AI Receptionist Guide: What It Can and Can't Do in 2026

An AI receptionist is a voice agent that answers a business's phone calls. It can greet callers, answer common questions, book appointments, take messages and transfer calls to a person. Vendors price the service per call, per minute or per unique customer, so costs are hard to compare. In Sierra's tau-Voice customer-service benchmark, the best voice agent reached 67 percent of tasks by April 2026, while a text agent in an earlier study reached 85 percent.

What an AI receptionist does today

Vendors describe these features on their own pages. Treat each one as a vendor claim.

Answering and sorting. Goodcall says its service answers calls 24/7, sorts callers into new, existing, urgent and spam, and asks qualifying questions.
Booking. Goodcall checks the calendar and books appointments, and Rosie advertises direct calendar booking.
Messages and leads. Goodcall sends leads by text and email, and Rosie provides call summaries, transcripts and recordings.
Transfers. Goodcall transfers calls to your line or a specific person when it cannot resolve a request. Rosie offers warm and live transfers, and Smith.ai says its warm transfers carry full context.
Human backup. Smith.ai says AI escalation to a human has no fee, and live-agent involvement is an add-on.

These features work best for narrow, predictable calls: a booking, a message, a question about hours. Calls that change records or depend on policy are harder, as the next section shows.

What it still gets wrong

Customer-service benchmarks are a demanding test. For the underlying technology, see what an AI voice agent is, and for the speed side of a live call, see the latency guide. The tau-Voice benchmark from Sierra uses 278 tasks in which the agent changes records through tools under a set of policies. The arXiv paper, submitted March 14, 2026, reports that voice agents completed 31 to 51 percent of those tasks under clean conditions and 26 to 38 percent with noise and diverse accents. The reasoning text agent completed 85 percent. The authors found that 79 to 90 percent of failures came from agent behavior, not from audio understanding.

Sierra's May 1, 2026 post reports a later snapshot. Voice task completion rose from 30 percent for OpenAI's gpt-realtime-1.0 in August 2025 to 67 percent for xAI's grok-voice-think-fast-1.0 in April 2026, and voice reached about 79 percent of text capability. Sierra's accents are generated with ElevenLabs personas.

xAI's post on Grok Voice 3.0, dated October 14, 2026, claims 71.6 percent on tau-Voice and says its Starlink call centers handle about 20,000 calls a day, with roughly 70 percent resolved without a human. Those are vendor claims. Most receptionist work is simpler than tau-Voice, but a caller with an unusual request still needs a clear handoff to a person.

AI receptionist pricing compared

The table compares entry plans as published in October 2026. The billing units differ, so the columns do not line up neatly.

| Vendor | Type | Billing unit | Entry plan | Included | Overage or notes | Source | |---|---|---|---|---|---|---| | Goodcall | AI | Unique customers | $79 a month | 100 customers | $0.79 per extra customer; no per-minute charges | goodcall.com/pricing | | Rosie | AI | Minutes | $49 a month | 250 minutes | Overage not shown on the pricing page | heyrosie.com/pricing | | ElevenLabs Reception AI | AI | Credits, one per inbound minute | $29 a month | 75 credits | $0.45 per credit | ElevenLabs billing | | Smith.ai, AI Receptionist page | AI with human option | Calls | Free; Pro $150 a month | 25 calls on Free | $3.00 per call on Free, $2.00 on Pro, $1.67 on Enterprise | smith.ai/ai-receptionist | | Smith.ai, pricing page | AI-first or human-first | Calls | $300 a month | 30 calls | $11.50 per call over 30 | smith.ai/pricing | | Ruby | Human, AI optional | Minutes | $250 a month | 50 minutes | Overage not stated | ruby.com/pricing |

Smith.ai's two pages show different price structures, so confirm the current plan before you budget. Goodcall's pricing page is the more detailed of the AI options: it defines a unique customer as one phone number per calendar month, so repeat callers do not add to the count, and spam and wrong numbers do not count at all.

AI or human receptionist

A human receptionist and a per-call AI service are different cost structures. The US Bureau of Labor Statistics reports a median pay of $38,010 a year, or $18.27 an hour, for receptionists in May 2025, across 947,500 jobs. The BLS outlook projects a 2 percent decline in employment from 2025 to 2035. That does not translate into a percentage saving, because a salaried receptionist covers hours and tasks that a per-call service may not.

Ruby sells live human receptionists and says its optional AI enhancements are available on all plans at no extra cost. Many buyers end up choosing a mix: AI for overflow and after-hours calls, and a person for complex or sensitive calls.

Rules to know: disclosure and consent

This is general information, not legal advice. Check the rules that apply to your calls with a lawyer.

United States, robocalls. The FCC ruled on February 8, 2024 that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act. The prior express consent standard applies. The ruling concerns robocalls, which are outbound, so do not assume it governs inbound answering without checking.
United States, state law. Utah's 2024 law, as enacted, requires a business using generative AI to disclose it when asked. Licensed occupations must disclose proactively at the start of spoken exchanges. The law has been amended since, so check the current text. Davis Wright Tremaine's summary covers the original version.
European Union. Article 50(1) of the AI Act requires disclosure that people are dealing with AI, unless that is obvious to a reasonably well-informed person. Goodwin reports that the transparency duties are now in force, with fines of up to EUR 15 million or 3 percent of worldwide annual turnover, under Article 99(4). Its analysis also notes that the AI Omnibus delayed high-risk rules, but not Article 50(1).

None of the platform pages checked for this article promise legal compliance. Microsoft's Voice Live features cover noise suppression, echo cancellation and end-of-turn detection, not disclosure. Check any receptionist product's compliance claims in writing.

How to set one up

Define the scope. List the calls the agent may handle, such as bookings, messages and hours, and the calls it must hand to a person.
Write the call flows. For each flow, decide what the agent asks, what it can change and when it transfers.
Record the fixed prompts. Greetings, disclosure lines and hold messages do not change, so produce them as audio files and review them once.
Test with real calls. Run test calls from different phones and in noisy places, since benchmark conditions are cleaner than most lines.
Set handoff rules. Decide which requests go to a person, how the transfer happens and what context the caller hears or the staff member sees.
Review transcripts. Check what the agent said and where it failed, then adjust the flows.

Where recorded voices fit

OndaLabs does not offer an AI receptionist or any call-answering product. It is a studio for recorded voices, and it produces the fixed audio around a receptionist line.

Phone-line prompts. Greetings, after-hours messages, voicemail greetings, menu prompts and a disclosure line such as "This call is answered by an automated assistant." Produce them in the text to speech studio. Commercial use is included from the Starter plan at $9 a month.
A branded voice. A consistent voice can be designed from a text description or cloned from a recording you have permission to use, at 1,500 credits each, through custom voices. The Creator plan includes 3 custom voices, Pro includes 10 and Business includes 30.
Bilingual greetings. The 30 prebuilt voices work in every supported language, and there are Spanish text to speech pages among 37 language pages.
Rehearsing call flows. The dialogue tool can script a sample caller and receptionist exchange to rehearse a call flow before you test it on a phone line.

The cost is simple arithmetic. A 200-character greeting costs 200 credits on Onda Lite at 1 credit per character, or 300 credits on Onda Flash at 1.5 credits per character. The Free plan's 2,000 credits cover several such greetings a month.

Market size estimates

One analyst firm, market.us, estimates the global AI receptionist market at USD 2.7 billion in 2025, rising to about USD 31.3 billion by 2035, a 27.8 percent compound annual growth rate. Its August 2026 report is one estimate among many, and other publishers give very different figures. Do not treat any single number as consensus.

Frequently asked questions

What is an AI receptionist?

Do I have to tell callers they are talking to an AI?

In the EU, Article 50(1) of the AI Act requires disclosure unless it is obvious to a reasonably well-informed person, and it applies from August 2, 2026. In the United States, the FCC's 2024 ruling covers AI-generated voices in robocalls, and Utah's law requires disclosure when asked. A clear disclosure line is the safe default, and this is not legal advice.

How much does an AI receptionist cost?

Entry plans range from free, with 25 calls a month on Smith.ai's Free tier, to $150 a month on Smith.ai's Pro tier. Goodcall starts at $79 a month, Rosie at $49 and ElevenLabs Reception AI at $29, while Ruby's human-backed plans start at $250. Billing units differ: some charge per call, some per minute and some per unique customer. Compare the total cost for your typical call volume rather than the entry price alone.

Can an AI receptionist transfer calls to a person?

Most products offer it. Goodcall transfers to your line or a specific person, Rosie offers warm and live transfers, and Smith.ai says its warm transfers include full context. OpenAI's Realtime SIP interface also exposes a refer endpoint for transfers. Test the handoff with real calls before you rely on it.

Is an AI receptionist better than a human receptionist?

It depends on the calls. An AI receptionist can cover after-hours and overflow calls on a per-call or per-minute plan, but benchmark results show agents still fail on complex, policy-bound tasks. A human receptionist handles nuance better in many cases. Many businesses use both, with a clear rule for when a call moves to a person.