Key takeaways
- Weigh the three kinds of review differently: ranked best-of lists least, detailed star reviews more, forum threads for how products fail.
- Affiliate commissions must be disclosed clearly and conspicuously; FTC guidance says a single disclosure on a home page is not enough[2].
- Fake and AI-generated reviews are banned under an FTC rule from August 2024, but a ban does not remove them from the page[3].
- Turn every claim you read into a test call, and every hours-saved claim into arithmetic with your own call counts.
- The most useful review comes from a business like yours, at a call volume like yours, about a product version you can date.
Three kinds of review, three different weights
Reviews of AI receptionists come in three kinds, and they deserve very different weight. Ranked best-of lists sit at the top of the search results and carry the least information. Star ratings on software marketplaces and app stores carry more, if you read them the right way. Forum threads, where owners argue with each other, are the best place to learn how products fail and the worst place to judge how often they do.
The habit of trusting reviews comes from somewhere sensible. In BrightLocal's 2026 survey of US consumers, 97% said they read reviews of local businesses, and 77% said negative reviews make them less likely to use a business[4]. Software reviews borrow that trust without always earning it, because the money works differently. A plumber does not pay a commission to the neighbour who reviews his work. A software comparison page often earns one on every sale it sends.
Best-of lists: follow the money first
Many ranked lists of AI receptionists make money when you click through and buy. That is legal, and some such lists are made honestly, but the money has to be disclosed. The FTC's Endorsement Guides say a connection that might affect the weight of a review, and that readers would not expect, must be disclosed clearly and conspicuously; the Guides' own example is a reviewer earning a share of sales through affiliate links[5]. The FTC's guidance adds that readers may not understand the words affiliate link, and that a single disclosure on a home page is not sufficient[2].
Read a list the way an editor would. These are signs the ranking follows commissions rather than testing:
- Every product scores between 4.5 and 5, and the cons are things like pricing could be clearer.
- There are no call recordings, transcripts or test scripts, only screenshots lifted from the vendors' own websites.
- The same product is best overall on several lists that share a layout and an affiliate program.
- An updated-this-month date sits on top, with no change you can find against the previous version.
- The disclosure hides behind a link at the bottom of the page instead of sitting near the recommendations.
Watch for lists a vendor runs itself, too. The FTC's 2024 rule bans a business from misrepresenting that a website it controls provides independent reviews[3]. A comparison page on a vendor's own site that ranks the vendor first is marketing. That is fine as long as it reads as marketing, and the same goes for ours.
Star ratings: read the middle, check the dates
Star ratings on marketplaces and app stores are closer to real use, but they need sorting. The FTC's final rule, approved 5-0 in August 2024, bans fake reviews including AI-generated ones, buying reviews of a particular sentiment, undisclosed reviews from insiders and suppressing negative reviews through threats[3]. A ban does not empty the page, though, and machine-written text is hard to spot: 76% of US adults say it is extremely or very important to tell whether text was made by AI or a person, yet 53% are not confident they can[1].
So stop trying to catch the fakes and read for detail instead. A useful review names the kind of business, roughly how many calls it gets and what the product did on a specific kind of call. Example: dental office, about 60 calls a day, answers insurance questions well, kept mishearing patients' last names until we added a spell-back step. That one line tells you more than fifty five-star reviews saying the product is amazing.
- Sort by most recent and read back six months. These products change quickly, and a review from before a major update may describe a different product.
- Read the three-star reviews first. They tend to come from people who used the product long enough to find its edges.
- Filter for your trade. A restaurant's experience with reservations tells a law firm little about intake calls.
- Note what each complaint is about. Setup and billing gripes matter less than wrong details on calls, which you cannot fix from your side.
- Check whether the vendor replies to criticism, and whether the reply is specific or a template.
Forum threads: honest, loud and unrepresentative
Owner forums and Reddit threads are where people describe what went wrong in plain words: the caller booked for the wrong day, the transfer that rang out, the invoice that doubled. That makes them the best source of test cases you will find. It does not make them a fair sample. People post when they are angry or delighted, rarely when something works quietly, and vendors and their partners post too. Under the FTC rule an insider's review has to disclose the connection[3], but a brand-new account with one glowing post about one product still deserves a raised eyebrow.
Mine the threads for failure modes, then test for them. If three owners describe the same failure, put that exact call in your test script. If a complaint is one person's bad week, note it and move on.
Turn every claim into a test call
Reviews tell you what to test, not what will be true on your phone line. Write down each claim as you read it, then turn it into a call you can make during a trial.
| Claim you read | What to ask | Test that checks it |
|---|---|---|
| Sounds completely human | Does it cope with noise, accents and interruptions? | Call from a car or a noisy room, interrupt it mid-sentence, and have a colleague with a different accent call |
| Books appointments | Into which calendar, and what stops double-booking? | Book, reschedule and cancel; have two people ask for one slot at the same time |
| Never misses a call | What happens with three calls at once, or a failed transfer? | Three calls together, then a transfer to a phone nobody answers |
| Integrates with everything | Native connection, webhook or a third-party tool? | Ask to watch one of your own records arrive in your system |
| Set up in minutes | Minutes to a demo, or to tested rules for your business? | Time how long your own rules take to write and test |
| Saved us 20 hours a week | At what call volume, and how long were the calls? | Put your own counts into the calculator further down |
Speech accuracy deserves extra care, because it is the claim a reviewer is least able to judge for your callers. In a study of five commercial speech recognition systems, 23% of audio snippets from Black speakers produced transcripts too poor to use, against 1.6% from white speakers[6]. A reviewer whose callers all sound like the reviewer has not tested your line.
- Your team
- Caller
- System
01Your team
Collect the claims
From lists, ratings and threads; note who said it and when.
02Your team
Drop the paid rankings
Set aside lists with no clear disclosure or no real test evidence.
03Your team
Shortlist three
Products whose detailed reviews come from businesses like yours.
04Caller
Run the same test calls
One script, every claim turned into a call, on each trial line.
05System
Read the records
Summaries and transcripts side by side; mark every wrong detail.
06Your team
Decide on your evidence
Choose on test results and keep the reviews as tie-breakers.
Example plan for an owner with a week to choose. Build the test script from your own call types.
Check the hours-saved claim against your own calls
The number that turns up most in reviews is hours saved, and it is the easiest one to check. Hours saved depend on three things: how many calls you get, how long they take and what share the system completes without your staff. Put your own numbers in and compare.
| Input | Example |
|---|---|
| Calls per working day | 35 |
| Minutes per call | 3 |
| Share completed without staff | 50% |
| Staff cost per hour | $36.73 |
| Calls a month | 760 |
| Staff hours on these calls | 38 |
| Hours handed off | 19 |
| Monthly value of that staff time | $697 |
How it is worked out: Calls a day × 21.7 working days × minutes per call ÷ 60 × share handled = hours; hours × hourly cost.
Assumptions
- Hourly cost of $36.73: private employers' average total compensation per hour worked for office and administrative support workers in June 2026.[7]
- 35 calls a day, three minutes each and half completed without staff are invented for a small clinic or trade office; use your own call log.
Example only: about 19 hours a month in this model, a little over four hours a week.
Now hold a review up against it. At three minutes a call with half the calls handled, saving 20 hours a week, about 87 hours a month, would take roughly 160 calls a day. If the reviewer runs a two-person office, either their calls are far longer than yours or the number is a guess. Neither makes it your number.
Vendor claims deserve the same treatment. When the FTC launched Operation AI Comply in September 2024, it said there is no AI exemption from the laws on the books[8]. One of the cases involved a service marketed as the world's first robot lawyer, which the complaint says was never tested against the level of a human lawyer[9]. Claims about what an AI product can do have to be backed up, and you are entitled to ask how.
Keep your own review sheet
Write down your call mix first
The five call types you get most, taken from your own log, before you read a single review.
Log each claim with its source
Who said it, when, about which version of the product, and whether they disclosed a connection.
Score test calls, not adjectives
For each product: did the call end the way your rules say, and were the name, number and time right?
Ask every vendor the same questions
One list for all of them, so the answers line up side by side.
Review again after 30 days
Write your own review of the product you chose, with call counts. It will be the most useful one you read.
For the questions themselves, use the 15 vendor questions, and for an honest list of trade-offs read AI receptionist pros and cons. The family law intake example shows a set of test calls that runs every branch before a real caller needs it, and the home services AI receptionist evaluation checklist has a version for the trades.
If price is pushing you toward a lesser-known product, read what free tools leave out first, and if you are moving from a current provider, plan the switch from your answering service around a test week. Whatever you choose, an SI Employee included, judge it the same way: on calls you made, records you read and a decision you can explain to your team.
Frequently asked questions
Are AI receptionist reviews on Reddit reliable?
They are useful for spotting failure modes, such as wrong bookings or transfers that ring out, and weak for judging how common those failures are. Treat each thread as a test case to try yourself, not as a verdict.
How can I tell if a best-of list is paid?
Look for a clear disclosure near the recommendations, not only a footer link. Uniformly high scores, no test calls or recordings, and the same winner across lookalike sites are further signs the ranking follows commissions.
Should I trust reviews on a vendor's own website?
Read them as marketing. Ask the vendor whether the reviewers were paid or given discounts, and whether you can speak to a business in your trade. Then rely on your own test calls.
How many reviews should I read before choosing?
Enough to collect the claims and failure modes worth testing, usually a few dozen across lists, marketplaces and forums. After that, more reading adds less than one hour of test calls on a trial line.
What if a product has few reviews?
Few reviews mean less evidence either way, not a worse product. Lean harder on a trial with your own calls, ask for a test line, and check that call records are complete enough to judge the results.
References
Pew Research Center (2025). How Americans View AI and Its Impact on People and Society (opens in a new tab) ↑
Federal Trade Commission (2026). FTC's Endorsement Guides: What People Are Asking (opens in a new tab) ↑
Federal Trade Commission (2024). Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials (opens in a new tab) ↑
BrightLocal (2026). Local Consumer Review Survey 2026 (opens in a new tab) ↑
Legal Information Institute, Cornell Law School (2026). 16 CFR 255.5: Disclosure of material connections (opens in a new tab) ↑
Proceedings of the National Academy of Sciences (via PubMed Central) (2020). Racial disparities in automated speech recognition (PNAS 117(14)) (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2026). Employer Costs for Employee Compensation, Table 4: private industry workers by occupational and industry group, June 2026 (opens in a new tab) ↑
Federal Trade Commission (2024). FTC Announces Crackdown on Deceptive AI Claims and Schemes (opens in a new tab) ↑
Federal Trade Commission (2024). FTC Announces Crackdown on Deceptive AI Claims and Schemes (DoNotPay complaint summary) (opens in a new tab) ↑