What an AI Receptionist Actually Does on a Patient Call, Step by Step

An AI receptionist on a practice line does five things on every call, in order: it answers, it works out who is calling and why, it reads the live schedule, it writes the result into your EHR (a booking, a change, a message, or a transfer), and it stops when the call is outside its rules. Everything else you read about “capabilities” is a variation on those five steps. This article walks one call end to end and shows what the agent has to read and write at each step, because that is where an AI receptionist either works or quietly fails.

The reason it matters: across the medical practice phone lines KloudMD tracks, 39% of inbound calls in the last twelve months ended without a live answer (sample: roughly 180,000 inbound calls to several hundred medical practices; aggregated, de-identified KloudMD platform data, Sep 2025 to Sep 2026). After hours the missed rate was 55%, and on weekends 56%. An AI receptionist is only useful if the calls it picks up end in something staff do not have to redo in the morning.

Step 1: Pick up and set expectations

The agent answers on the first ring, as the practice, with the greeting you wrote. It does not pretend to be a person, and if a caller asks, it says it is an AI. That is a policy choice, not a technical limit: a receptionist that lies about being human creates a trust problem the first time a patient figures it out.

At this point the agent has one piece of data: the caller ID. In a well-built system that number is already matched against your contact records before the agent says a word, so a returning patient is greeted with their name, and a new number is treated as a new patient. The match is a starting point, not proof of identity. Because families share phones, the agent will still confirm name and date of birth before touching a chart (more on that in step 3).

Step 2: Understand the request in free speech

This is the part that separates an AI receptionist from a phone tree. The caller says “I need to move my cleaning next Tuesday to later in the week” and the agent has to extract four things: the intent (reschedule), the appointment being referenced (the cleaning next Tuesday), the constraint (later that week), and anything missing (which day, which time of day).

The requests that come up on a practice line fall into a short list:

• Book a new appointment (new or existing patient)
• Reschedule or cancel an existing appointment
• Confirm an appointment or ask what time it is
• Hours, directions, parking, what to bring
• “Do you take my insurance?”
• Prescription, results, or a clinical question
• Billing question
• Something urgent

The first five are answerable from data the agent can read. The last three are either routed to a person or handled with a message. A good agent knows which list an utterance belongs to before it says anything back.

Step 3: Identify the patient properly

Before the agent books, changes, or reads anything about an appointment, it confirms who the patient is. The rule that matters most here comes from EHR integration work, not from the phone side: a phone number or email identifies a household, not a patient. A mother calling about her son’s appointment calls from her own number. If the agent matched on phone alone, it would book the mother.

So the agent asks for the patient’s name and date of birth and uses those, together with the contact, to find the right chart. Some EHRs enforce this at the door: ModMed and athenahealth will not create a patient chart without a real date of birth, so any booking path that feeds them, phone or web, has to collect DOB. The same rule is why a dependent can be booked under a parent’s phone number and still land on the child’s chart, with the parent recorded as the responsible party.

Step 4: Read the live schedule

Now the agent needs real availability, and “real” means three things at once:

• The right service. A new-patient consult is not a follow-up. Each has its own duration and its own set of providers who perform it.
• The right provider and location. Only providers who actually do that service at that office, on the days they work there.
• Minus what is already booked. Slots that are free on the provider’s template but taken in the EHR are not offered.

Where this data comes from depends on your software. Cloud EHRs with an API expose provider hours and booked appointments directly. On-premise systems such as Dentrix and ChiroTouch have no cloud API and expose no provider schedules at all, so working hours are configured manually and booked appointments are read through a connector installed in the office. Either way, the agent should be reading the same scheduling layer your online booking page uses. If the phone agent and the web booking page compute availability differently, you will double-book.

The agent then offers two or three concrete options (“I have Thursday at 2:10 or Friday at 9:40 with Dr. A”) rather than reading a list, and it offers times on the EHR’s own grid. If your appointment book runs on 10-minute increments, an agent offering only :00 and :30 will hide bookable time.

Step 5: Write the result to the EHR while the caller is on the line

This is the step most callers never think about and most vendors gloss over. Once the patient picks a time, the agent has to create the appointment in the EHR with the correct provider, service code, duration, and location, on the correct patient chart, and get confirmation back before it tells the caller “you’re booked.”

The failure mode to ask about: does the agent wait for the EHR to accept the booking, or does it say “done” and hope the write goes through later? If the EHR rejects the slot because it was taken a moment ago, the agent should be able to say so and offer the next option. That only works when the write happens during the call.

Reschedule and cancel follow the same path but are harder in practice, because the agent has to update or cancel the same appointment record in the EHR, not create a second one. That is covered in detail in why reschedule and cancel are the hard part.

Step 6: Know when to stop

An AI receptionist earns its keep by handling the routine calls, and it stays safe by refusing the rest. The stop rules on a healthcare line are:

Call type	What the agent does
Clinical question or advice	Does not answer. Documents the request and routes it to your team.
Emergency or crisis language	Directs to 911 or 988 per your rules, then notifies on-call.
Call types you mark “escalate”	Warm transfer during clinic hours; detailed message plus follow-up task after hours.
Anything it cannot resolve	Takes a message with a callback number and creates a task, instead of guessing.

The important word is “your rules.” Which call types page on-call, which transfer, and which become a morning task is configuration, not something the vendor decides for you.

Step 7: Leave a record

When the call ends, staff should find three things without doing anything: the appointment (or change) already in the EHR, a transcript and recording attached to the patient’s record, and a task for anything the agent could not close. Compare that to a live answering service, where the morning routine is re-typing messages from a portal into the schedule.

What the call volume data says about where this helps

The same platform data shows where an always-on receptionist changes outcomes most. Missed rates by hour are flat at 35 to 38% from 8am to 4pm, then climb: 42% at 4pm, 46% at 5pm, 53% at 6pm, and 55% at 8pm. Early mornings are worse (56% missed at 7am, 65% at 6am). Weekend calls are missed 56% of the time.

Window (practice local time)	Share of inbound calls	Missed
Business hours, 8am to 5:59pm	89%	37%
After hours, 6pm to 7:59am	6%	55%
Weekends	5%	56%

Source: aggregated, de-identified KloudMD platform data across several hundred medical practices, Sep 2025 to Sep 2026.

Note the business-hours figure. Most missed calls do not happen after hours; they happen while the office is open, because the desk is already on another line and business hours carry the overwhelming majority of the volume. An AI receptionist used as overflow during the day, not only as an after-hours service, is where most of the recovered calls come from. For the by-day breakdown see which days and hours practice phones are busiest.

Questions to ask any vendor

• Does it check live availability from my EHR, or from a copy of my schedule?
• Does the booking land in my EHR during the call, on the correct chart, with the right service code and duration?
• How does it identify a patient who calls from a shared family number?
• Can it reschedule and cancel the existing appointment, or only create new ones?
• Which call types will it refuse, and can I change that list?
• Does it say it is an AI when asked?

How KloudMD handles this

KloudMD Voice AI sits on your existing practice number, answers on the first ring as the practice, and says it is an AI when asked. It checks live availability and books, reschedules, or cancels in the EHR while the caller is on the line, using the same scheduling layer as KloudMD’s online scheduling and the same pinned booking channel, so the phone and the website never compute availability differently. Clinical, urgent, and any call types you mark as escalate are transferred during clinic hours or turned into a message and follow-up task after hours, with 911 and 988 handled per your rules. Every call leaves a transcript and recording on the patient record. It connects to 50+ EHR and practice-management systems; see the AI medical receptionist overview and the Voice AI demo lines to hear a real call.

Get started with KloudMD and hear it answer your own phones in a 15-minute walkthrough.