AI receptionists and voice agents in 2026: what they cost, what they handle, and the rules

What is an AI receptionist?

An AI receptionist is a voice agent that answers your business phone line. It listens to the caller, works out what they want, answers from your own FAQs and documents, and takes a message or transfers the call to a person when it can’t help. Under the hood it is speech recognition, a language model and a synthetic voice, connected to a phone number.

It is different from the old phone menu (“press 1 for sales”), which only routes, and from a virtual receptionist service, where a human answers on your behalf. AI voice agent is the broader term: the same technology answering a support line, an after-hours line or a booking line.

What does an AI receptionist handle well, and badly?

It is good at the calls that repeat:

• Hours, prices, locations and “are you open on Sunday?”
• Availability and simple booking questions
• How-to and policy questions answered in your help articles
• Taking a clear message, and routing the caller to the right person
• Every call after hours, or when everyone is already on the phone

It struggles with exceptions, noisy lines, strong emotion and anything your documents don’t cover. Self-service in general resolves less than vendors suggest: in a Gartner survey of 5,728 customers, only 14% of service issues were fully resolved without a person. Speed is the other limit. People reply to each other within a few hundred milliseconds, while Twilio’s launch targets for its voice agents are a median of about 1.1 seconds from the caller finishing to hearing a reply. A good agent feels slightly slow; a bad one talks over people.

Callers want a way out. In a 2026 Gartner survey, 87% of customers said companies using generative AI must give them access to a human agent. In a May 2026 survey by AnswerConnect, which sells human answering services, 31% said they would hang up if connected to AI.

Voice AI has had public failures. McDonald’s ended its AI drive-thru test with IBM in 2024, and Taco Bell rethought its AI drive-thru in 2025 after glitches and pranks such as an order for 18,000 waters. Drive-thrus are noisier than phone lines, but the lesson carries over: test with real callers in real conditions before you trust it.

What does an AI receptionist cost?

Prices from each provider’s US pricing page on October 5, 2026. Services price per call, per minute or per customer; platforms you build on price per minute.

Here is an illustration, not a quote: 1,000 inbound calls a month, averaging three minutes (3,000 minutes), every call billable.

The per-minute platforms are cheapest on usage, but that leaves out the build, the testing and keeping it running. Packaged services include all of that, and their per-call prices assume their own idea of a typical call. Run your own numbers at your real volume and average call length; long calls favor per-call pricing, short calls favor per-minute.

Should you buy a service or build your own?

Buy a receptionist service if

• Your calls are standard: hours, prices, messages and simple bookings.
• Your volume is modest, and their booking integrations fit your calendar.
• You want it answering this week without any engineering.

Build your own voice agent if

• It must answer in depth from your own documents, policies and help articles.
• It needs your own systems: order status, account lookups, your CRM.
• You need custom transfer rules, languages or a voice the services don’t offer.
• Your volume is high enough that per-call fees outrun usage costs.
• You want to own the call logs, the data and the logic.

We build the second kind: an agent that answers from your FAQs and articles goes live in three days, and one that books or looks things up takes about a week. See AI voice agents and receptionists. If a packaged service fits your calls, we’ll tell you.

Are AI phone calls legal?

Answering calls with AI and making calls with AI are treated differently in the US. This is a summary, not legal advice.

• Outbound calls. In February 2024 the FCC ruled that AI-generated voices are “artificial” under the Telephone Consumer Protection Act, so calling someone with one needs their prior express consent. Telemarketing calls to mobile phones need prior express written consent, and the message must identify your business at the start and give a callback number.
• Inbound calls. In its follow-up proposal the FCC said plainly: “The TCPA’s requirements do not extend to technologies used to answer inbound calls.” An AI receptionist answering your own line is not a robocall. That proposal, which would also require AI disclosure at the start of outbound calls, was not final as of October 2026. Courts are no longer bound by the FCC’s reading of the TCPA since McLaughlin Chiropractic v. McKesson (2025), so get consent for outbound calls regardless.
• Recording. Many states, including California, Florida, Illinois, Maryland, Massachusetts, Pennsylvania and Washington, require every party’s consent to record a call. If your agent records or transcribes calls, say so in the greeting.
• Your vendor listening. In Ambriz v. Google (2025), a federal court let a California wiretapping claim proceed over Google’s Contact Center AI, because the vendor had the capability to use call data for its own purposes. Check what your voice vendor may do with recordings, and say who processes calls in your privacy notice.
• Large businesses in California. From January 1, 2027, AB 1609 bars businesses with more than $500 million in revenue from presenting a customer service bot as human, including on the phone, and requires a good-faith effort to connect callers to a person within 15 minutes or book a time within one business day.
• Europe. Since August 2, 2026, the EU AI Act requires telling people they are interacting with AI at the latest at the first interaction, unless it’s obvious.

The safe default everywhere: say it’s an AI assistant in the first sentence, mention recording if you record, and let callers reach a person by asking.

How do you launch one safely?

1. Count your calls. List the ten reasons people called last month, from call logs or voicemails. Those are the agent’s job.
2. Write the answers down. The agent answers from your documents, so the FAQ comes first.
3. Decide the transfers. Who takes which calls, during which hours, and what the agent says after hours.
4. Write the greeting. Name the business, say it’s an AI assistant, mention recording if you record, and tell callers they can ask for a person.
5. Test by phone. Call it with 50 real questions, in different accents, on a bad line and from a car.
6. Start after hours. Put it on overflow and after-hours calls first, then the main line once the logs look right.
7. Cap and log. Set a monthly cost ceiling, keep a log of every call, and review a sample every week.

What next?

Pull last month’s call log and count the calls you missed and the ones that asked the same question. If either number is large, an AI receptionist will pay for itself quickly. When you’re ready, send us your requirements or see how a three-day voice agent build works.