Exam Room Setup for AI Scribes: Microphone, Wifi, and Patient Positioning

The boring setup that determines everything

A lot of AI scribe rollouts fail not because the AI is bad but because the audio coming into it is bad. The vendor's accuracy benchmarks were measured with clean studio-quality audio. Your exam room has an HVAC vent running, a sink that gets used, a hallway door that opens, a baby crying through the wall, and a laptop with a built-in microphone sitting four feet from the patient.

The gap between the vendor's accuracy and your real-world accuracy is mostly explained by audio quality. Fixing the audio is the highest-leverage thing you can do, and most practices skip it because it feels like a setup detail rather than a clinical decision.

This article is the audio setup guide. What microphone to use, where to put it, what wifi setup to verify, and how the exam room layout affects the result.

The microphone choice that matters most

A laptop's built-in microphone is the default for most early adopters. It's also the worst common option. The mic is designed for video calls where the user is one to two feet from the screen, not for capturing two-way conversation across a four-foot exam room.

Three better options, in order of cost and quality:

Option 1: A USB lapel mic clipped to the provider's coat. Cost: $40 to $80. Pros: captures the provider's voice very clearly, follows them around the room. Cons: only captures one side of the conversation well; patient audio is still distant. Best for visits where the provider does most of the talking (procedure narration, dictation-style visits) or when the patient sits close to the provider.

Option 2: A USB conference mic on the desk or counter. Cost: $80 to $200. Pros: captures both provider and patient evenly, no wearable hardware, durable. Cons: needs to be positioned centrally between provider and patient. Best for typical primary care, internal medicine, and most outpatient specialties.

Option 3: A dedicated dual-channel system with separate mics for provider and patient. Cost: $300 to $800. Pros: best accuracy, especially for patients with quiet voices or accents. Cons: more cables, more setup, more things to break. Best for psychiatry with longer visits, specialty consults with detailed history-taking, and clinics where audio accuracy has been the dominant complaint.

For 80% of primary care practices, option 2 is the right call. It's a one-time $100 to $150 investment that improves AI accuracy noticeably. Decent options in 2026 include the Jabra Speak 410, the eMeet M2, and the Anker PowerConf S330.

Where to put the microphone

The microphone choice matters; the placement matters almost as much. A great mic in the wrong place underperforms a mediocre mic in the right place.

For a desk-style USB conference mic, the placement that works:

• On a flat surface, not on a surface that vibrates. A wood desk is fine; the EKG cart that has wheels is not, because the wheels translate every footstep into mic noise.
• Equidistant between provider's typical seating position and the patient's typical seating position. If the provider sits at the desk and the patient sits in the chair next to the desk, the mic goes on the corner of the desk closest to the patient chair, not in the middle of the desk where it's farther from both.
• At chest height for seated speakers. Most mics work best at chest height; mics on the floor or on a high shelf perform worse because the geometry of voice projection is downward and outward, not upward.
• Away from the HVAC vent. A mic placed directly under a ceiling vent picks up the fan noise more than the conversation. Move the mic across the room from the vent if there's a choice.
• Away from the laptop fan. Modern laptops have audible fans that the mic will pick up. Distance the mic at least two feet from the laptop body.

For a lapel mic, the placement that works:

• Clipped to the provider's coat near the second shirt button. Higher up the chest gives better voice capture than lower.
• Cable routed under the coat, not draped over. Cable rustle is a real and annoying source of audio artifacts.
• The mic should NOT face the patient. Counterintuitively, the mic faces upward toward the provider's mouth. The mic still picks up the patient's voice through ambient capture, but the provider's voice (the more important channel for diarization) is clear.

The wifi reality check

Most AI scribes stream audio to the cloud in real time or near-real time. Wifi reliability in exam rooms varies dramatically. Many practices discover during onboarding that the exam room at the end of the hall has marginal wifi that drops the AI session every few minutes.

The diagnostic that catches this:

1. Stand in the exam room with a phone or laptop.
2. Run a speed test (fast.com or speedtest.net).
3. Wait 30 seconds.
4. Run it again.
5. Repeat for each exam room.

You're looking for two things:

Consistent throughput of at least 5 Mbps upload. Most AI scribes need 1 to 3 Mbps to stream audio reliably. 5 Mbps gives margin. If any exam room is under 5 Mbps consistently, that room will be a friction point.

Stability over time. A room that shows 50 Mbps on one test and 2 Mbps on the next has a bad signal even though the peak is fine. Stability matters more than peak speed.

For rooms that fail the diagnostic, three remediations:

• Move the wifi access point closer. Often the access point is in the wrong physical location for the actual usage. Moving it 15 feet can resolve a problem.
• Add a second access point. For larger clinics, one access point is rarely enough. A mesh wifi system covering all exam rooms costs $200 to $500 and pays for itself in avoided friction.
• Run ethernet to the laptop. If wifi can't be made reliable, ethernet always works. A USB-C ethernet adapter and a long cable solve the problem for $30, with the downside of cable management.

A practice with multiple AI scribe failures per day from connection issues is a practice with a wifi problem, not an AI problem. Diagnose before assuming the vendor is at fault.

Exam room layout that helps

The physical layout of the exam room affects audio capture more than most practices realize. Layouts that work well for AI scribes share a few features:

Patient seated, not on the exam table during the conversational part of the visit. Voice projects forward when seated; voice on the exam table often projects to the ceiling or to the wall. Many practices already seat patients during history-taking and move to the exam table for the physical exam; this also helps audio quality.

Provider seated facing the patient, not facing the laptop. When the provider faces the laptop, their voice projects toward the wall behind the laptop instead of into the room. A small adjustment in seating geometry can improve provider voice capture meaningfully.

Hard surfaces softened where possible. A room with all hard surfaces (drywall, linoleum, no rug) bounces sound and creates echo that degrades audio quality. A small area rug, a fabric chair, or even acoustic panels on one wall can dampen the room significantly.

The door closed during conversational time. An open door admits hallway noise (other patients, staff conversations, intercom announcements). The door closes anyway during exams; closing it earlier helps audio.

None of these require renovation. They're seating choices and minor furniture additions.

Special cases

Pediatric exam rooms. Children move around, parents add a third speaker, and toys make noise. A lapel mic on the provider plus a smaller secondary mic clipped to the parent (if they consent) sometimes outperforms a single conference mic. Accept that pediatric audio will always be harder than adult audio.

Telehealth visits. Wifi reliability shifts from "yours" to "the patient's." Test the patient's audio quality at the start of the visit ("can you hear me clearly?") and bail on AI scribing if their audio is bad. Bad audio from the patient end is a degraded note, not a saved one.

Procedure rooms. Procedure rooms have suction, monitors, music sometimes. A throat-mounted mic for the provider (rare but available) outperforms a room mic in these environments. For procedures, dictating the procedural narrative explicitly tends to work better than ambient capture, regardless of mic choice.

Shared exam rooms. A clinic where multiple providers rotate through the same exam rooms benefits from having the audio hardware live in the room, not travel with the provider. A counter-mounted conference mic and a docking station for the laptop means each provider can sit down and start without setup time.

The end-of-visit audio check

For the first two weeks of using an AI scribe, run a small end-of-visit audio check on a random sample of visits:

1. Open the AI scribe's audio playback or transcript.
2. Listen to 30 seconds from the middle of the visit.
3. Rate: can you clearly hear both the provider and the patient?

If the answer is "no" on more than 1 in 5 visits, the audio setup needs attention. Most accuracy complaints during early adoption trace back to audio quality, and fixing audio fixes the complaint without any platform changes.

After two weeks, this check is no longer necessary; the audio either works or it doesn't. But during the early calibration window, the audio check is the single most useful diagnostic.

What the vendor won't tell you

AI scribe vendors usually don't push hard on audio setup because it makes their product sound more demanding. The vendor wants you to believe their AI is robust to typical room audio. It is, mostly; but the gap between "robust to typical audio" and "best accuracy with optimized audio" is wider than the vendor advertises.

Practices that invest the $100 to $300 in audio hardware in the first week typically see meaningfully better accuracy than practices that don't. The vendor won't tell you this is necessary because they don't want it to be a barrier to adoption. It isn't a barrier; the laptop's built-in mic does work. But it works at 85% accuracy when the same room with a better mic would work at 95%, and that 10 percentage points is the difference between a tool you trust and a tool you double-check.

The 30-minute setup check

If you're starting with an AI scribe and you have 30 minutes to invest in setup, spend it on:

• 5 minutes choosing a microphone (use the recommendations above)
• 10 minutes running wifi speed tests in every exam room
• 10 minutes physically arranging the mic and adjusting seating in your two highest-use rooms
• 5 minutes doing one practice recording with a colleague and listening to the playback

That 30 minutes typically returns 30 minutes of saved frustration in the first month alone. It is the single highest-leverage piece of preparation in the entire rollout.

---

For the broader first-week onboarding plan that audio setup fits into, see the first week with an AI scribe. For why audio quality matters specifically for medication accuracy, what to do when your AI scribe mishears a medication covers the medication-specific failure modes.