When JAMA published the largest real-world study of AI scribes on April 1, the results fell well short of what the industry had been promising. Among the 1,809 clinicians who adopted ambient documentation across five academic medical centers, the average time saved on note-writing was about 16 minutes per eight hours of patient care.
The study also did not find a relevant reduction in the time clinicians spent in the electronic health record after hours.
Two months earlier, KLAS Research had said that they had never seen a technology adopted by hospitals as quickly as ambient AI, and both findings can be true at the same time. A September analysis of 412 ambient scribe implementations helps explain the gap, finding that only 56.8 percent of licensed seats were still being used regularly a year after the start of the contract.
The leading solutions can deliver real value, but how much a health system gets from them depends on what happens after the software is installed.
Abridge Leads KLAS as Epic Builds Their Own Scribe
Abridge won the Best in KLAS award for ambient speech for the second year in a row in February, with a rating of 94.7 out of 100. Their parent company, Ambience Healthcare, also received the KLAS and CHIME Trailblazer Award a few weeks before that for technology that suggests billing codes as clinicians document patient visits.
In February, Epic announced their own AI charting feature, built on Microsoft’s Dragon ambient technology, to draft notes and prepare orders for clinicians to sign.
For standalone vendors, having a capable scribe included in the EHR a health system already pays for makes each renewal conversation harder, because they have to show why their product is worth an additional contract and support that case with specialty depth and coding accuracy that can be difficult to match in a first-generation native tool.
The Benefits Extend Beyond Time Saved
The case for these tools becomes clearer when the studies look beyond time saved on documentation, with several findings pointing to benefits for clinicians and patients. In a combined Mass General Brigham and Emory University analysis of more than 1,400 clinicians, burnout prevalence at Mass General Brigham fell by 21.2 percentage points among clinicians who were using the tools 84 days after implementation.
UC San Diego Health found that AI-assisted drafting did not shorten physicians’ response times to patient messages, but it did reduce the mental effort required to write each response. Much of the burden of patient messaging comes from having to start from a blank page later in the day, something that time studies only partly capture.
The JAMA researchers also found that the largest gains went to clinicians who used the scribe in more than half of their visits, yet only about a third of adopters reached that level, while the rest returned to their old habits.
Unused Seats Extend the Payback Period
Those unused licenses still represent a real cost, and a September analysis of 38,600 clinicians found that 77.6 percent of seats were active three months after go-live, with usage declining at every checkpoint that followed. Independent practices, which typically license only the clinicians who request the tool, had 74 percent of their seats still active after a year, while multi-hospital systems that tend to purchase licenses in bulk had only 49 percent still in use.
Active means at least four notes in the preceding month. Source: Healthcare Industry Reviews, Ambient AI Scribe ROI Statistics: 2026 Report.
For each clinician who actively uses the tool, the report estimates an annual return of $10,500 against a $4,200 license cost, putting the payback period at less than five months; when that same return is spread across every licensed seat, however, the payback period stretches to about 8.5 months. Improved coding capture accounted for 27 percent of the return, although many organizations could not verify the gain because they had not recorded their evaluation and management coding levels before launch.
Where AI is cutting admin work, 2026
| Admin task | Leading tools | Best evidence so far |
|---|---|---|
| Visit notes | Abridge, Ambience, Microsoft Dragon Copilot, Epic AI Charting | 16 minutes saved per 8 hours of care (JAMA, 2026) |
| Portal messages | Epic in-basket draft replies | No time saved, lower cognitive burden (UC San Diego, 2024) |
| Coding | CodaMetrix, coding-aware scribes | CodaMetrix is 2026 Best in KLAS for autonomous coding |
| Prior authorization | Infinitus, AKASA (provider); Cohere Health (payer) | Cohere reports up to 90% auto-approval (company data) |
Sources: KLAS Research; JAMA (April 2026); JAMA Network Open (April 2024); company disclosures where noted.
What to Negotiate in the Next Scribe Contract
KLAS scores show how these tools perform when clinicians use them, but they reveal much less about what happens in the second year. For health systems signing contracts this year, that makes utilization worth building into the agreement itself, with active-use thresholds tied to spending and a six-month audit to identify dormant licenses that can be reassigned to clinicians on the waiting list.
An E/M baseline should also be recorded before the first microphone is turned on, since coding gains that cannot be measured are difficult to defend when the budget comes back under review. In the September data, deployments that audited use at six months and reassigned idle seats recovered most of the lost return within a quarter, giving health systems a clearer way to protect ROI than any award listed in a vendor’s sales deck.
