent workflow. Practices were unsure how to introduce recording to patients, what to say when patients declined, and how to handle sensitive visit types - mental health, reproductive health, domestic violence disclosures - where ambient recording is inappropriate. More practices have now developed standardized consent workflows, and the clinical community has more shared experience with which visit types are well-suited to ambient documentation.
What remains unchanged
Ambient AI does not diagnose. It documents what the physician says and does during the visit. This distinction is important and sometimes blurred in how these tools are described. The model generates a note based on what it hears; it does not independently assess clinical findings or recommend treatment. The physician's clinical judgment remains entirely in the loop.
Ambient AI also does not eliminate note review time. Physicians typically spend 1 to 3 minutes reviewing a generated note before finalizing it. This is much less than the 5 to 10 minutes required to write the note, but it is not zero. Practices that evaluate these tools expecting a complete elimination of documentation time tend to be disappointed; practices that expect a substantial reduction tend to be satisfied.
The subjective section - what the patient reports - is generally where ambient tools perform best. The objective section - physical examination findings - varies more, because it depends on whether the physician narrates findings aloud during the exam. Physicians who develop the habit of saying findings out loud ("lungs clear bilaterally, no wheezing") produce better objective sections than those who examine silently. This is a behavioral adaptation that most physicians make naturally within a few weeks, but it is worth noting that the tool performs better with physician narration than without it.
The outpatient-specific considerations
Ambient documentation has followed two parallel paths: hospital-based deployment through large health system contracts, and outpatient deployment through per-provider subscriptions to independent practices. The outpatient setting has distinct characteristics that make some ambient tools poorly fitted to it.
Outpatient visits are shorter (15 to 20 minutes on average), more varied in chief complaint, and higher in daily volume per provider compared to inpatient rounds. A primary care physician seeing 22 patients in a day is generating 22 separate documentation events, each requiring visit context that the tool must track independently.
Hospital-focused tools are often optimized for longer, more complex notes - inpatient history and physicals, discharge summaries, progress notes on complex patients. These differ structurally from an outpatient SOAP note for a patient presenting with hypertension follow-up or an acute upper respiratory infection. Outpatient physicians evaluating ambient tools should confirm that the tool was trained on outpatient visit audio, not predominantly on inpatient clinical text.
What to watch in 2026
The next phase of ambient documentation will likely involve tighter integration between the documentation layer and the clinical decision support layer in EHRs. Tools are beginning to flag coding opportunities, suggest missing HCC diagnoses, and identify documentation gaps that could affect reimbursement during the same workflow where notes are generated. Whether this creates value or cognitive load is a question practices will answer in the coming 18 months.
There is also an ongoing conversation about note compression. Ambient AI tends to produce longer notes than physician-typed notes, because it captures more of the spoken visit. This can be valuable - more complete documentation - or it can be noise, adding length without clinical value. The pressure to produce shorter, denser notes that satisfy coding requirements without reading like a transcript is driving some interesting model development.
For independent outpatient practices, the practical question in 2025 is simpler than the industry conversation suggests: does this tool reduce my documentation burden, can I integrate it with my EHR, and can I trust the data handling? Those three questions, answered well, define the current state of the market.