Ambience's Chart Prep Surfaces 92.2% of Critical Pre-Visit Facts

Ambience Healthcare's newest AI chart-investigation harness surfaced 92.2% of critical pre-visit clinical facts on a 46-case benchmark, up from 78.7% for its production system. The company announced Ambience Prepare, a pre-visit workstream built around the Chart Prep feature, on November 12, 2026.

Chart Prep turns longitudinal patient records into evidence-linked briefings. It orients the clinician in the short window before a visit, restores the relevant care thread, flags changes and unresolved questions, and leaves clinical decisions with the clinician. Ambience also ships Chart Chat in Prepare, which lets clinicians ask natural-language questions about a patient's chart.

Why the medication case matters

Ambience illustrates the problem with a urologist preparing to see a patient with a recent multifocal stroke before a planned transurethral resection of the prostate. Cardiology had planned to change medications if atrial fibrillation was detected. A later test confirmed atrial fibrillation — but subsequent records left unclear whether apixaban was started and clopidogrel stopped.

Each record entry could be summarized accurately without ever resolving the medication question central to surgical preparation. Chart Prep is built to connect the procedure with medications relevant to bleeding risk, follow the cardiology plan, check the rhythm-monitor result and compare the intended change against the later medication reconciliation. It shows what the record supports, identifies conflicts and preserves unresolved questions for clinician review.

The takeaway: useful preparation makes the unresolved plan visible before surgery, but deciding how to reconcile it stays with the clinician.

What does Chorus add?

Chart Prep draws on Chorus, the context system Ambience announced on August 24, 2026. Chorus gives the agent tools to survey a record, follow a clinical thread, open underlying evidence, compare sources and trace conclusions to their origins. It maintains a shared, source-linked understanding of the patient that different Ambience workflows can reuse, navigating notes, medications, labs, admissions and diagnostic reports while reconciling conflicting sources and timelines.

Ambience argues that simply loading as much chart text as possible into a model doesn't work: long records can exceed context windows, crowded context gets harder to use reliably, and the model still needs direction about which threads to inspect. The electronic health record remains authoritative; Chorus supplies a shared context layer between that record and Ambience's AI workflows.

How was it measured?

Ambience's evaluation starts with board-certified clinicians reviewing de-identified charts as they existed immediately before an upcoming visit. They identify and rank encounter-relevant facts; a second clinician independently reviews each case, and disagreements or omissions get adjudicated. Automated reviewers compare outputs against those clinician-established references only after reaching acceptable agreement with clinician judgments. Each case runs multiple times to check consistency.

The numbers come from 46 de-identified patient cases with three independent runs per case:

• Critical-fact recall: 92.2% with the newest context harness vs. 78.7% for production; 95% confidence interval of 87.2% to 96.3%.
• Recall of all critical and high-importance facts rose from 66.3% to 72.3%.
• Share of remaining misses attributed to failure to retrieve necessary evidence fell from 19.9% to 9.3% — a 53% reduction, per Ambience.

The comparison held the generation model, patient cohort, clinician references and evaluation suite constant, making the Chorus-built context harness the principal change. The measured difference concerns how the agent investigates the chart, not a change in the underlying model.

What does this mean for health systems?

That separation gives health systems a specific measure of the product's usefulness: whether better investigation retrieves and prioritizes facts clinicians consider consequential, while preserving evidence and uncertainty for human review. For now, Ambience's benchmark says better context navigation moves that number by 13.5 percentage points.