What an AI receptionist does, what it costs and how to set one up

What is an AI receptionist?

An AI receptionist is a phone assistant run by artificial intelligence (AI). It picks up when someone calls your business. It listens, works out what the caller wants and replies in a natural voice.

It is different from an old phone menu. A phone menu asks callers to press 1 or 2. An AI receptionist lets them speak in their own words, such as "Can someone fix my water heater tomorrow?"

You can usually keep your number. You forward calls to the AI, or the vendor gives you a new number. Goodcall, for example, includes a number on every plan and also lets you forward your existing one.

What does an AI receptionist do on a real call?

Picture a call to a small plumbing company at 7 p.m. A good AI receptionist handles it like this:

• Greets the caller: It answers with the business name and says it is an AI assistant. This sets honest expectations from the first second.
• Finds out what they need: It asks short questions. What is the problem? What is the address? How soon do you need help?
• Answers common questions: It gives your hours, service area and any prices you have told it to share.
• Books the visit: If it connects to your calendar, it offers open times and books one. Then it can text the caller a confirmation.
• Passes urgent calls to a person: A burst pipe goes to the plumber on call right away.
• Takes a message: If nobody is free, it records the name, number, address and reason for the call.
• Sends you a summary: After the call, you get a short written summary, a transcript and often a recording.

Rosie, for example, lists call summaries, transcripts, recordings, and email and text alerts on all of its plans. Dialzara lists appointment booking, lead qualification and call routing from its lowest plan.

What can an AI receptionist not do well?

AI handles routine calls well. It still has limits you should plan for:

• Judgement calls: Complaints, refunds and prices for complex jobs still need a person.
• Emotional calls: A caller who is upset or scared usually wants a human voice.
• Facts you never gave it: It only knows what you load into it. If your website is out of date, its answers will be too.
• Wrong answers said with confidence: AI models can sometimes state a wrong fact as if it were true. Limit it to answers you have written and checked.
• Bad lines and noise: Noisy calls can lead to mistakes in names and numbers. A good setup reads key details back to the caller.

How do you judge whether an AI receptionist is actually good?

Every vendor's demo call sounds great, because demo calls are written for the demo. The real test is a question specific to your own business, asked the way a real caller would ask it, not a tidy script. Before you pick a plan, run the free trial through a short checklist of your own.

• Ask it something messy. Not "What are your hours?" Ask the confusing question your customers actually ask, half-finished and out of order, the way people really talk on the phone.
• Listen for what it does when it does not know. A good AI receptionist says so and offers to find out or pass the call on. A bad one guesses, confidently, and a confident wrong answer is worse than no answer.
• Interrupt it. Talk over it mid-sentence, the way an impatient caller would. See whether it recovers cleanly or loses the thread.
• Check how easy it is to change what it says. Prices change, hours change, a new technician joins the team. Find out whether you can update that yourself in a few minutes, or whether every change is a support ticket.
• Ask to hear a real call, not a demo call. A vendor confident in their product will let you hear an anonymized real call from a business like yours, not only the one built for the sales page.

If a vendor cannot answer these cleanly before you have paid them anything, that tells you something.

How much does an AI receptionist cost?

Vendors charge in different ways. Some charge by the minute of talk time. Some charge for each real call. Goodcall charges by the number of different people who call each month, and it does not meter minutes.

These prices come from each vendor's own pricing page as of October 2026.

Product	Lowest paid plan	What that plan includes	Other plans
Dialzara	$29 a month (Business Lite)	60 minutes, then $0.48 a minute	$99, $199 and $349 a month
Rosie	$49 a month (Professional)	250 minutes	$149 for 1,000 minutes, $299 for 2,000 minutes
Goodcall	$79 a month (Starter)	100 unique customers, about 150 calls	$129 and $299 a month
Smith.ai AI Receptionist	$150 a month (Pro)	75 real calls, then $2.50 a call	A free plan with 25 calls, and Enterprise from $500 a month

The plan you need depends on your call volume. Check your phone bill or call log for last month. Count the calls and how long they lasted. Then pick a plan that covers that number with some room to spare.

Higher plans often add the features that matter most. Rosie adds calendar booking and live call transfers on its $149 Scale plan. Dialzara adds warm transfers and API access on its $349 Business Elite plan. Smith.ai keeps custom prompting and custom integrations for its Enterprise plan.

How does an AI receptionist compare with a human receptionist?

A person brings judgement and warmth, and costs much more. The US Bureau of Labor Statistics (BLS) reports that the median receptionist earned $38,010 a year, or $18.27 an hour, in May 2025.

Wages are only part of the cost. In June 2026, benefits made up 30% of total pay costs for private industry workers, the BLS says. If benefits add a similar share for a receptionist, the full cost comes to about $54,000 a year. That buys about 40 hours of cover in a week that has 168 hours.

Human answering services sit in the middle. Ruby, a US virtual receptionist service, lists $250 a month for 50 minutes and $720 a month for 200 minutes as of October 2026. My guide to virtual receptionists vs AI receptionists compares all three options in detail.

What US rules apply to an AI receptionist?

This section is general information. Ask your own lawyer about your setup.

Call recording consent

Most AI receptionists record and transcribe calls. Federal law allows a recording when one person on the call agrees. That rule sits in section 2511 of Title 18 of the US Code. Some states go further and require consent from everyone on the call. Justia's 50-state survey lists California, Florida, Maryland, Massachusetts, New Hampshire, Pennsylvania and Washington as all-party consent states. Nevada needs consent from everyone on a phone call too. Illinois, Montana and a few other states have their own versions of the rule.

You cannot always tell where a caller is. So the safe habit is a short notice at the start of every call. For example: "Hi, this is the AI assistant for Lake Plumbing. This call is recorded."

Callbacks, reminders and marketing calls

On 8 February 2024, the Federal Communications Commission (FCC) ruled that AI-generated voices count as "artificial" voices under the Telephone Consumer Protection Act (TCPA). The ruling took effect right away. It means the consent rules for robocalls also cover AI voice calls. For telemarketing robocalls, the TCPA already required prior express written consent from the person you call.

This matters when your AI starts making calls. Callbacks, reminders and sales calls all count. Keep a record of each person's consent before your AI calls or texts them. My guide to AI voice agents covers outbound calls in more depth.

Medical, dental and law offices

Medical and dental offices must follow HIPAA, the federal health privacy law. HHS says that if a vendor handles patient health information for you, you need a business associate agreement (BAA) with that vendor first. A BAA is a contract that says how the vendor will protect that data. My guide to a medical answering service covers this in more detail.

Law firms owe clients confidentiality. In July 2024, the American Bar Association released Formal Opinion 512 on generative AI. It says lawyers must protect client information under Model Rule 1.6 and understand the risks of the tools they use. Ask any vendor where it stores recordings and who can hear them. See my guide to a legal answering service.

How do you set up an AI receptionist well?

Write down your most common questions

Look at your last 50 calls, emails and texts. List the questions people ask most often. Write a short, correct answer to each one. This becomes the core of what the AI says.

Set clear rules for what it can say

Tell it what it may share, such as hours, service area and booking steps. Tell it what it must never do. Examples include quoting a price for a job it cannot see, promising a time you cannot keep, or giving medical or legal advice.

Plan the handoff to a human

This is the most important part of the setup. Decide which calls go to a person, and which person. Common reasons to transfer are an emergency, an angry caller, a caller who asks for a human and a large sales lead.

There are two main ways to pass a call:

• Warm transfer: The AI tells your staff member who is calling and why, then connects them. The caller does not have to repeat their story.
• Blind transfer: The AI sends the call straight through with no introduction. It is faster, but your staff member starts with no context.

Also plan for when nobody picks up. The AI should take a full message and tell the caller when to expect a call back. It should alert you by text right away. Then you must call back quickly, or the setup fails the caller.

Connect it to your tools

An AI receptionist is much more useful when it can see your calendar and save each call into your CRM (your customer list). Check which tools each plan connects to. Goodcall allows 1 external tool on Starter, 5 on Growth and 15 on Scale.

Test before you go live

Call it yourself many times. Ask hard questions. Pretend to be angry. Ask for a person. Smith.ai's free plan includes 50 simulated test calls a month. Rosie and Dialzara each offer a 7-day free trial.

Review calls every week

Read the call summaries every week for the first month. Fix any wrong answers. Add the new questions callers ask. Expect to make many small changes early on.

What goes wrong when businesses set these up badly?

Most of the problems I see with AI receptionists are not the AI's fault. They are the setup. A few mistakes show up again and again.

• Letting it promise what it cannot deliver. An AI that quotes a price or a same-day slot it is not actually authorized to guarantee puts the business in the position of either honoring a promise it never agreed to or disappointing a customer who was told something different. Write down exactly what it is allowed to commit to, and nothing else.
• No real way out for an upset caller. If every path loops back to the AI and a frustrated caller can never reach a person, you have built a wall, not a receptionist. There should always be a fast, obvious way to a human.
• Setting it up once and forgetting it. Hours change, prices change, a service gets discontinued. An AI receptionist only knows what you tell it, and a business that updates its website but not its AI ends up giving two different answers to the same question.
• Treating disclosure as optional. A caller who later realizes they were talking to an AI the whole time, and were not told, tends to feel misled, whatever the law in your state actually requires. It costs nothing to say so upfront.
• Testing it once, in ideal conditions, and calling it done. A quiet room and a clear voice is not a real test. A bad phone line, a strong accent, a caller talking over it, these are what will actually happen, and they should be what you test for before a real customer finds the gap for you.

When should you pay for a custom AI receptionist?

A software subscription is the right choice for most small businesses. Buy one if your calls are simple and your tools are common.

A custom AI receptionist makes sense when one of these fits you.

• Your job software has no connection: The AI needs to check stock, job status or account details in a system the products do not support.
• Your calls follow unusual steps: For example, a detailed intake for insurance claims that a standard script cannot follow.
• You run many locations or brands: Each one needs its own rules, and the monthly fees add up.
• You must control the data: You want recordings and transcripts in accounts your business owns.
• Your call volume is high: At thousands of calls a month, per-call fees can cost more than running your own voice agent on a platform that charges by the minute.

How does Blur Studio approach an AI receptionist?

When a business asks me for an AI receptionist, I first check whether a product it can buy already does the job. Often one does, and I say so. When none fits, I build an AI receptionist that connects to the business's own calendar, CRM or job software, so it can actually see your schedule and your customer records instead of guessing. You can read more about that work in my guides to AI agent development and AI integration services.

I write the call rules with the owner, set up the handoff to staff and test it on real calls before launch, the way I described above, not just once. The code and the accounts stay in the business's name, and I keep an eye on the system after launch so the answers stay correct as your business changes.