The State of AI Clinical Documentation, 2026

1. Why documentation became the front door

Every other category of clinical AI asks a clinician to trust a new judgement. Documentation asks for something easier: give me back my evening. That is why it became the first generative AI category adopted at health-system scale.

For every hour physicians provide direct clinical face time to patients, nearly 2 additional hours is spent on EHR and desk work within the clinic day.

That time study framed the decade. Eric Topol's Deep Medicine later argued that AI's most valuable contribution might be returning time and presence to the clinical encounter. Ambient documentation is the first technology to attempt that at scale.

2. What HealthAratus data shows

HealthAratus tracks 35 records in the Documentation category. 22 carry a live HealthAratus Score. 13 are core ambient documentation tools; the remainder are speech recognition, human scribe services and specialty documentation systems. The ranking is public on HealthAratus Rankings and its weighting across Evidence, Deployment, User Experience and Innovation is set out in the methodology.

Three observations stand out.

• A single clear leader. Abridge scores 92.0, nearly 25 points ahead of the next tool. No other category on HealthAratus has a gap this wide at the top.
• A crowded established cluster. Six tools sit within seven points of each other (61.9 to 67.5). At this density, small shifts in real-world experience will reorder positions.
• Identity fragmentation. Several companies appear under more than one product identity, notably Nuance and Microsoft (Dragon Medical, DAX, DAX Copilot, now Dragon Copilot). Rebranding is constant in this category, which is why HealthAratus resolves each tool to one canonical record.

3. The tools and companies

The leader. Abridge combines health-system scale with deep Epic integration and a strong evidence posture, including the Kaiser Permanente deployment.

The platform incumbents. Microsoft, through DAX Copilot and Dragon Medical, owns the longest installed base of clinical voice. Epic Ambient Documentation represents the EHR vendor itself entering the category, the most important structural change in the market.

The independents. Corti, DeepScribe, Suki, Ambience, Heidi, Nabla and PatientNotes compete on specialty depth, price, speed of deployment and, increasingly, on coding and revenue cycle. Augmedix, now part of Commure, shows the hybrid human and AI scribe model folding into a broader platform. Eleos serves behavioural health, where note structure and privacy differ markedly.

4. What has happened

The category moved from pilot to procurement. Capital concentrated in a few companies, the largest EHR vendor entered directly, and Microsoft unified its voice products.

The Permanente Medical Group's ambient AI rollout reached thousands of physicians within months, the fastest adoption of any technology in the group's history, with substantial estimated time savings.

Sources: Microsoft on Nuance, Microsoft on Dragon Copilot, Kaiser Permanente on Abridge.

5. Where the money went

Documentation is the category where healthcare AI capital has concentrated most. Rock Health counted $14.2 billion of US digital health venture funding in 2025, up 35% on 2024, and AI-enabled companies took 54% of it. Within healthcare AI spending specifically, Menlo Ventures estimates that $600 million of $1.4 billion in 2025, about 43%, went to ambient documentation, which it calls healthcare AI's first breakout category, with revenue up 2.4 times year on year. Coding and billing automation took a further $450 million.

Put the two together and roughly three quarters of healthcare AI spend went to capturing the encounter and turning it into a claim. That concentration explains the pressure on vendors to grow beyond the note.

6. From the note to the claim: horizontal expansion

The pattern is now consistent. A vendor gets into the workflow through the note, then moves into the functions that rely on the note: coding, clinical documentation integrity, prior authorisation and revenue cycle.

• Abridge into revenue cycle. In September 2026 Abridge launched pre-bill review for CDI and coding teams. It compares the coded inpatient record against the documented complexity of care before submission. The company says it does not change documentation, codes or claim status, and CDI teams decide whether to hold, correct or release a claim.
• Abridge into prior authorisation. Abridge is co-designing point-of-care prior authorisation with Highmark Health and announced a real-time prior authorisation partnership with Availity in January 2026.
• Ambience as coding-first. Ambience sells documentation, real-time coding and revenue integrity as one platform.

Our partnership with Abridge started with documentation and has kept growing because each step earned the next one.

Implications. First, the commercial case is moving from saving clinicians time to recovering revenue, which is easier to measure for a CFO. Second, it changes who carries the risk: a note that feeds a claim is also a billing record. A policy brief on ambient scribes and the coding arms race warns that more intensive coding can raise payer costs and invite payer counter-measures. Third, it favours platforms: the vendor that holds the note has a structural advantage downstream, which pushes standalone scribes toward consolidation.

7. How the technology changed the product

The category has gone through three technical generations, each of which changed what the product could do, not just how accurate it was.

1. Dictation (1990s to 2010s). Front-end speech recognition such as Dragon Medical turned spoken words into text. The clinician still had to structure the note.
2. Hybrid ambient (late 2010s). Remote human scribes, sometimes helped by machine learning, as in the Augmedix model, drafted notes from recorded visits. Quality was good, but cost and turnaround limited how far it scaled.
3. Generative ambient (2023 onwards). Large language models made it possible to summarise a whole conversation into a structured, specialty-specific note within seconds, with deep EHR write-back. Once the encounter was understood as structured data, the same output could feed codes, orders, patient instructions and prior authorisation. That is the technical basis for the expansion described above.

The gains are real but uneven. The Peterson Health Technology Institute found ambient scribes are likely to reduce documentation time and cognitive load, while early evaluations found less burnout but no clear financial return on investment. A randomised trial in clinical practice adds controlled evidence, but long-term data by specialty and outcome is still scarce.

8. Hospital adoption

Adoption has outpaced evidence. An analysis in AJMC found that nearly two thirds of US hospitals using Epic have adopted ambient AI, with higher uptake in larger, not-for-profit systems. KLAS lists ambient documentation among the most widely adopted healthcare AI use cases. At Kaiser Permanente, The Permanente Medical Group made it the fastest-adopted technology in its history (see Section 4).

Being deployed is not the same as being used. One academic health system reported that its AI scribe was used in just 24% of eligible encounters. Licences sold and organisations signed up overstate how much of the work the tool is actually doing. This is the gap between deployment and use that HealthAratus measures through clinician signals.

9. Where the risk lives

A note that looks complete can still be wrong. The failure modes in Figure 3, omission, hallucination, speaker attribution error, edit fatigue and coding intensity, are rarely visible in a demo. They show up across thousands of encounters, in specific specialties and at specific sites. This is the category where vendor-reported time savings and observed reality are most likely to diverge, and where Drift Watch and the Signal Network matter most.

The ambient scribe won the argument about time. The next argument is about truth.

10. Why HealthAratus tracks this category

Documentation is the category where AI touches the most encounters, changes most often and fails most quietly. Clearance, procurement and pilots all happen before the moment that matters: months into routine use, after the vendor has shipped its third model update. These are the reasons it sits at the centre of what HealthAratus observes.

11. What we predict

1. Consolidation around platforms. EHR-native and platform players will take the default seat; independents will survive by specialty depth or by owning downstream workflow.
2. From notes to actions. The product will expand from drafting notes to queuing orders, referrals, patient instructions and codes. Value, and risk, move downstream.
3. Coding scrutiny. As ambient notes capture more detail, payers and regulators will examine whether documentation intensity is changing billing.
4. Price compression. As ambient capture becomes a feature of the EHR, standalone per-clinician pricing will fall.
5. Evidence becomes the differentiator. Buyers will ask for independent, post-deployment evidence: edit rates, omission rates, clinician trust over time and outcomes by specialty. Those are exactly the signals the HealthAratus Score is built to observe.

Related analysis: Is the FDA's 510(k) pathway rigorous enough for adaptive AI? and The hidden failure modes of clinical AI. Explore the live category on the Intelligence Map, or share your own experience of a documentation tool.

Conclusion: post-deployment performance is now the question

For a decade the question about clinical documentation AI was whether it could work. That question is settled. The ambient scribe is deployed across most large health systems, funded at a scale no other healthcare AI category matches, and is now moving into coding, claims and prior authorisation.

What is not settled is how these tools perform after they go live: across specialties, across sites, and across the model updates that arrive every few months without fresh review. Pre-market evidence cannot answer that. Vendor dashboards will not answer it independently. Only continuous observation of real use can.

Our recommendation is unambiguous: every health system, buyer, regulator and vendor should now treat post-deployment performance as the primary measure of a documentation tool. Track it continuously, by specialty and by version, and make the result independent.

That is the record HealthAratus exists to build, and it is only as good as the experience of the people who use these tools every day.

Frequently asked questions