Ambient scribes have changed how many clinicians experience documentation. Instead of typing late into the night, they can speak naturally with patients while the system drafts their notes in the background. For many organizations, that alone feels transformative.
But as these tools move from pilot to routine use, a deeper question has emerged. Is capturing the conversation enough to produce documentation that reflects the patient’s full clinical story — how their conditions are evolving, what the relevant labs show, how they’ve responded to recent medication changes, what the current plan is, and everything else that gives the visit its clinical meaning?
In a recent study, our medical team set out to explore that question. We compared documentation generated from ambient transcripts alone with documentation created when those transcripts were paired with patients’ longitudinal medical histories: prior notes, labs, imaging, medications, consults, and trends over time.
The difference was substantial. With ambient transcripts alone, clinicians captured about 40% of the clinically relevant information. When we combined ambient data with longitudinal context, documentation completeness rose to 83%, and overall documentation quality improved by around 18%.
These findings don’t mean ambient technology is flawed. They highlight something more basic: the visit transcript reflects only one moment in a much longer clinical story.
Why the transcript can’t stand alone
Most clinicians don’t assume that the conversation tells them everything they need to know. They have lived for years with the reality that key information is scattered across systems and time.
A patient’s lab trajectory, for example, may be more important than a single value mentioned in a visit. A consultant’s assessment might sit in a scanned PDF that the patient never brings up. Medication changes from another health system might alter the risk profile of a seemingly routine complaint.
Much of this context never appears verbatim in the room.
Ambient tools excel at capturing what is said and structuring it into a note. What they do not provide is the longitudinal context that clinicians rely on when documenting the patient’s current state and, often, when shaping decisions about the plan going forward. Our research helps quantify the gap between “this is what we talked about” and “this is everything clinically relevant about this patient right now.”
From documentation burden to cognitive burden
The initial promise of ambient scribes focused on reducing documentation burden: fewer clicks, less typing, more face time with patients. Those are real gains. But the feedback we hear from clinicians suggests that the more persistent burden is cognitive.
When AI focuses only on the in-room conversation, that investigative work doesn’t disappear. Before or after visits, clinicians still spend time hunting for relevant labs, imaging, and consults, reconstructing timelines across multiple encounters and settings, reconciling conflicting or outdated information – often during after-hours, on their own time, long after the visit has ended.
This has implications beyond efficiency. Incompleteness at the documentation level can contribute to missed patterns, overlooked risk factors, and gaps in quality performance, especially in models that depend on proactive management of chronic conditions. While our study did not directly measure downstream outcomes, it highlights how easy it is for important signals to fall through the cracks if AI systems don’t help surface them.
AI implementation pitfalls
As AI becomes more common in clinical workflows, differences in implementation are starting to matter as much as differences in technology. The organizations we see struggling with AI share a few recurring patterns.
First, treating ambient scribes as a replacement for pre-visit chart review.
If ambient tools are introduced with the implicit message that “the system will take care of the note,” clinicians may reasonably pull back from pre-charting – even though crucial information still lives outside the transcript.
Second, underestimating the role of data infrastructure.
Even when there’s a clear desire to bring longitudinal data into the exam room, fragmented systems and inconsistent data formats get in the way. The best AI solutions are those that can integrate and reconcile multi-source, multi-modal data in a way that actually fits clinical workflows.
Third, focusing on novelty over cognitive fit.
Clinicians are more likely to adopt tools that mirror their natural reasoning process: identifying problems, trends, and risks within a broader context. If AI surfaces too much information or presents it in ways that don’t align with clinical thinking, users quickly disengage.
It’s also worth noting that data fragmentation is only one of several constraints. Governance, reimbursement, training, and change management all shape how much value organizations ultimately derive from AI tools.
True clinical AI
Taken together, our findings point toward a more grounded framework for what clinical AI should aspire to do.
Ambient transcription provides immediacy. It captures the content and structure of the encounter as it unfolds.
Historical data provides orientation. It shows whether today’s findings are new, stable, or part of a longer trend.
Clinical AI creates value when it connects the two. The goal is not to replace clinical judgment, but to give clinicians a more complete, coherent view of the patient at the moment documentation is created and care plans are formed.
Many organizations are beginning to see that ambient documentation is most effective when it is paired with the clinical context that sits outside the conversation. As healthcare leaders think about the next phase of AI adoption, the conversation will likely shift from whether AI can write notes to whether it can support better thinking – helping clinicians see more of what matters, with less effort, and with fewer hidden gaps, so that documentation truly reflects the richness and continuity of each patient’s story.
Dr. Michael Zuckerman
Dr. Michael Zuckerman is Head of Clinical AI Research at Navina. He has a multidisciplinary background spanning electrical engineering, physics, and medicine (MD), along with clinical training as a family medicine resident and academic experience as a university lecturer in biostatistics.
