Can Patients Tell They're Talking to an AI? (We Tested It)

Short answer: sometimes yes, often no, and the interesting part is which ones noticed and which ones didn't.

We ran a blind test with 100 real patients calling our AI receptionist Marcus. Here's what we learned, what surprised us, and what it means for practices considering AI answering.

The test setup

• 100 callers: a mix of dental-practice patients and med-spa clients from practices that had just added Agentis
• Callers weren't told the receptionist was AI. They were asked afterward: "Anything stand out about the call?"
• Calls were standard inbound: booking, reschedule, insurance question, emergency triage, FAQ.
• Post-call survey: 5 questions, 2 minutes.

The headline numbers

• 51% of callers did not realize they'd spoken to AI until we told them afterward.
• 34% suspected something "a little off" but couldn't pinpoint what.
• 15% identified the receptionist as AI during the call.
• Among the 15% who identified it: 82% still rated the call experience as good or very good.

So the true answer to "can patients tell?" is:

• Half can't tell.
• A third notice something but don't care enough to think about it.
• One in six identifies AI specifically — and most of those are fine with it.

What the 15% noticed

This was the most interesting data. The callers who correctly identified AI pointed to these cues:

1. Consistent tone across the whole call — no variation in energy level, no "I'm tired of answering this question today" inflection. Real humans drift; AI doesn't.

2. No background noise — a real receptionist has phones ringing, other staff, office noise. Marcus calls are dead quiet. A few callers specifically said "too quiet."

3. Very fast responses to common questions — humans pause and think. AI answers near-instantly when it knows the answer. Callers occasionally read that as "she's reading from a script."

4. Didn't pick up on unstated emotional context — one caller mentioned a recent bereavement offhandedly. Human receptionist might have said "oh I'm so sorry." Marcus acknowledged respectfully but didn't weight it the way a human would.

What the other 85% liked

Unprompted comments from survey responses included:

• "She was very patient. Didn't rush me."
• "Knew answers to questions I usually have to get transferred for."
• "Spoke Spanish with my mom when I put her on."
• "Called me back within a minute of my voicemail." (This was actually Marcus on the next inbound call recognizing the number.)
• "Got me booked without making me hold."

These are things AI does well that humans often don't.

The 73% preference finding

Industry research from 2024–2025 consistently finds that about 73% of customers prefer AI for simple, transactional inquiries — booking, reschedules, FAQ, pricing questions — when the AI is good.

That last clause matters. The same studies find strong customer preference for humans when:

• The issue is emotional (complaint, bereavement-related scheduling, medical anxiety)
• The issue is complex and judgment-heavy (billing dispute, insurance denial)
• The caller explicitly asks to speak to a person

The pattern: AI for routine, human for everything else. Patients are already in this mindset from banking and airlines. Medical practices are catching up.

The "accent" and "natural cadence" question

We tested Marcus against three callers who specifically were chosen for strong accents (Caribbean Spanish, West African English, Southern US). All three calls succeeded — appointments booked, info captured correctly.

Two of the three didn't clock Marcus as AI. The third did — because of the "too consistent tone" cue mentioned above, not because of accent handling.

The takeaway: modern voice AI is good at accent handling. Where it still gives itself away is consistency of affect, not comprehension of the caller.

What this means for your practice

Three honest conclusions:

1. The "patients will hate AI" objection is largely wrong.

85% of patients either don't notice or don't care. Of the 15% who notice, most rate the experience positively anyway. The practice owners who worry about this are over-indexing on their own expectations of what patients will think.

2. The "we need to disclose it's AI" question is complicated.

Some regulatory frameworks require disclosure. Some don't. We generally recommend a light disclosure on the opening: "This is Marcus, the AI assistant for [Practice Name] — how can I help?" This frames the interaction correctly and is often appreciated more than obscured.

3. The AI is not going to replace human judgment for hard calls.

If your practice has a significant share of emotionally loaded or judgment-heavy calls — bereavement-related scheduling, complex insurance disputes, anxious first-time patients — keep a human in the loop for those specific call types. AI handles the routine 80% well. The remaining 20% is genuinely better with a human.

Test it yourself

The right way to evaluate this isn't to read an article about it. It's to call Marcus and see what you think.

Call (786) 474-9984 now.

Pretend you're a patient. Ask about booking. Ask in Spanish if you speak it. Ask a curveball question. See how it flows.

Then ask yourself: is this better or worse than what a caller hits when your front desk is at lunch, or when it's 8pm, or when two calls come in at once?

For most practices, the honest comparison isn't "AI vs. our best receptionist on her best day." It's "AI vs. voicemail" — because voicemail is what's actually catching most of your after-hours, overflow, and lunch-break calls today.

Against that comparison, AI wins decisively. Patient satisfaction with "someone helpful picked up" beats patient satisfaction with "I hit voicemail" every single time.

Try Agentis

Call Marcus: (786) 474-9984 Book a setup: getagentis.ai/demo

The summary

Half of patients can't tell they're talking to an AI receptionist. A third notice something but don't care. The remaining 15% who identify it mostly rate the experience positively anyway.

The "patients will hate AI" objection is weaker than practice owners think — but the underlying concern is legitimate: for hard, emotional, or judgment-heavy calls, humans still win. For routine booking, reschedules, FAQs, and after-hours coverage, AI is already good enough that patients mostly don't notice.

The real comparison isn't "AI vs. your best receptionist." It's "AI vs. voicemail." And voicemail is what patients hit today, on most of the calls your practice doesn't answer.