What an AI receptionist costs in 2026

"AI receptionist" is sold on four different units, so two quotes can differ by a factor of ten and still describe the same amount of work. One is a monthly subscription, one is a rate per minute, one is a rate per call, one is a rate per member of staff. Until you know which, the headline numbers are not comparable at all. This page sets out the shapes, the figures the vendors who publish any actually publish, and the costs that sit outside every subscription.

Every external figure below is linked to the page it came from and dated 30 October 2026, the day it was read. Prices move. The link is the authority; the number is a reading.

Two rules this page keeps. We do not state anybody else's price as fact: where a figure belongs to another company it is quoted as what that company's own page said on that date, never as what they will charge you, and third-party reports of a price are left out entirely. And we do not restate our own prices here, because they live on one page that reads the live list, and a copy of a number is a copy that goes stale.

The four pricing shapes

A flat subscription, usually per location. You pay the same whether it is a quiet month or a brutal one, up to some included amount of usage. This is how we price and how most of the all-in-one platforms price. It makes budgeting trivial and it is the wrong shape if you only get six calls a month.

Per minute. You buy a block of minutes. Overruns are billed on. This is how human-staffed services price, because you are buying somebody's time and a busy month is more of it.

Per call. You buy a number of calls, whatever length. Simpler to forecast than minutes if your calls are a consistent length, and worse if one caller talks for twenty minutes.

Per seat. You pay per member of staff. Common in general software, rare in this category, and almost always the wrong shape for a front desk, because the whole point is that the thing answering is not a member of staff. Worth spotting, because a cheap-looking per-seat price multiplies.

What the vendors who publish prices publish

Human-staffed services, which are the honest cost anchor, because they are what you would otherwise be paying. From Ruby's pricing page and Smith.ai's pricing page, both read on 30 October 2026:

The right-hand column is our division of their published figures, not a rate either company advertises. Ruby's page states that the only difference between its four plans is the number of minutes, so the features do not change.

The all-in-one local platforms mostly publish nothing. Checked on 30 October 2026:

You will find third-party blogs and resellers quoting what these platforms supposedly charge, sometimes with confident ranges. Those numbers are not on this page on purpose. They are reports about somebody else's price rather than that company's own published figure, we cannot verify them, and we would object if somebody published a guess at ours. If you want a number from one of the three, ask them, get it in writing, and ask what it is per.

What we charge is on our pricing page, and deliberately only there. It is a flat monthly subscription per location across four tiers plus a negotiated Enterprise one, and that page lists each tier's included locations, team members, conversations, AI runs, SMS segments, emails and automation runs, with the per-unit rate that applies past each. Every plan starts with the same 14 day free trial, with no card. The per-unit rates are the same on every paid tier, deliberately, so the next tier up is always the better deal at volume rather than a trap.

The numbers are not repeated here on purpose. That page reads the live price list and refreshes within the hour, so a figure typed into this guide would be wrong the first time one changed, and nothing would notice. Our own rule is that a price lives in one place, and we would rather send you to it than keep a second copy.

The costs that are not the software

This is where a quote becomes misleading, and it catches people out in every direction.

Carrier fees for texting. A US business sending texts from a 10-digit number has to register with the carriers: A2P 10DLC brand and campaign registration, which carries its own one-off and recurring carrier charges separate from any software. We are not quoting a figure because the carriers set it and it changes; what matters is knowing it exists and asking who pays it. Our A2P 10DLC guide covers what the process involves.

Per-message and per-minute carrier rates. Somebody pays the carrier for each text and each minute of each call. If a vendor bundles this, it is inside their margin and you cannot see the rate. With Frontiva the texting and voice numbers sit on your own Twilio account, so those charges reach you at carrier rates and the number stays yours, which also means you keep it if you leave. The trade is honest and cuts both ways: we cannot provision a number for you and we cannot file your registration. The integrations page spells out what each connection needs from you.

Number rental. A small monthly charge per phone number, from the carrier.

Your time. The real one, and it is never in a quote. An assistant that answers from your approved facts needs those facts written down: services, prices, hours, policies, the questions you answer twenty times a day. That is an afternoon, not a project, and it is an afternoon regardless of which vendor you pick. Picking your industry at signup seeds a starting set, and the industries page lists exactly what all twenty templates seed so you can see how much of it is already written.

A worked example

A single-location dental practice, 150 calls a month averaging three minutes, plus maybe 200 texts and web chats.

On published per-minute pricing, 450 minutes is past the 500 minute band's smaller neighbours, so you are at the top plan or into overages. On published per-call pricing, 150 calls is past the 90 call plan and into overages at the published per-call rate. On a flat subscription, the 150 calls and the 200 texts and chats both sit inside an included allowance, and 300 calls next month would cost the same.

Now change the practice to one taking eight calls a month. The smallest human plan beats almost any software on price, and gets a better answer on all eight.

The crossover is real and it is not close to the middle. Low volume favours people. Repetitive volume favours software. Your phone bill already has the number that decides it, which is more than any page like this can tell you.

The other half of the example is the part the per-minute comparison misses entirely: the 200 texts and web chats. A service billed by the phone minute does not touch them.

Six questions that make two quotes comparable

1. What is the price per? Per location, per user, per call, per minute, per month. Two numbers on different units are not a comparison.
2. What is the overage rate, in writing? Then multiply it by your worst month, not your average one.
3. Does it book against my live calendar, or a copy? Only one of those is safe. Software booking against a stale export will double book you.
4. Which channels does it cover? If it only answers the phone, price the rest separately, because the rest is most of your enquiries.
5. What happens when a caller asks for a person? Test it before you sign, then describe an emergency and test it again.
6. Where does the record end up? In your own CRM, or in a portal somebody has to remember to check. A conversation nobody reads is one you paid for twice.

Two things worth being skeptical about

A per-minute price that looks like a software price. Some AI voice products are billed per minute at rates that look small next to a human service and large next to a subscription. Multiply by your real minutes before you believe either feeling.

A free tier that cannot answer anything. A trial that does not let you load your own services, prices and hours is not a trial of the thing you are buying, because answering from your own approved facts is the whole product. Ours switches everything on at a single-location size for 14 days and needs no card.