How to Dictate in Epic on a Mac

Short answer: Most Mac clinicians reach Epic through a remote session, so dictation has to run on the Mac itself and deliver finished text as keystrokes into the session window. Speak the narrative sections, type the discrete fields, and let SmartPhrases fire from the keyboard.

If you searched for this, you have probably already tried the obvious thing. You opened a progress note, pressed whatever dictation shortcut you use elsewhere on your Mac, spoke a sentence, and watched nothing appear. Or worse, watched half a sentence appear and then stop. The note field sat there empty while your Mac's microphone indicator glowed helpfully in the corner.

This is not a bug in your setup. It is a structural consequence of how most people run Epic on Apple hardware, and once you understand it the fix is straightforward. This guide covers what is actually happening, the architecture that works, which parts of a chart are worth speaking and which are not, and where the honest limits are.

Why Epic on a Mac is a different problem

Epic's full desktop client is a Windows application. Organizations that support Mac users almost always do it one of three ways: a published Windows desktop or application delivered through Citrix, a virtual desktop through VMware Horizon or a similar broker, or a browser-based route where the health system has enabled it. A minority of clinicians work on a Mac for everything except Epic, and keep a Windows machine or a shared workstation for charting.

In the first two cases, and they cover most Mac users, Epic is not running on your Mac. It is running on a server somewhere in your organization's data center, and your Mac is displaying a picture of it. Your keyboard and mouse events get forwarded across the network into that session. Your microphone, by default, does not.

That distinction is the whole story. A dictation tool that expects to talk to the application it is typing into finds nothing to talk to, because the application is not there. Audio redirection into a remote session is a separate feature that has to be configured deliberately, is frequently disabled by IT for bandwidth and security reasons, and even when enabled tends to introduce enough latency and compression that recognition quality drops noticeably.

We wrote a longer piece on this class of problem in dictation software for Citrix and VDI. The short version is that the fight is not worth having.

The architecture that works: speak locally, type remotely

Instead of pushing your voice into the remote session, do the recognition entirely on the Mac side and hand the remote session what it already knows how to accept: keystrokes.

The sequence is simple. You put your cursor in an Epic note field inside the session window. You hold your hotkey on the Mac, speak a few sentences, and release. The transcription happens on the Mac's side of the connection. The finished text is then delivered at the cursor as though you had typed it, which means the Citrix or Horizon client forwards it into the session exactly the way it forwards every other keystroke you make.

From Epic's perspective, nothing unusual happened. A user typed. There is no plugin to install in Hyperspace, no change to your organization's Epic build, nothing for IT to package, and nothing that breaks when Epic is upgraded. This matters more in healthcare than almost anywhere else, because clinical machines are managed, locked down, and updated on a schedule you do not control.

Voice Keyboard Pro works this way by design. It lives in the Mac menu bar, holds a hotkey, and types finished text at whatever cursor is active system-wide. It does not know or care whether that cursor is in Apple Mail, a browser, or a remote session window.

Setting it up

1. Install the Mac app and grant it microphone and accessibility permission when macOS asks. Accessibility is what allows text to be delivered at the cursor rather than only into the app's own window.
2. Pick a hotkey that your remote session will not swallow. More on this below, because it is the most common first-run failure.
3. Open your Citrix or Horizon session and log into Epic normally.
4. Open a test patient or a scratch note. Click directly into the note text so the caret is blinking inside the field, not merely hovering over it.
5. Hold the hotkey, say one short sentence, release. Confirm the text lands in the field.

If step five produces nothing, the near-certain cause is that the caret was not actually in a text field. Remote session windows can look focused without the inner application having focus. Click into the field, type one character by hand to confirm, delete it, then try again.

The hotkey collision problem

Remote desktop clients are greedy about keyboard input. That is their job. Depending on how your session is configured, a large share of your modifier combinations get forwarded straight into the Windows session rather than being handled by macOS, which means a hotkey that works everywhere else on your Mac may do nothing, or may trigger something unexpected inside Epic.

Two practical rules. First, prefer a function key or a lone right-side modifier over a multi-key chord. Keys in the F13 to F19 range are unclaimed almost everywhere and survive being forwarded. Second, test your chosen key inside the session before you rely on it, not just on the Mac desktop.

Epic itself compounds this. Clinical builds lean heavily on single-key and function-key shortcuts for chart navigation, and your organization may have added more. If your dictation hotkey happens to collide with a navigation command, you will not get an error message. You will get a chart section you did not ask for, in the middle of writing a note. Test deliberately, and pick something boring.

Where dictation actually pays off in a chart

Not every field in Epic deserves your voice. The value is concentrated in a handful of surfaces, and it is concentrated there for a specific reason: those are the places where the text is prose, where nobody else can write it for you, and where the alternative is typing it at the end of a clinic day when you are already behind.

1. The narrative sections of the note

History of present illness and the assessment and plan are where the thinking lives, and they are the two sections most likely to get compressed into fragments because writing them properly costs time you do not have. This is the single highest-value place to speak instead of type.

The arithmetic is not subtle. Ordinary adult typing runs around 40 words per minute. Conversational speech runs 130 to 150. A five-sentence assessment that takes two minutes to type takes well under a minute to speak, and the spoken version tends to be more complete, because you are not unconsciously shortening it to save keystrokes.

If you work in a SOAP structure, we covered the section-by-section approach in detail in how to dictate SOAP notes faster, and the method transfers directly to an Epic note template.

2. In Basket messages

In Basket is where charting time quietly goes to die. Individually each message is short. Collectively, result notes, patient advice requests, staff messages, and refill questions can consume a substantial part of a day, and every one of them is prose that has to be written by a clinician.

These are ideal for voice precisely because they are short. Two or three spoken sentences per message, sent, next. You are not composing a document, you are answering a question, and answering a question aloud is something you have done thousands of times today already.

3. Patient instructions and the after-visit summary

Patient-facing instructions are the section most likely to be left generic, because writing tailored instructions takes time and the boilerplate is already there. That is a real loss for the patient. Three specific spoken sentences about what this patient should watch for, when to call, and what changes today beats a template paragraph that could apply to anyone.

This is also the surface where speaking helps the writing itself. Instructions dictated aloud come out in plainer language than instructions typed, because you naturally speak to a person rather than composing for a chart.

4. Telephone and result-note documentation

The call happened, the result came back, and now it needs a line of narrative saying what you decided and why. These notes are almost always written from memory hours later, and they get thinner the longer you wait. Speaking one takes fifteen seconds while the call is still in your head.

5. Everything around the chart

Referral letters, secure messages to colleagues, prior authorization narratives, and the email you owe the practice manager all live outside Epic but inside the same workday. Because dictation runs at the Mac level rather than inside one application, the same hotkey works in your mail client, your browser, and your documents without any change of tool.

Speak the narrative, type the discrete fields

This is the rule that separates clinicians who make dictation stick from clinicians who abandon it in week two.

Prose defends itself. If a word comes out wrong in your assessment, you see it, because the sentence stops making sense. Discrete data does not defend itself. A dose, a lab value, a blood pressure, a date, a medical record number, a code: if a digit lands wrong, the field still looks completely plausible. It reads as correct forever, and it propagates into every downstream place that field feeds.

So the boundary is not about capability, it is about the cost of a silent error. Type anything numeric or coded that another system will consume. Speak the reasoning that only you can supply.

A wrong word in your narrative announces itself. A wrong digit in a dose does not.

The same logic applies to orders. Do not dictate into an order field and move on. Ordering is a discrete, safety-critical workflow with its own confirmation steps, and those steps exist for a reason.

Keeping SmartPhrases and templates intact

Most clinicians who have used Epic for any length of time have built a personal library of SmartPhrases, and those dot phrases are the thing they least want to lose. The good news is you do not lose them. The important detail is the order of operations.

Type the dot phrase. Let it expand. Then dictate into the placeholders it leaves behind.

Trying to speak a dot phrase is the wrong move for a mechanical reason: expansion is triggered by the way the client sees the characters arrive and by the accept keystroke that follows, and spoken punctuation is unreliable at reproducing that exactly. It is also unnecessary. A dot phrase is six characters. Typing it is faster than saying it correctly.

Once the template has expanded, you have a structure full of placeholders waiting for content, and filling those placeholders is exactly what voice is good at. Jump to the first one with your usual navigation key, speak the sentence that belongs there, jump to the next. The template gives you the skeleton, your voice gives you the substance, and the note comes out both structured and specific instead of structured and empty.

One caution worth stating plainly: watch what happens the first time you dictate into a field where a template control has focus rather than a plain text area. Behavior varies by build. Test on a scratch note, not on a real chart.

The vocabulary problem, and the fix

Clinical language is where general-purpose dictation tools usually lose people. Drug names collide with ordinary English. Referring physicians have surnames that no transcription engine has a reason to guess. Your organization has internal names for clinics, units, protocols, and forms that exist nowhere outside your walls.

The wrong response is to correct the same three words every day for a year. The right response is to fix them once, at the source.

Voice Keyboard Pro's Smart Vocabulary is a personal dictionary with replacement rules. You add the terms you actually use, in the spellings you actually need, and they come out right from then on. A realistic starter list for most clinicians is short: your prescribing set, the ten or fifteen referring names you write most often, your local unit and clinic names, and any abbreviation that keeps expanding wrongly.

Build it opportunistically. Every time you fix the same word twice, add it. Within two weeks the list stops growing and the corrections stop happening. Our post on handling drug names and coding vocabulary goes deeper on how to structure that list.

The privacy and compliance boundary

This section is deliberately blunt, because vague reassurance is worse than useless in a clinical setting.

Voice Keyboard Pro's servers store operational pings only. No audio and no transcript content leave as stored records. That is a real structural property of the product and it is a meaningful one, but it is not the same thing as a compliance decision, and we will not pretend otherwise.

If you intend to dictate content that includes protected health information, the questions that matter are organizational: whether a Business Associate Agreement is in place, what your institution permits on managed devices, and how your Mac itself is secured. Those are decisions for your compliance officer and your IT team, not for a vendor's blog post, and not for you alone. We wrote about the shape of that conversation in HIPAA-minded voice to text for clinical notes, including why any claim of a HIPAA "certification" should be treated with suspicion.

There is also a non-technical exposure that no software addresses. Dictation is audible. In a shared workroom, at a nursing station, or in a hallway, the risk is the person two metres away, not the network. Speak clinical detail where you would say it aloud anyway, and nowhere else.

What about Epic on the iPhone?

Honest answer: proceed carefully and check with your organization first.

Voice Keyboard Pro's iPhone product is a custom keyboard with a built-in mic button, which works in text fields across iOS apps. But clinical mobile apps and the mobile device management profiles that healthcare organizations deploy frequently restrict third-party keyboards outright, and some fields refuse them by design. If your organization's policy prohibits third-party keyboards on devices that touch patient data, that policy is the answer, and there is nothing to configure around it.

Where the iPhone keyboard does earn its place for clinicians is the non-clinical half of the day: capturing a thought between rooms, replying to a colleague, drafting the paragraph you owe someone before you forget it. That is genuine value, and it does not require putting patient detail on a phone.

Common questions

Do I need my IT department to install anything?

Not on the Epic side. The app installs on your Mac like any other Mac application and nothing is added to Hyperspace, to the Citrix server, or to your organization's Epic build. Whether you are permitted to install software on a managed Mac is a separate question, and your IT policy governs it.

Will this interfere with the dictation system my hospital already licenses?

They operate at different layers, so in practice they coexist, but the two things worth checking are hotkey conflicts and microphone contention. Do not run both listening at once. Pick one for a given task.

Does it work if my Epic access is through a browser?

Yes, and it is simpler than the remote-session case, because a browser text field is an ordinary text field. Click in, hold, speak, release. The same caveat about clicking into the field first still applies.

Can I dictate while the session is laggy?

Recognition happens on the Mac, so session lag does not degrade accuracy. What lag affects is delivery: on a poor connection, text arrives into the session more slowly. Speak in shorter bursts and let each one land.

Does it work offline?

No. Transcription requires a connection. In practice this is rarely the binding constraint, since Epic through a remote session requires a connection too.

Should I dictate the whole note?

No, and clinicians who try usually quit. Dictate the narrative, type the discrete data, let templates handle the structure. The goal is not to eliminate the keyboard. It is to stop typing the part of the note that costs the most and carries the most meaning.

The realistic bottom line

Dictation in Epic on a Mac fails for most people on the first attempt because they try to send their voice somewhere it cannot go. Once you flip the architecture and let the recognition happen locally while the session receives ordinary keystrokes, the technical problem mostly disappears and what remains is a workflow question.

Answer that question conservatively. Speak the history of present illness, the assessment and plan, the In Basket replies, and the patient instructions. Type the doses, values, codes, and dates. Fix your recurring vocabulary once instead of every day. Settle the compliance question with the people whose job it is before patient detail is involved.

Done that way, the thing you get back is not a technology upgrade. It is the part of the evening you were spending on notes. Voice Keyboard Pro has a free tier with daily limits, which is enough to test the whole architecture on a scratch note before you commit to anything. Pro is $4.99 a month or $34.99 a year.