Start with the plain definition. An AI receptionist is software that handles the first contact with your business, works out what the person needs, answers from information you approved, and either finishes the job (books the slot, takes the details) or hands the conversation to a person. Some answer the phone with a synthetic voice. Others answer by text. The channel is the main difference between them, and it shapes what the product can do for you.
This guide explains how AI receptionists work step by step, where the voice and text versions part ways, what they cannot do, and what to set up before one talks to a customer. FromKai builds a texting agent, so the examples lean that way, and the voice differences are called out plainly. Product guidance, not legal advice.
What is sold as an "AI receptionist"
Three things carry the label. A phone tree with better recordings ("press 1 for sales") understands nothing and books nothing. An auto-reply, a fixed text or recording that promises a callback, responds in seconds and resolves nothing. A conversational agent reads what the person said in their own words, answers it, asks what it still needs to know, and takes an action: books, routes, captures, or escalates. Only the third one is a receptionist, and everything below is about it.
How AI receptionists work, step by step
Use these six steps to evaluate any AI receptionist, voice or text. Vendors build each step differently, and a product that skips one of them is not a receptionist.
1. A trigger starts the conversation
Something tells the software a person is waiting: an inbound call, an inbound text, a missed call, a web form. For a voice receptionist the trigger is the call itself, forwarded by your phone system. For a texting receptionist it is the text that arrives, or a missed-call event your phone system passes along so the agent can text the caller back. The missed call text back guide explains why that second trigger matters to a service business.
2. The agent reads what the person said
A voice receptionist first transcribes the audio with speech recognition, then works from the transcript. A texting receptionist skips that step; the message is already text. From there both do the same thing: a language model reads the message, in context with the rest of the thread, and works out the intent. "I'm interested but at work, can someone call me around 4?" contains interest, unavailability, a channel preference and a time. A keyword matcher sees "interested." A model reads all four.
3. It answers from approved information
The agent does not answer from the open internet. It answers from a knowledge base the business set up: services, service area, hours, the prices you are willing to state, scheduling rules, tone, and the topics it must not touch. Ask any vendor what happens when the question is not in that knowledge base. The right answer is "it says it will check and hands off," not "it guesses."
4. It asks the qualifying questions you set
A good front-desk person does not just answer; they ask. What is the address? Repair or replacement? The agent works from a list the business defined and asks only what is still unanswered, so a lead who already said "120 feet of wood fence in Dumont" is not asked the length again.
5. It takes an action
This is the step that separates a receptionist from a chat window. The agent:
- -Books. Checks real calendar availability and offers open times. The AI appointment setter guide covers what a real booking motion needs.
- -Captures. Writes the lead's details and the thread into the CRM.
- -Routes. For voice, transfers the call to a person with a summary. For text, hands the thread to a person with a summary.
- -Schedules a follow-up. "Not until next month" becomes a message next month, not day three of a drip. See AI lead follow up.
6. It hands off, stops when told to, and logs everything
Every serious AI receptionist has a rule for when a person takes over: the customer asks for one, the topic is out of scope, the agent is unsure, or the conversation turns sensitive. The handoff should carry a summary; a raw transcript to scroll while the customer waits is not a handoff. When someone asks to stop hearing from you, the agent has to stop; for texts covered by the TCPA that is a legal requirement, not a feature. And every conversation gets a transcript, an outcome and an opt-out status, because what you cannot read back, you cannot fix.
Voice AI receptionists vs text AI receptionists
Judge both kinds on the six steps above. Where they differ is the channel, and the channel changes the product.
Be honest about which one your customers want. If your phone rings all day and callers expect a live answer, a voice receptionist is the right category. If your leads text in, fill out forms, call after hours or screen unknown numbers, text is where the conversation already is. A Pew Research Center survey of U.S. adults conducted in July 2020 and published that December found that eight in ten do not generally answer their cellphone when an unknown number calls, which is the problem any voice product calling a lead back has to solve.
Plenty of businesses run both: a phone system or voice product on the line, and a texting agent on everything the phone drops. The two do not compete for the same conversation.
What an AI receptionist cannot do
Set expectations before the first customer does.
- -Answer what you did not teach it. The knowledge base is the ceiling. An hour of setup here is the difference between an agent and a parrot.
- -Exercise judgment. An upset customer, a pricing exception, a request that fits no script: those go to a person, and the agent's job is to recognize the moment.
- -Make an unreachable lead reachable. A voice agent calling a number that screens calls reaches voicemail. A texting agent sending from an unregistered business number can be filtered or blocked by the carriers. And a lead who never consented to texts is not someone it should text for marketing at all, whether or not the message would deliver; consent is your obligation and your record, not something the network checks for you.
- -Make you compliant. It can honor opt-outs, keep records and run pre-send checks. How you obtained the number, what consent you can show and what your messages say are still yours.
AI receptionists for a small business: what to set up first
If you run a plumbing company, a clinic, a law office or a two-location gym, the receptionist you need is the one that covers the hours nobody is at the desk. Set these up, in this order:
- 01Decide the trigger. Missed calls, inbound texts, web forms, or all three.
- 02Write the knowledge base. Services, area, hours, what you will say about price, the questions that qualify a lead, and the topics that are off limits.
- 03Connect the calendar and CRM. A booking that only exists in a text thread is not booked.
- 04Set the handoff rules. Who takes over, when, and what they receive.
- 05Measure one number. Not texts sent. How many first contacts ended in a booked appointment, including the ones at 9 pm.
The speed to lead guide collects the research on why that last number swings so hard on the first few minutes, with its sources.
The rules that apply to a texting receptionist
A texting AI receptionist sends business texts from a business number. That puts it under the same rules as any other business texting, and no vendor can describe it as free or built in without a catch.
- -Registration. Business texts from a standard local 10-digit number in the United States go through A2P 10DLC, the carriers' registration system run through The Campaign Registry. You register a brand and a campaign; traffic from an unregistered number can be filtered or blocked, depending on the carrier and the route it takes. What is 10DLC walks through it.
- -Consent. When the customer texts first, a prompt single reply that contains only relevant, responsive information is likely conversational under CTIA's Messaging Principles and Best Practices; anything promotional, or a recurring sequence, needs its own opt-in. For marketing, CTIA's guidelines call for express written consent, and automated marketing texts can fall under the TCPA, which requires prior express written consent for many of them; ask counsel how that applies to how you send. Keep records; the SMS consent records guide covers what to capture.
- -Opt-out. Two layers apply. The carriers' side: CTIA's guidelines, the wireless industry's best practices, call for honoring STOP and the other standard keywords as well as plain-language requests like "please stop texting me." The federal side, for texts covered by the TCPA: the FCC's February 2024 order currently requires that a revocation be accepted in any reasonable manner and honored within a reasonable time, no more than 10 business days. From November 25, 2026, a later order (FCC 26-67) lets a sender designate and clearly disclose exclusive ways to revoke, such as replying STOP or using a stated website or phone number; a sender that does not designate one must still accept any reasonable means. Check the current rule with counsel.
- -Cost. A texting receptionist costs the platform fee plus per-message carrier and platform charges plus registration fees. Price the whole loop, not the headline number.
Talk to counsel about your own program. A voice receptionist has its own rules around call recording and consent; those are outside this guide.
Where Kai fits
Kai is FromKai's AI texting agent. It is a text receptionist, not a voice one: FromKai does not answer phone calls, and your existing phone system stays where it is. What Kai is built to do, once you set it up: when a call is missed, your phone system or CRM tells FromKai through a webhook, FromKai runs its screening checks (do-not-call, litigator, number validation and suppression lists), and Kai picks the caller up by text. When a text comes in, Kai reads the thread, answers from information your business approved, asks the qualifying questions you set, offers times from the calendar you connect (booking is on Kai Pro and Kai Max), schedules a follow-up for the date the lead named, and hands the thread to a person with a summary when a person should take over. If someone on your team starts typing in a thread, Kai pauses. Kai works over SMS, web chat and missed-call text back.
Setup includes 10DLC brand and campaign registration; SMS credits are purchased separately. The missed call text back software page shows the full workflow.
Want to test the loop yourself? Text Kai's demo line at (878) 7TXT-KAI, (878) 789-8524, ask something a customer would ask, then ask something that is not on the website and watch what it does.
Frequently asked questions
How do AI receptionists work?
A capable AI receptionist works through six steps: a trigger (a call, a text, a missed call or a form) starts the conversation, a language model reads what the person said in context, the agent answers from a knowledge base the business approved, asks the qualifying questions the business set, then books an appointment, captures the lead or hands off to a person with a summary. Voice versions add speech recognition on the way in and a synthetic voice on the way out; text versions work on the message as written.
What is the difference between a voice AI receptionist and a text AI receptionist?
A voice AI receptionist answers inbound phone calls with a synthetic voice. A text AI receptionist answers inbound texts and texts back missed calls. Both read the message with a language model, answer from approved information and book or hand off. Voice fits customers who expect a live phone answer; text fits leads who text, screen calls or reach out after hours, and leaves them a thread to scroll back through.
Can an AI receptionist book appointments?
Yes, if it is connected to a real calendar. A good one checks live availability, offers open times, books the slot inside the conversation and writes the booking to your calendar and CRM. One that only sends a scheduling link is passing the work back to the customer.
What can an AI receptionist not do?
It cannot answer questions outside the knowledge base you gave it, exercise judgment on an upset customer or a pricing exception, reach a lead who screens calls, text a lead who never consented, or make your messaging program compliant on its own. It should recognize those moments and hand the conversation to a person.
Does a text AI receptionist need 10DLC registration?
If it texts US numbers from a standard local 10-digit business number, yes. Carriers require application-to-person traffic to come from a registered brand and campaign through The Campaign Registry, and sends from an unregistered number can be filtered or blocked, depending on the carrier and the route. Toll-free numbers go through their own verification instead.
Does FromKai answer phone calls?
No. FromKai is a texting platform. Your phone system or CRM detects a missed call and triggers FromKai through a webhook, and Kai takes over by text from there. You keep the phone system you already have.
Sources: Pew Research Center, "Most Americans don't answer cellphone calls from unknown numbers," December 14, 2020 (web survey of U.S. adults, July 13-19, 2020). CTIA, Messaging Principles and Best Practices, May 2023. Federal Communications Commission, TCPA consent revocation Report and Order (FCC 24-24), February 2024. Federal Communications Commission, FCC 26-67, 91 FR 67214 (October 26, 2026), effective November 25, 2026 (exclusive revocation methods). Telephone Consumer Protection Act, 47 U.S.C. 227. The Campaign Registry, A2P 10DLC brand and campaign registration documentation. Research on response timing and follow-up cadence is cited, with sources, in the linked FromKai guides. Example details are fictional.