Dragon Copilot for radiologists: what changed and what didn't
Dragon Copilot merges Dragon Medical One and DAX ambient AI. For radiologists, here is what the change means, where it fits, and why on-device front-end dictation often serves better.
By The RadMyk team
In March 2025, Microsoft unified two separate Nuance products, Dragon Medical One and DAX Copilot, into a single offering called Dragon Copilot. If you have been using Dragon Medical One for radiology dictation, or evaluating it, the question is worth asking directly: what actually changed, and does any of it matter for how radiologists work?
The short answer is that a lot changed for primary care and ambulatory physicians. For radiologists, most of the architectural issues that made Dragon Medical One a poor fit are still present in Dragon Copilot.
What is Dragon Copilot?
Dragon Copilot is Microsoft’s unified clinical AI assistant. It combines three capabilities under one per-provider subscription:
- Front-end voice dictation: the core Dragon Medical One engine, which converts spoken words into text at the cursor in real time.
- Ambient documentation: the DAX Copilot capability, which listens passively to a clinician-patient encounter and generates a draft clinical note from the conversation.
- Generative AI actions: summarization of visits, referral letters, after-visit instructions, and order suggestions, all built on Azure OpenAI Service.
All three run in the cloud. Audio and conversation data travel to Microsoft Azure for processing. The product is sold as a subscription, priced per provider, and distributed either directly or through the Microsoft CSP (Cloud Solution Provider) channel for smaller practices.
How Dragon Copilot differs from Dragon Medical One
Dragon Medical One was dictation only. You pressed a button, spoke your report, and the text appeared in your EHR or PACS field. The product had no ambient listening, no AI note generation, and no conversation analysis.
Dragon Copilot adds the ambient layer on top of that same dictation core. When you see a patient, Dragon Copilot can listen to the conversation in the background and produce a structured draft note without you manually dictating anything. For primary care physicians seeing 20 to 30 patients a day, this is a meaningful change. Each visit produces documentation automatically, reducing after-hours charting significantly.
That ambient capability is where most of Dragon Copilot’s new value sits. The dictation engine underneath is recognizably similar to Dragon Medical One.
Why the ambient layer adds little value for most radiologists
Radiology workflows are different from the clinical encounters Dragon Copilot’s ambient AI is designed for.
A primary care physician sits with a patient, takes a history, examines, explains, and orders. Dragon Copilot can capture all of that conversation and draft the note from it. An interventional radiologist consenting a patient before a procedure has a similar scenario.
Most diagnostic radiology is different. You sit in a reading room, review images on a PACS workstation, form findings, and dictate a structured report. There is no ongoing patient conversation to capture. The “documentation” is not a summary of an encounter: it is a precise technical description of what the images show. You are not looking for AI to generate that from a conversation you had. You are looking for a fast, accurate tool to transcribe the specific language you are producing.
This is front-end dictation. Dragon Copilot includes it, but the ambient layer, the generative summaries, and the AI actions are largely irrelevant to a chest radiologist or a neuroradiologist working through a worklist. You are paying for a full clinical AI platform when the part you need is the dictation engine.
What hasn’t changed for radiologists
No native Mac support
Dragon Medical One required a Windows environment. Radiologists on macOS had to run it inside Parallels or Boot Camp. Dragon Copilot inherits the same architecture. If your reading workstation is an Apple Silicon Mac, that situation has not changed. Virtualization adds latency, changes how global shortcuts behave, and can conflict with PACS client software.
This matters more than it might seem. Apple Silicon Macs are fast, quiet, and increasingly common in radiology. The M-series chip handles heavy PACS workloads well. No native Dragon Copilot client means Mac radiologists are still working around a Windows dependency.
Audio leaves the machine
Dragon Copilot processes voice and conversation in Microsoft Azure, as Dragon Medical One did before it. Every dictation you make, and every ambient recording if you use that feature, travels to a cloud server for processing. This is the fixed architecture of the product.
For most hospitals, a Business Associate Agreement with Microsoft manages the HIPAA obligations. But the audio still leaves the workstation. Radiologists who want patient voice to stay on-device have no mechanism to change that within Dragon Copilot.
Subscription pricing
Dragon Copilot is a per-provider subscription. Microsoft restructured pricing in May 2026, reducing the per-user rate and consolidating the product tiers into a single standard offering. But it remains a subscription. Each year, a new bill arrives. If the product changes pricing again, you adjust or leave.
For radiologists in large hospital systems where IT manages the procurement, this may not be a daily concern. For solo radiologists, small groups, and teleradiologists who buy their own tools, a recurring per-user fee for a platform that includes features you will never use is a structural mismatch.
General medical vocabulary depth
Dragon Medical One was a general clinical dictation product. Radiology’s specific vocabulary, subspecialty anatomy, imaging modalities, measurement conventions, laterality patterns, and structured reporting cadence, lives in a separate Nuance product line: PowerScribe. Dragon Copilot’s dictation core carries forward Dragon Medical One’s general medical vocabulary. If your dictation involves highly specific radiology language, you may still find gaps that need custom vocabulary work.
Compare this with a tool trained specifically on radiology: one that already knows “right lower lobe consolidation” and “no evidence of pneumothorax” without supplementation.
When Dragon Copilot makes sense for radiology
There are scenarios where Dragon Copilot is a reasonable choice for a radiologist.
If you work in interventional radiology or have a significant clinical consultation practice, the ambient documentation layer becomes genuinely useful. Consenting conversations, post-procedure discussions, and clinic notes are exactly the encounter-based documentation Dragon Copilot is built for.
If your institution runs Epic and is already in the Microsoft enterprise ecosystem, Dragon Copilot has deep EHR integrations and workflow hooks that a standalone dictation tool cannot replicate. Where the EHR is deeply embedded in the workflow and IT manages the toolchain, Dragon Copilot is a coherent choice.
If you are already a Dragon Medical One subscriber and your institution is moving to Dragon Copilot as part of a broader Microsoft agreement, the transition may be straightforward with no new purchasing decision required.
When it doesn’t
Dragon Copilot is a poor fit if any of the following describe your situation.
You are on a Mac. No native client means a virtualization workaround. That has costs: setup time, latency, and software compatibility.
You need offline capability. Reading rooms lose network access. Hospitals have VPN instabilities. If your dictation must survive a network outage, a cloud-only product is an operational risk. Dragon Copilot does not function without connectivity.
You want to own your software. Subscriptions are ongoing costs. A per-provider dictation subscription can total $3,000 to $4,000 or more over three years, before add-ons, at indicative reseller pricing. If you change institutions, the subscription travels with the provider (Dragon Medical One’s profile portability carries forward), but the bill also does.
You do diagnostic radiology and do not see patients. The ambient AI layer, which is the primary new feature Dragon Copilot adds over Dragon Medical One, is not useful for image-based reporting workflows. You are paying for a capability that does not apply to your work.
You want audio to stay on your machine. The cloud architecture is not configurable out of the product’s standard offering.
The alternative: on-device radiology dictation
RadMyk is built for exactly the radiology workflow Dragon Copilot was not designed around. It is front-end dictation only. Speak, and your words appear at the cursor in your PACS field, RIS, EHR text box, Microsoft Word, or any other app where you can type.
The speech engine runs fully on your machine. Voice never leaves the workstation. It works when the network is down. It runs natively on macOS Apple Silicon and Windows.
Accuracy is 96.1%, measured out of the box, with a radiology-tuned vocabulary. Transcription runs at approximately 220 words per minute. Guided calibration tunes the model to your voice and microphone after a short first-run session.
The pricing model is different from Dragon Copilot in one significant way. RadMyk costs $199 one-time for the first 100 users (Rs 14,999 in India), rising after that. No subscription, no renewal, no per-provider meter. Radiology trainees, residents and fellows, get access free for the full length of their training. Everyone else gets a 28-day free trial at radmyk.com/pricing with no credit card required.
If you are moving from Dragon Medical One, or evaluating Dragon Copilot and finding that the ambient AI adds complexity you do not need, RadMyk is worth comparing directly. Our medical dictation software guide covers the broader category, and the comparison of RadMyk and Dragon Medical One goes further on the specifics if you want the side-by-side.
The honest summary
Dragon Copilot is a meaningful upgrade for primary care physicians, ambulatory clinicians, and radiologists with active patient contact. Microsoft bundled real AI capability into a product that was previously dictation only.
For diagnostic radiologists working through a worklist in a reading room, the ambient AI layer is mostly overhead. The core dictation engine, the cloud architecture, the Windows dependency, the subscription model, and the general medical vocabulary are all carried forward from Dragon Medical One.
If you need clinical AI for encounter documentation and are already in the Microsoft ecosystem, Dragon Copilot delivers that. If you need fast, private, radiology-tuned dictation on a Mac or Windows machine with no ongoing subscription, you are paying for capabilities you will not use.
A 28-day free trial of RadMyk runs on your actual reading workstation, in the PACS and apps you already use, with no credit card required. That is the right comparison before committing to either product.