The new-patient bias hiding inside most ambient scribes
Ambient documentation has spent the last two years being benchmarked, marketed, and demonstrated almost exclusively on the new patient visit, which happens to be the encounter most flattering to a transcription model. The history of present illness is verbalized at length, the review of systems is methodical, the physical exam is narrated in full, and the assessment plan is dictated as if the chart is a blank page. Anyone evaluating an ambient scribe on those terms will conclude that the technology is essentially solved, which is precisely the conclusion that vendors are happy to encourage. The follow-up visit, however, is where the conversational signal collapses, where shorthand replaces narration, and where the model has to know things that were never spoken aloud during the encounter.
What a follow-up encounter actually sounds like
In a real follow-up visit, the physician walks in and says something like, the metformin is fine, we are continuing the lisinopril at twenty, and the rash is gone. There is no review of systems verbalized, no formal history of present illness, and no exam beyond a brief auscultation that the physician comments on in two words. A naive ambient scribe will dutifully transcribe the spoken sentences and produce a note that is mechanically accurate and clinically empty, because the meaning of the visit lives in the chart, not in the air. The patient is on metformin five hundred twice daily because of the visit nine weeks ago, the lisinopril dose was titrated up two visits ago to control morning systolics, and the rash was the reason for an urgent visit eleven days earlier. None of that context was spoken in the room, but all of it must be in the note.
Chart-aware generation versus chart-naive transcription
The distinction that matters is whether the ambient layer is generating a note from a closed-loop transcript or from a transcript fused with the longitudinal chart. The transcription-only approach produces a note that reads as if every visit is the first time the patient has been seen, since the model has no prior context to draw on. The chart-aware approach, which a smaller number of platforms have actually implemented at the architectural level rather than the marketing level, treats the transcript as a delta against the existing chart, then resolves references using the medication list, the active problem list, the recent results, and the prior plan. The output of the chart-aware approach reads like a clinician's note, with appropriate referencing of dose adjustments, problem status changes, and continued plans, rather than a verbatim retelling of an opaque conversation.
Hero EMR's implementation, examined under follow-up conditions
The reviewer tested Hero EMR's ambient scribe across a controlled set of fifteen simulated follow-up scenarios, ranging from a stable diabetic encounter to a complex psychiatric medication management visit to a post-hospitalization heart failure check. The pattern that emerged was consistent. When the physician said, the metformin is fine, the generated note recorded that the patient is tolerating metformin five hundred milligrams twice daily without gastrointestinal side effects and that the regimen is being continued without change, with the dosing pulled directly from the active medication list. When the physician said, we titrated up last time and her morning numbers look better, the note correctly attributed the prior titration to the previous visit, summarized the home glucose log that had been uploaded between visits, and framed the assessment as continued response to the recent change. None of that context was in the verbal exchange.
Where the system still requires supervision
It would overstate the technology to claim that chart-aware ambient documentation is fully autonomous, and the reviewer encountered three failure modes that warrant explicit attention. The first is reference resolution when the patient mentions a symptom that overlaps semantically with a prior problem but is clinically distinct, such as a new shoulder pain that the model briefly attributed to a months-old shoulder problem before the physician corrected the assessment field. The second is plan inheritance from a prior visit that was no longer accurate, which happened when a labs follow-up that had already been completed appeared in the new plan section. The third is the boundary between physician verbalization and patient verbalization when both speakers say similar things, since the model occasionally attributes a sentence to the wrong speaker, although this is rare in the current build.
What separates the chart-aware approach from competing implementations
Other ambient scribes tested for this article, including the standalone vendors that have raised significant funding and the embedded modules in two enterprise EMRs, did not pass the follow-up test consistently. The standalone vendors generally have no access to the chart and therefore cannot resolve references at all, leaving the physician to either dictate context that the chart already contains or accept notes that read as if every patient is being seen for the first time. The enterprise embedded modules technically have chart access but appear to use it sparingly, in the sense that the generated notes still treat the transcript as the primary source rather than the chart. The Hero EMR implementation behaves differently because the scribe is not a feature bolted onto the EMR, it is an integrated layer that treats the chart as part of the input, and the resulting notes reflect that architectural difference.
Why the follow-up problem is economically significant
Follow-up visits represent the majority of encounters in most primary care and subspecialty practices, often by a wide margin, which means the value of an ambient scribe is overwhelmingly determined by how it performs on those visits rather than on the marketing-friendly new patient case. A scribe that adds friction on a follow-up because the physician has to manually add context the chart already knows is a scribe that costs time rather than saving it, and the cumulative effect across a panel of patients can erase the documentation benefit entirely. Practices evaluating ambient documentation should be running the technology through several days of real follow-up patients before committing to a contract, since the new patient demo will not surface the failure modes that matter most.
A practical evaluation protocol for prospective buyers
The reviewer suggests a simple, reproducible evaluation that any practice can run during a vendor trial. Identify five returning patients with stable chronic conditions and an active medication list of three or more entries, then conduct the visits as usual without modifying the verbal communication to accommodate the scribe. After each visit, review the generated note for three specific properties, including whether medications and doses are correctly referenced without being spoken aloud, whether prior visit context is appropriately summarized in the assessment, and whether the plan reflects continued, modified, or discontinued therapies rather than treating each visit as a clean slate. Any system that fails on two or more of these properties is not ready for the daily reality of follow-up documentation, regardless of how impressive its new patient demonstration looked.
The honest verdict
The follow-up problem is not solved, but Hero EMR's chart-aware ambient scribe is meaningfully closer to solving it than any other commercial implementation the reviewer has tested in the past twelve months. The combination of chart fusion, longitudinal context resolution, and clinician-style note generation produces output that holds up under the conditions where most ambient scribes quietly fail. Other implementations will likely converge on this approach over time, since the architectural insight is not proprietary, but the practical execution gap remains significant as of this writing. Practices that document a high proportion of follow-up visits, which is to say most practices, should weight follow-up performance heavily in any ambient scribe evaluation, and they should treat the new patient demo as the marketing exercise it usually is.