Home · Guides

AI medical answering service: what it does, what it costs, and when a live service is better

Published September 24, 2026 · Last reviewed October 12, 2026
By the ClinicConvo team

An AI medical answering service is software that answers a medical office's phone calls or messages with a conversational AI agent instead of a live operator. In the published prices below it costs less per call than a live service, and it answers at any hour, but it should only handle routine administrative requests, and anything clinical or urgent must reach a person. Whether AI, a live service, on-call staff or a simple auto-reply fits your practice depends on how many calls and messages you get, how clinical they are and what you can spend.

If you are weighing AI on WhatsApp rather than on the phone, our comparison of WhatsApp chatbots for clinics covers the messaging side.

This guide explains what an AI answering service does, puts published prices next to live services, sets out where AI must stop, and lists the US rules to check before you switch. The second half compares the five realistic ways to cover after-hours patient messages. It is written to help you choose, including when the right answer is not software.

What is an AI medical answering service?

A traditional medical answering service puts a live operator on your phone line when the office is closed or busy. The operator takes a message, follows your script and pages the on-call clinician when the script says so. An AI medical answering service does the same front-line job with a voice agent: it picks up, talks to the caller in natural language and either completes a routine request or records it for your staff. The same idea exists for written channels, where an AI assistant answers patient messages instead of calls.

The jobs an AI agent is suited to are administrative: opening hours, address and parking, which insurance you accept, how to prepare for a visit, and collecting an appointment request with the patient's name and preferred time. The jobs it is not suited to are clinical: deciding whether a symptom is urgent, interpreting results, or advising on medication. The evidence for that limit is in the next sections.

AI vs live medical answering service

Many practices end up with both: AI for the routine calls, and a person (your own staff, an on-call clinician or a live service) for everything the AI hands off. If you are comparing vendors, ask each one to show you the handoff, not only the happy path.

How much does an AI medical answering service cost?

The prices below are what each vendor's own page showed when we read it on October 12, 2026. They are examples, not recommendations, and they change. Two of the three products are sold to businesses in general, not only to medical offices.

• Smith.ai AI Receptionist. A free plan with 25 calls a month, then US$3.00 per call. The Pro plan is listed at US$150 a month, shown at US$2.00 per call at the 75-call level, with extra calls at US$2.50. The pricing page does not mention HIPAA (smith.ai/pricing/ai-receptionist).
• Freed Front Desk. Freed says its service starts at US$149 a month for a block of 250 calls, with no charge for spam or unanswered calls, and describes it as HIPAA compliant and SOC 2 Type II certified (getfreed.ai, article dated July 13, 2026).
• My AI Front Desk. US$99 a month, or US$79 a month billed annually, with 200 voice minutes; extra minutes cost US$0.25 each. The pricing page does not mention HIPAA (myaifrontdesk.com/pricing).

For comparison, the two live services we priced on September 24, 2026 started at US$250 a month for 50 receptionist minutes (Ruby) and US$300 a month for 30 calls (Smith.ai). Using Smith.ai's own figures, that is US$10.00 per call with a live receptionist and US$2.00 per call with its AI plan.

The units differ (calls, minutes, blocks of calls), so headline prices are not directly comparable. Count one month of your calls and their average length, then ask each vendor for a written quote at that volume, including setup, integrations and overage rates.

Where should an AI answering service stop?

At anything clinical. In a 2015 BMJ audit of 23 online symptom checkers, appropriate triage advice was given in 57% of cases. In a 2026 randomized study in Nature Medicine of 1,298 UK participants, people using large language models chose the right course of action in fewer than 44.2% of cases, no better than the control group. An AI agent that answers your phone should take the message, give the emergency number and reach your on-call person. It should not decide whether a caller's symptom can wait.

Trust is the second limit. In a nationally representative survey of 2,039 US adults fielded in June and July 2023 and published in JAMA Network Open in 2025, 65.8% reported low trust in their health care system to use AI responsibly, and 57.7% reported low trust that it would make sure an AI tool would not harm them. Telling callers at the start that they are talking to an AI, and giving them a way to reach a person, addresses that directly.

What rules should a US practice check first?

This is a summary, not legal advice. Check with your lawyer before you switch.

• HIPAA. A vendor that handles protected health information for you generally needs a written business associate agreement. A product built for general business may not offer one, so ask before any patient information goes through it.
• California. Since January 1, 2025, AB 3030 (Health and Safety Code section 1339.75) requires health facilities, clinics and physician practices that use generative AI to communicate patient clinical information to include a disclaimer saying so, with instructions for reaching a human. For audio, the disclaimer must be given verbally at the start and the end of the interaction. Administrative matters such as appointment scheduling and billing are excluded, and so are communications read and reviewed by a licensed provider (summary by the law firm ArentFox Schiff).
• Outbound calls. On February 8, 2024, the FCC ruled that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act (FCC news release). If you plan to use an AI voice to call patients, for example for reminders, check the consent rules that apply to artificial-voice calls first.

How many patient messages arrive after hours?

We could not find a published, clinic-only figure for the share of patient chat messages that arrive after closing time. The fastest way to get your own clinic's figure is to measure it: the Inbox X-Ray reads your exported WhatsApp chats and shows what share of patients wrote outside your opening hours. The numbers we have measured ourselves, with the method, are on our data page.

The best way to size the problem for your clinic is to count it yourself: export one month of messages and mark which arrived after closing, on weekends and on holidays. That single number will tell you which option below makes sense.

How fast do patients expect an answer?

Research on patient expectations mostly covers patient portals and email rather than WhatsApp, so treat it as a guide, not a rule:

• In a 2026 survey at an academic geriatric primary care practice (786 respondents: 683 patients and 103 caregivers), 54.5% of patients and 59.8% of caregivers expected a reply within one business day or faster. Delayed or inadequate replies were among the concerns respondents raised.
• An older 2001 family practice survey asked when patients expected email replies about lab results: 21% said less than 9 hours, 53% said 9 to 24 hours and 26% said more than 24 hours.
• For comparison, a study of more than 5 million secure messages at Mayo Clinic (2010 to 2017) found clinicians replied to 72.2% of patient messages within 1 day and 90.6% within 3 days.

The practical reading: in both surveys, more than half of respondents expected a reply within about a day (54.5% of patients within one business day or faster in 2026; 74% within 24 hours for lab results in 2001). Because expectations vary, the most useful thing any after-hours setup can do is state the reply time so nobody has to guess.

After hours: the five options side by side

The rest of this guide is about patient messages that arrive when the office is closed. Prices in this table are what each vendor's own pricing page showed when we read it on September 24, 2026. They are examples, not recommendations, and they change.

1. Doing nothing

Without any setup, the default is to reply the next morning. It costs nothing up front and it is honest: nobody is pretending to monitor the inbox. The risks are that a patient with something urgent gets no instruction about where to go, and that a patient who wanted to book may book elsewhere. There is also a technical cost on the WhatsApp Business Platform: a free-form reply is only possible within 24 hours of the patient's last message. A message sent at 9 pm on Friday and answered at 9 am on Monday falls outside that window, so you can only reply with a pre-approved template, which Meta charges per message at rates that vary by country. Since October 1, 2026, Meta also charges per message for replies sent inside the 24-hour window (after 1,000 free a month per number), which were free until September 30 (Meta: non-template message pricing).

2. An auto-reply

An automatic "we're closed" message is the minimum every clinic should have. On WhatsApp, the Business app lets you set an away message that only sends outside the business hours in your profile (WhatsApp Help Center). Check whether your email and patient portal can do the same. A good auto-reply fixes the worst risk of doing nothing, because it tells an urgent patient where to go. It does not answer the question or capture a booking. For a clinic with few after-hours messages, this is often enough. Sample texts are further down.

3. On-call staff

A staff member or clinician checks messages at set times or gets alerts. This is the only option in this list where a qualified person can make a clinical judgment at night, which matters for practices whose patients may message with post-procedure symptoms, medication problems or mental health crises. The costs are real: staff time, burnout, and, in the US, possible pay obligations for on-call time under federal Department of Labor rules when the employee must stay so close to work that they cannot use the time effectively for their own purposes. Rules differ by country and contract, so check with your employment adviser.

4. An answering service

A human answering service gives patients a live person at any hour. The two services below publish their prices; others quote on request, so ask for a written quote based on your real message volume.

• Ruby lists "24/7 live answering" and "HIPAA-compliant services" among plan features and says the only difference between its receptionist plans is the number of minutes. Receptionist plans start at US$250/month for 50 minutes, and live chat plans at US$143/month for 10 chats (ruby.com/pricing, read September 24, 2026).
• Smith.ai prices its human-first (live receptionist) plans from US$300/month for 30 calls, with extra calls at US$11.50 and add-ons such as appointment booking at US$1.50 per call; its AI-first service is priced on a separate page (smith.ai/pricing, read September 24, 2026).

Both pricing pages centre on phone calls and website chat. If your patients mostly write on WhatsApp, ask any service whether it can answer on that channel before you sign.

5. An AI assistant

An AI assistant can answer routine questions instantly (hours, address, prices, insurance accepted, how to prepare for a visit) and collect appointment requests for staff to confirm in the morning. Omnichannel inbox tools such as respond.io include AI agents on some plans, and clinic-specific assistants also exist. Whatever you evaluate, ask the vendor to show you exactly what the assistant does with a message about symptoms.

The limit is clinical judgment. In a 2015 BMJ audit of 23 online symptom checkers, appropriate triage advice was given in 57% of cases (80% for emergency cases). In a 2026 randomized study in Nature Medicine of 1,298 UK participants, the models tested (GPT-4o, Llama 3 and Command R+) identified relevant conditions in 94.9% of scenarios on their own, but participants using them identified relevant conditions in fewer than 34.5% of cases and chose the right course of action in fewer than 44.2%, no better than the control group. An AI assistant should never be the part of your process that decides whether something is urgent.

What should happen with urgent or clinical messages after hours?

Professional bodies word it differently, but the common thread is that patients must be told the limits of messaging, including delays in response, and the BMA's guidance for England says explicitly that patients must be told where to go instead when it is urgent or an emergency.

• United States. The AMA Code of Medical Ethics (Opinion 2.3.1) says physicians who communicate electronically should tell patients about its inherent limitations, including possible breaches of privacy and possible delays in response. The American College of Physicians' 2020 guidance says electronic messages should supplement, not replace, in-person care, that expectations and appropriate uses should be discussed with patients, and that messages should be documented in the medical record. The US emergency number is 911, and 988 is the 24/7 suicide and crisis line (call, text or chat).
• United Kingdom. The BMA's guidance on online consultation tools (January 2026) says on-screen messaging should tell patients with an urgent problem to phone the practice or use NHS 111, and to attend A&E or call 999 for a medical emergency. This BMA guidance is written for GP practices in England. 999 is the emergency number across the UK. NHS 111 is for urgent help that is not an emergency in England; Scotland uses NHS 24 on 111 and Wales uses NHS 111 Wales; for Northern Ireland, the NHS website points people to nidirect.
• European Union. 112 is the free emergency number across the EU.
• Everywhere else. Use your country's emergency number and any official non-emergency line, and check your medical council's or college's guidance on electronic communication.

In practice that means three rules, whichever option you pick: every after-hours reply names the emergency number; nothing (no auto-reply, no answering script, no AI) implies that someone is watching the inbox when they are not; and clinical questions are passed to a clinician rather than answered by whoever or whatever is on duty.

What about privacy?

Rules differ by country. This is a summary, not legal advice: check with your regulator or lawyer.

• US (HIPAA). HHS says the Privacy Rule allows providers to communicate electronically with patients if they apply reasonable safeguards. Any outside vendor that handles protected health information for you (an answering service, a messaging platform, an AI vendor) generally needs a written business associate agreement. Ask each vendor whether it signs one.
• EU and UK (GDPR, UK GDPR). Health data is special category data and needs a specific legal condition to process (ICO guidance; GDPR Article 9). Ask vendors where data is stored and for their data processing agreement.
• Other countries have their own health-privacy and data-protection laws. Check with your local regulator.

Sample after-hours auto-replies

A good after-hours reply does four things: says the clinic is closed and the message was received, says when a person will reply, says the channel is not monitored for emergencies, and gives the exact number to call instead. Adapt the examples below to your clinic. Replace every bracket and confirm the emergency numbers for your country.

Which option fits your clinic?

• Few after-hours messages, mostly admin: an auto-reply is usually enough. Software or a paid service may not be worth it.
• Many after-hours messages, mostly routine (booking, prices, hours): an AI assistant or an answering service can answer them instantly. Compare the cost against the bookings you currently lose overnight.
• Patients likely to message with clinical problems (after surgery, during cancer treatment, in mental health care): you need a clinician on call or a nurse triage line. An AI assistant or a non-clinical answering service is not the right tool for those messages, though it can sit in front of them for routine ones.
• Patients who prefer to phone: a telephone answering service, live or AI, fits better than a chat tool.

Checklist

• Count one month of messages: how many arrive after hours, and how many are clinical?
• Set an auto-reply today, whatever else you decide.
• Put your country's emergency number in every after-hours reply.
• Publish a reply time ("by 10 am next working day") and meet it.
• Decide who handles clinical messages at night, and write it down.
• Ask every vendor about data storage, and in the US whether they sign a business associate agreement.
• Test the handoff: send a fake urgent message at night and see what happens.
• Review after 30 days: missed bookings, complaints, and any urgent message that slipped through.

Sources

1. Rydberg MG et al. Patient and caregiver expectations for patient portal communication with providers. J Am Geriatr Soc. 2026. PMID 42544725, doi:10.1111/jgs.70620.
2. Couchman GR et al. E-mail communications in family practice: what do patients expect? J Fam Pract. 2001. PMID 11350705.
3. Huang M et al. Characterizing patient-clinician communication in secure medical messages. J Med Internet Res. 2022. doi:10.2196/17273.
4. Meta. WhatsApp Cloud API: customer service window and WhatsApp Business Platform pricing. Read Sep 24, 2026.
5. WhatsApp Help Center. How to use away messages. Read Sep 24, 2026.
6. US Code of Federal Regulations. 29 CFR 785.17, On-call time.
7. Ruby. Pricing. Read Sep 24, 2026.
8. Smith.ai. Pricing. Read Sep 24, 2026.
9. respond.io. Pricing. Read Sep 24, 2026.
10. Semigran HL et al. Evaluation of symptom checkers for self diagnosis and triage: audit study. BMJ. 2015. doi:10.1136/bmj.h3480.
11. Bean AM et al. Reliability of LLMs as medical assistants for the general public: a randomized preregistered study. Nat Med. 2026. doi:10.1038/s41591-025-04074-y.
12. American Medical Association. Code of Medical Ethics Opinion 2.3.1, Electronic communication with patients.
13. American College of Physicians. Ethical guidance for electronic patient-physician communication, June 29, 2020.
14. British Medical Association. Online consultation tools: managing patient care safely, January 2026.
15. NHS. When to use 111.
16. European Commission. 112, the EU's emergency number.
17. US Department of Health and Human Services. HIPAA FAQ: email with patients and Business associate contracts.
18. Information Commissioner's Office. Special category data; Regulation (EU) 2016/679 (GDPR).
19. 988 Suicide & Crisis Lifeline. 988lifeline.org.
20. Smith.ai. AI Receptionist pricing. Read Oct 12, 2026.
21. Freed. Medical answering service cost, July 13, 2026. Read Oct 12, 2026.
22. My AI Front Desk. Pricing. Read Oct 12, 2026.
23. Nong P, Platt J. Patients' trust in health systems to use artificial intelligence. JAMA Netw Open. 2025;8(2):e2460628. doi:10.1001/jamanetworkopen.2024.60628.
24. ArentFox Schiff. California requires disclaimers for health care providers' AI-generated patient communications (AB 3030, Health and Safety Code section 1339.75). Read Oct 12, 2026.
25. Federal Communications Commission. FCC makes AI-generated voices in robocalls illegal, news release, February 8, 2024.

Keep reading

• AI Front Desk for Clinics: What It Handles, What Staff Do
• AI Receptionist for Medical Offices: 2026 Buyer's Guide
• AI Scheduling Assistant for Clinics: Can AI Book?
• AiSensy Alternatives for Clinics (2026)
• Appointment Reminder Texts: What Cuts No-Shows
• DenGro Alternatives for UK Practices (2026)
• HIPAA Compliant Chatbot: What It Takes and How to Check
• Medical Chatbot for Clinics: Rule-Based vs AI
• Pabau Engage Alternatives for UK Clinics (2026)
• Respond.io Alternatives for Clinics (2026)
• Wati Alternatives for Clinics (2026)
• WhatsApp Business API Pricing and HIPAA for Clinics
• WhatsApp Chatbots for Clinics: 9 Tools Compared (2026)
• Our data: when patients message clinics
• 1 in 4 bad Google reviews of clinics are about not being able to reach them

Prebuilt WhatsApp agents by specialty · All guides