Epic AI vs. Microsoft Copilot: a decision guide for clinicians

Last reviewed: August 2026

Epic AI and Microsoft Copilot solve different problems, and the split is cleaner than most rollouts make it sound. Epic AI works inside the electronic health record, where patient data, documentation, and billing live, so it handles charting, coding suggestions, and patient messages. Microsoft Copilot works outside the EHR, across Word, Excel, Teams, and Outlook, so it handles meeting prep, research synthesis, presentations, and education materials. The deciding question is one sentence: does this task touch a patient chart or billing? If yes, stay in Epic. If no, Copilot is probably the faster tool.

I built this framework because clinicians at the bedside kept asking the same question in different words: "which AI am I supposed to use for this?" Hospitals are deploying both tools at once, usually with two separate training decks and no bridge between them. This guide is the bridge.

Table of contents

• What AI tools actually live inside Epic?
• Who are Emmie and Penny?
• Where does Microsoft Copilot fit in a hospital?
• How do you decide which tool to use in the moment?
• What about PHI and HIPAA?
• What are the limits of each tool?
• Frequently asked questions
• Where this fits, and what to do next
• Further reading

What AI tools actually live inside Epic?

Epic's generative AI suite is embedded in the workflows clinicians already use, which is exactly why it is the right home for anything chart-adjacent. As of HIMSS 2026, Epic reports that more than 85% of its customer organizations actively use its AI tools, so if you work in an Epic shop, some of this is likely already turned on (HIT Consultant).

The suite clusters into four groups:

Documentation and charting. Ambient voice documentation for inpatient and outpatient notes, drafted hospital course sections for discharge summaries, invasive procedure narratives, and end-of-shift care plan notes. Epic's clinician-facing assistant is called Art, and its ambient charting is now expanding into bedside nursing (Epic, Art for Clinicians). Nurses at Mayo Clinic save about 30 seconds per patient message with drafted replies, and clinicians at John Muir Health report saving 34 minutes a day on notes. Those numbers come from Epic's own customer reporting, so treat them as vendor-published rather than independent, but the direction matches what peer-reviewed work is finding.

Coding and billing. Clinical documentation improvement suggestions, CDI nudges for physicians, level of service suggestions, and risk adjustment coding assistants that run before and after visits.

Insights and extraction. Inpatient and outpatient insights, extraction of social drivers of health from notes, and cancer staging extraction.

Communication. In Basket Art drafts responses to patient messages, results notes, and translations, which the clinician reviews, edits, and sends.

The pattern across all four groups is the same: Epic AI drafts, the clinician judges. Nothing goes to a patient or a chart without your review, and that review is not a formality. It is the safety mechanism.

Who are Emmie and Penny?

Emmie and Penny are Epic's named AI agents, announced at Epic's 2025 user group meeting alongside Art (Advisory Board).

Emmie is the patient-facing agent inside MyChart. She chats with patients before a visit to surface needs, collect history, and confirm that pre-visit tasks like labs are done, then hands the clinician a summary. After the visit she helps patients understand their care plan and complete follow-up documents. She also drafts responses to routine patient inquiries for provider review. You will rarely open Emmie yourself, but she changes your day anyway, because work that used to arrive as inbox messages gets resolved before it reaches you. Rush University Medical Center saw a 58% sustained reduction in billing-related customer service messages after deploying her (Epic).

Penny is the revenue cycle agent. She drafts appeal letters for denied claims by pulling relevant clinical evidence from the chart, generates medical necessity justifications for utilization review, summarizes clinical information for prior authorization requests, and suggests codes based on documentation. Summit Health cut prior authorization submission time by 42% with Penny, and 92% of her drafted responses were accepted without edits (Epic).

For nurses, the practical takeaway is simple: Emmie and Penny run in the background of your organization. Your job is knowing they exist, so that when documentation you wrote feeds an appeal letter or a prior auth summary, you understand where it went and why accuracy in the original note matters twice.

Where does Microsoft Copilot fit in a hospital?

Microsoft Copilot is the tool for everything that happens outside the chart. It lives across Word, Excel, Teams, Outlook, and PowerPoint, and it is genuinely good at the administrative layer of clinical work (Microsoft, healthcare scenario library). Be careful with the name, because three different Microsoft products carry it: Microsoft 365 Copilot for work outside the chart, Dragon Copilot for ambient clinical documentation that writes into the chart, and a consumer Copilot Health preview that has no place in patient care.

Concretely, that looks like:

• Summarizing staffing meetings in Teams and drafting the follow-up email to leadership
• Turning unit metrics in a spreadsheet into a readable report or a presentation
• Synthesizing literature for a committee, a conference abstract, or a practice council
• Drafting patient education materials, including multilingual versions, from approved source content
• Building schedules, checklists, and templates that would otherwise eat an afternoon

The scale of this shift is not small. NHS England is rolling Copilot out to more than 505,000 clinicians and staff after a trial found it saved an average of 43 minutes of administrative time per person per day (Healthcare Digital). Microsoft and Epic also have a standing collaboration, so the two ecosystems are designed to coexist rather than compete (Microsoft).

Sample prompts that earn their keep:

• "Summarize key staffing challenges from these meeting notes and draft an email for leadership."
• "Create a patient-friendly guide for managing hypertension after discharge."
• "Generate a PowerPoint summarizing ED wait times and readmission trends from this spreadsheet."

Notice what those prompts have in common: none of them contains a patient's name, MRN, or chart data. That is the line, and the next section is about why.

How do you decide which tool to use in the moment?

Ask one question: does this task touch a patient chart or billing? Chart or billing means Epic AI. Everything else means Copilot. Here is the framework as a scenario table:

Three habits make the framework stick:

1. Use Epic AI for anything that touches patient charts or billing.
2. Use Copilot for meeting prep, research, and creating education materials.
3. Validate every AI output before sharing it or acting on it, in either tool.

There is a one-page version of this framework you can keep on your phone or hand to your unit educator: download the ten-card summary.

What about PHI and HIPAA?

This is the part of the framework with no gray zone. Never enter protected health information into Copilot unless your organization has explicitly approved a secure, HIPAA-covered configuration for it. Consumer AI tools do not sign Business Associate Agreements, and entering PHI into a tool without one is a reportable problem, not a shortcut (HIPAA Journal).

Epic AI sits inside your organization's existing HIPAA infrastructure, which is why chart-connected tasks belong there. Copilot deployments vary by organization: some are enterprise configurations with agreements in place, and some are not. You will not know by looking at the interface, so the operating rule for clinicians is to keep patient-identifiable data out of Copilot prompts entirely and let your organization's policy, not the tool's marketing, tell you when that changes.

De-identified work is fine. "Draft a hypertension discharge guide" needs no PHI. "Draft discharge instructions for Mr. Alvarez in bed 12" does, and that request belongs in Epic or nowhere.

What are the limits of each tool?

Epic AI's strengths are deep EHR integration, real-time patient context, and measurable gains in documentation and coding accuracy. Its limits are flexibility, since it is built for EHR workflows and little else, plus alert fatigue risk and its constant need for clinician oversight.

Copilot's strengths are its reach across Microsoft 365 and its speed at summarization, drafting, and analysis. Its limits are the mirror image: it has no connection to Epic patient data, it carries HIPAA risk if PHI is entered, and its clinical claims need validation because it was never designed as a clinical tool.

The oversight point deserves one more beat, because it applies to both tools and it is where the research keeps landing. A randomized trial of two ambient AI scribes found documentation time savings alongside occasional inaccuracies that required ongoing vigilance (medRxiv preprint, 2025). A 2026 JAMIA study found that ambient scribes improved clinician engagement with patients but created new reconciliation work, since clinicians had to check AI text against their own documentation standards (JAMIA). The tools are useful and the tools are fallible, and both things stay true at the same time. Your review is the control.

Frequently asked questions

Can I use Copilot to write my nursing notes? No. Clinical documentation belongs in Epic, where ambient documentation and drafting tools are integrated with the chart and covered by your organization's HIPAA infrastructure. Copilot has no access to Epic data, and pasting chart content into it creates PHI exposure.

Do I need to learn both tools? Realistically, yes, but not equally. Learn the Epic AI features in your role's workflow first, since they touch patient care. Then learn three or four Copilot patterns for your administrative work, like meeting summaries and report drafting. That covers most days.

Whose job is it to check AI output? Yours, every time. Both Epic and Microsoft position their tools as draft generators with human review, and the clinical research so far supports exactly that framing. An unreviewed AI draft is an unfinished task.

What if my hospital has not deployed these tools yet? The framework still helps, because it tells you what to ask for and what to ask about. Good first questions for your informatics team: which Epic AI features are turned on for my role, and is our Copilot deployment approved for any PHI at all?

Where this fits, and what to do next

You can run this decision framework tomorrow without buying anything: sort your next shift's tasks into chart-touching and everything else, route them accordingly, and validate every output before it moves. If your organization is mid-rollout, share the scenario table with your unit educator, because most AI friction on the floor is really a routing problem.

If you want the deeper version, charmthirteen teaches it. Chart Smarter is the course for nursing and clinical education teams, and it covers this same ground with more room: the zero-PHI rule, safe prompt structure, spotting hallucinations, drift, and bias, ambient documentation workflows, and managing alert fatigue, with human-in-the-loop vigilance throughout. It is CE-eligible pending an accreditation partnership. For the people who own the policy rather than the shift, Guardrails covers governance, procurement, and vendor evaluation for teams buying healthcare AI. Both live on the workshops page, and both get scoped for your team in a free Vibe Check call. When the gap is bigger than training, we also run healthcare AI consulting and AI governance engagements.

Further reading

1. Real Results, Right Now: How Epic AI Is Reducing Costs, Improving Care, and Helping Patients (Epic): Epic's own outcomes reporting for Art, Penny, and Emmie, including the Rush and Summit Health numbers cited above. Vendor-published, so read it as a capability map with marketing gravity.
2. Art for Clinicians (Epic): the current feature list for Epic's clinician-facing AI, useful for checking what your organization could have turned on.
3. Epic debuts new AI tools at annual meeting (Advisory Board): independent coverage of the UGM 2025 announcements where Emmie, Penny, and Art were introduced.
4. The Agentic EHR: Epic Unveils Agent Factory and Custom Models (HIT Consultant): where this is heading next: health systems building their own agents on Epic's platform.
5. Microsoft and Epic expand AI collaboration (Microsoft): the partnership that explains why the two ecosystems are built to coexist.
6. Healthcare scenario library (Microsoft Adoption): Microsoft's catalog of Copilot healthcare use cases, organized by function.
7. NHS deploys Microsoft AI to cut administrative burden (Healthcare Digital): the 505,000-clinician rollout and the 43-minutes-per-day trial finding.
8. HIPAA, Healthcare Data, and Artificial Intelligence (HIPAA Journal): how HIPAA's technology-neutral rules apply to generative AI, and what workforce training should cover. Read this one first if you only read one.
9. A Randomized-Clinical Trial of Two Ambient Artificial Intelligence Scribes (medRxiv preprint, 2025): documentation efficiency and burnout findings, with the honest caveat that inaccuracies occurred and vigilance stayed necessary. Preprint, not yet peer-reviewed.
10. Listening to the note: clinician perspectives on ambient AI scribes (JAMIA, 2026): the peer-reviewed view of what ambient AI actually changes about clinician work, including the new review burden.

Changelog

• 2026-08-16: First published. Framework from the author's clinician AI usage playbook; Emmie and Penny profiles added; references verified live as of this date.