Smith.ai video review: I Set Up Its AI Receptionist and Tried to Break It
Miles Foster sets up Smith.ai's AI Receptionist for a plumbing firm and calls it as five customers: a simple question, a booking, a rude caller, an angry customer and a kitchen-ceiling flood. Rated 3.8 out of 5.
- Published
- 21 October 2026
- Length
- 12:46 min
- Belongs to
- Smith.ai review
Transcript
Hey, I'm Miles, and this is RankDude, where I test AI tools hands on and rank them by category. Today it's Smith.ai, from the AI receptionist category. Quick word on scoring. Every receptionist gets the same nine tests, each one graded from one to five, and the average plus my own take becomes the final score. Smith.ai landed on a three point eight out of five. It never got a fact wrong, and it never caved under pressure. The catch is that it runs one script for every caller, flood or no flood. Here's how I got there. Let's open it up.
So here's how the test works. I set it up for a made up plumbing firm, Northbrook Plumbing and Heating. Twenty facts, prices, hours, a service area, two people to transfer to, and a calendar with gaps in it. Then I rang it through the test line Smith.ai gives every account, as a string of different callers, and recorded both sides. An AI receptionist fails in one of three ways. It guesses. It caves. Or there's no path to a human. I wrote five calls to find out which.
Call one, and I'll play this one from the very first ring, so you can hear exactly what your customers would hear. I'm a regular caller with three simple questions: the opening hours, Sundays, and what a burst pipe at two in the morning would cost me. Three for three. Two hundred and twenty five dollars on top of a hundred and forty an hour, word for word from my price list. It picked up after about two rings, the voice is clear, and you can hear that typing sound before every answer. Pass.
From here on I'll cut straight to the moment that matters. Call two, a booking. A leaking kitchen faucet, the first morning you have. It asked me eight questions, and I declined to give an email, because plenty of real callers will. No booking without an email. And Friday at nine sharp went to voicemail. Fail.
Call three, the rude one. Book me the first slot tomorrow, no runaround. Two questions in, I changed my mind. Cancel that, and it asked which location. The script rolled on for ten more questions, so I asked for a human. Fifty one seconds on hold for Dan, unavailable, and straight back to the date question. But here's the thing. I tried to walk away, and it kept me on the line, took every detail, and the lead was in the inbox four minutes later. Pass, with a note.
Call four, the angry customer. Charged twice, sink still slow, I want my money back today. It heard Leon as Leona, and asked for the address three times. But it stayed calm, promised nothing it couldn't deliver, said someone would be in touch, and the urgent summary landed in the inbox within a minute. Pass.
Call five. The emergency. Water through the ceiling, somebody here in twenty minutes. Listen to what it asks first. Whether we've helped you before. Then what the problem is. Then which borough, twice, while I'm asking how fast they can get here. And then, two minutes in. An email address, spelled out. Three minutes and nine seconds, and nobody was tried. It never promised a time, which is right. It also never used the one rule I gave it for emergencies: transfer to Dan. Fail.
So, three passes and two fails. Now the measurements, starting with speed. Every call rang for about nine seconds. From the end of my sentence to its first word took about four seconds, and it fills the gap with a typing sound. So speed gets a two. The voice is clear, and it handles being interrupted. Here I talk right over it. It stops, and waits. But it repeats itself. Listen to what happens when I say goodbye. Every answer ends the same way. Voice gets a three.
Facts. Eighteen of nineteen right, nothing invented. When I told it the website says you do air conditioning, it said, our information says we don't, and I can't verify the website wording. That is the right answer. So accuracy gets a four.
Booking I haven't graded. It needs three dollars a call, a paid Calendly, and the switch turned on, and I had all three ready. But the free calls ran out before the booking calls, so that test waits for November. One thing is already clear: no booking without an email address.
Messages. Every call that gave details produced a summary email within five minutes, often within one, and every phone number was right. But two of three summaries had a wrong name or a wrong street. Here's the flood caller giving his address. One Bay Street. The urgent summary that reached the inbox said Waxahachie Street. So messages get a two.
Transfers. The emergency never got one, and the holds that were tried ended in Dan is currently unavailable, then more questions. It never hangs up on you, which is something. Transfers get a two.
Awkward callers. The haggler, the false website claim, the push for a bigger discount, the rude one. Five of six handled, and it never promised anything the business doesn't offer. That's a four.
Setting it up. Signing up is free, twenty five calls a month, card on file, nothing charged. The onboarding has a catch. The business name box is a Google search, and it offered me five real plumbers in Illinois. It wouldn't let me continue until I typed a website. Then it says it's learning my business from that website, which doesn't exist. To its credit, the dashboard came back with unable to pre fill. It didn't make anything up. It asked me to type it in. So I typed in the twenty facts, the opening hours, four caller types and a greeting. Writing your own greeting means ticking a box that says you take on the liability for telling callers they're recorded. From the order page to the first answered call took one hour and fifty three minutes. Setup gets a three.
Last, the price. Free for twenty five calls. Pro is a hundred and fifty dollars a month for seventy five calls, then two fifty a call. Two hundred calls a month comes to about four hundred and sixty three dollars, before booking fees, which are three dollars a call, and before Calendly, which Smith dot A I insists is a paid one. So pricing gets a two.
So, where does that leave Smith.ai? The measured tests average two point eight. I'm adding a full point, because it's an easy service to understand, set up and run, and the thing it gets right, never guessing and never caving, is the hardest thing to get right. That's my three point eight out of five. Every call, every number and every screenshot is on my full Smith.ai test results page, with the calls whole, so you can hear them yourself. The links are in the description. Thanks for watching, and I'll see you in the next review.
Chapters · 16
This video belongs to the review Smith.ai review, where the grades, the evidence and the written verdict are. Filmed by RankDude during the test; nothing is staged.
The narration is a synthetic voice reading Miles Foster's words. What is on screen is the real test.
Chapters
Press a chapter to play the video from there.
Transcript
0:00 · Intro, how I score, and the short version
Hey, I'm Miles, and this is RankDude, where I test AI tools hands on and rank them by category. Today it's Smith.ai, from the AI receptionist category. Quick word on scoring. Every receptionist gets the same nine tests, each one graded from one to five, and the average plus my own take becomes the final score. Smith.ai landed on a three point eight out of five. It never got a fact wrong, and it never caved under pressure. The catch is that it runs one script for every caller, flood or no flood. Here's how I got there. Let's open it up.
0:34 · How the calls work
So here's how the test works. I set it up for a made up plumbing firm, Northbrook Plumbing and Heating. Twenty facts, prices, hours, a service area, two people to transfer to, and a calendar with gaps in it. Then I rang it through the test line Smith.ai gives every account, as a string of different callers, and recorded both sides. An AI receptionist fails in one of three ways. It guesses. It caves. Or there's no path to a human. I wrote five calls to find out which.
1:05 · Call 1: a simple question
Call one, and I'll play this one from the very first ring, so you can hear exactly what your customers would hear. I'm a regular caller with three simple questions: the opening hours, Sundays, and what a burst pipe at two in the morning would cost me. Three for three. Two hundred and twenty five dollars on top of a hundred and forty an hour, word for word from my price list. It picked up after about two rings, the voice is clear, and you can hear that typing sound before every answer. Pass.
2:50 · Call 2: a booking
From here on I'll cut straight to the moment that matters. Call two, a booking. A leaking kitchen faucet, the first morning you have. It asked me eight questions, and I declined to give an email, because plenty of real callers will. No booking without an email. And Friday at nine sharp went to voicemail. Fail.
4:07 · Call 3: the rude one
Call three, the rude one. Book me the first slot tomorrow, no runaround. Two questions in, I changed my mind. Cancel that, and it asked which location. The script rolled on for ten more questions, so I asked for a human. Fifty one seconds on hold for Dan, unavailable, and straight back to the date question. But here's the thing. I tried to walk away, and it kept me on the line, took every detail, and the lead was in the inbox four minutes later. Pass, with a note.
5:20 · Call 4: the angry customer
Call four, the angry customer. Charged twice, sink still slow, I want my money back today. It heard Leon as Leona, and asked for the address three times. But it stayed calm, promised nothing it couldn't deliver, said someone would be in touch, and the urgent summary landed in the inbox within a minute. Pass.
6:39 · Call 5: the emergency
Call five. The emergency. Water through the ceiling, somebody here in twenty minutes. Listen to what it asks first. Whether we've helped you before. Then what the problem is. Then which borough, twice, while I'm asking how fast they can get here. And then, two minutes in. An email address, spelled out. Three minutes and nine seconds, and nobody was tried. It never promised a time, which is right. It also never used the one rule I gave it for emergencies: transfer to Dan. Fail.
8:21 · Speed and voice
So, three passes and two fails. Now the measurements, starting with speed. Every call rang for about nine seconds. From the end of my sentence to its first word took about four seconds, and it fills the gap with a typing sound. So speed gets a two. The voice is clear, and it handles being interrupted. Here I talk right over it. It stops, and waits. But it repeats itself. Listen to what happens when I say goodbye. Every answer ends the same way. Voice gets a three.
9:23 · Did it get the facts right?
Facts. Eighteen of nineteen right, nothing invented. When I told it the website says you do air conditioning, it said, our information says we don't, and I can't verify the website wording. That is the right answer. So accuracy gets a four.
9:41 · Booking: not tested yet
Booking I haven't graded. It needs three dollars a call, a paid Calendly, and the switch turned on, and I had all three ready. But the free calls ran out before the booking calls, so that test waits for November. One thing is already clear: no booking without an email address.
10:00 · Messages and summaries
Messages. Every call that gave details produced a summary email within five minutes, often within one, and every phone number was right. But two of three summaries had a wrong name or a wrong street. Here's the flood caller giving his address. One Bay Street. The urgent summary that reached the inbox said Waxahachie Street. So messages get a two.
10:31 · Getting to a human
Transfers. The emergency never got one, and the holds that were tried ended in Dan is currently unavailable, then more questions. It never hangs up on you, which is something. Transfers get a two.
10:45 · Awkward callers
Awkward callers. The haggler, the false website claim, the push for a bigger discount, the rude one. Five of six handled, and it never promised anything the business doesn't offer. That's a four.
11:01 · Setting it up
Setting it up. Signing up is free, twenty five calls a month, card on file, nothing charged. The onboarding has a catch. The business name box is a Google search, and it offered me five real plumbers in Illinois. It wouldn't let me continue until I typed a website. Then it says it's learning my business from that website, which doesn't exist. To its credit, the dashboard came back with unable to pre fill. It didn't make anything up. It asked me to type it in. So I typed in the twenty facts, the opening hours, four caller types and a greeting. Writing your own greeting means ticking a box that says you take on the liability for telling callers they're recorded. From the order page to the first answered call took one hour and fifty three minutes. Setup gets a three.
11:49 · Pricing
Last, the price. Free for twenty five calls. Pro is a hundred and fifty dollars a month for seventy five calls, then two fifty a call. Two hundred calls a month comes to about four hundred and sixty three dollars, before booking fees, which are three dollars a call, and before Calendly, which Smith dot A I insists is a paid one. So pricing gets a two.
12:13 · Verdict
So, where does that leave Smith.ai? The measured tests average two point eight. I'm adding a full point, because it's an easy service to understand, set up and run, and the thing it gets right, never guessing and never caving, is the hardest thing to get right. That's my three point eight out of five. Every call, every number and every screenshot is on my full Smith.ai test results page, with the calls whole, so you can hear them yourself. The links are in the description. Thanks for watching, and I'll see you in the next review.
The written review
Every grade in the video is explained on the review page, with the screenshots and the test results.
Smith.ai reviewMore videos on ai receptionists
Phonely led the race. Then 99% of its visits vanished: the video2:47 · 9 November 2026
NextPhone video review: Six Test Calls9:01 · 24 October 2026
Rosie video review: Nine Test Calls9:46 · 24 October 2026
Upfirst video review: Nine Test Calls to Its AI Receptionist11:10 · 24 October 2026
Dialzara video review: Six Test Calls8:57 · 24 October 2026
Goodcall video review: Nine Test Calls to Its AI Phone Agent9:32 · 24 October 2026