AI Receptionists · updated Sep 29, 2026
How this index works
Every figure in this index is read from a source you can visit, dated on the day it was verified, and corrected in public — free — when it turns out wrong. Here is exactly how.
Who we are. This is the AI Receptionists section of The Front Desk Review. We publish structured, verified market data about AI receptionist and AI answering services. We are not a vendor or an agency, and no company listed here pays us.
Where the data comes from. Every data point on this site carries its own provenance: the source URL, the date we captured it, and how it was verified. We collect data three ways:
- Public sources. We monitor vendors' published pricing and
feature pages on a weekly schedule. We respect robots.txt and rate limits. Each
day we ask the Internet Archive's Save Page Now to capture pricing pages we cite
and have read in the past month, up to a daily limit — never a page that refuses
our crawler or that robots.txt closes to us — and each capture is recorded with
the date it was taken. Where a second source confirms a
figure, we attach it as corroboration — read on its own day, not necessarily the
day the first source was read, which is why both access dates travel with the row
(
accessed_atandcorroboration_accessed_at). - Direct measurement. Benchmark data, where we publish it, comes from scripted test calls placed to vendors' public demo lines. Recordings and transcripts are retained, and the methodology for each run is published with it.
- Corrections. Anyone — vendor, reader or competitor — can send a correction by email, free of charge; no account, and no vendor payment is ever required or accepted. Verified corrections are applied and logged in the public corrections record with the date.
Corrections. Anyone can flag an error. We investigate and correct verified errors within 5 business days, free of charge, and every correction is recorded in our public corrections log.
-
AI Receptionists — Trillet — Studio plan. Price/quota corrected from a seeded $49 / 150 min to the live pricing page's $99 / 100 min.
-
AI Receptionists — AIRA & Upfirst. Flagged byte-identical pricing tables (an apparent shared pricing engine); both recorded independently with their own sources.
-
AI Receptionists — AnswerConnect. Recorded as quote-only after its pricing page returned 404 at capture; an unverifiable ~$325/mo third-party figure was not published.
Read the full corrections log — every entry dated.
Freshness. Pricing rows show their "last verified" date. Rows we have not been able to re-verify within 30 days are flagged STALE rather than silently retained.
A row counts as re-verified when a page it cites is read again and still publishes the same figure — the source page or the independent corroboration, whichever was read most recently. Both are listed with their own dates, so you can check either. A new point on our own price curve does not count and never clears the flag: we write that ourselves, and a figure cannot vouch for itself.
The cost normalizer
Flat, per-minute and per-call plans cannot be compared until the workload is fixed. Our headline figure is the effective monthly cost at 200 calls per month at 3 minutes per call. For each vendor we take the cheapest plan that can be priced at that workload:
- Flat plans contribute their base fee, plus any published overage beyond an included minute or call quota.
- Per-minute / subscription-plus-usage plans contribute their base fee plus the published per-minute rate times the workload's talk-minutes.
- Per-call plans contribute their base fee plus the published per-call overage beyond the included call quota.
- Unlimited plans contribute their base fee, independent of volume.
When pricing a workload would require an overage rate the vendor never published, we return no figure for that plan rather than fabricate one. A vendor whose every plan needs an unpublished rate — or that is quote-only — appears with a dash, an honest gap.
Named measurements
Beyond pricing, we publish our own named measurements of how each service performs on a real call. Each is defined once, measured by scripted live test calls, and gated: a vendor's value is published only when it is backed by at least 10 test calls — below that it stays "insufficient data," never an estimate.
- Booking-Success Score — Share of decisive scripted test calls in which the AI receptionist completed a booking/appointment action. Published per vendor only at n ≥ 10 test calls.
- Voice-Agent Latency Index — Median (p50) time from call connect to the agent's first spoken response across scripted test calls. Published per vendor only at n ≥ 10 test calls.
Measurement set version Jun 11, 2026. Values download as JSON.
Frequently asked
Who publishes this index and what is it?
The Front Desk Review publishes structured, verified market data about AI receptionist and AI answering services. We are not a vendor or an agency, and no company listed here pays us.
Where does the data come from?
Every data point carries its own provenance: the source URL, the date we captured it, and how it was verified. We monitor vendors' public pricing and feature pages on a schedule, respecting robots.txt and rate limits, and attach a second independent source where one confirms a figure. Each day we ask the Internet Archive's Save Page Now to capture pricing pages we cite and have read in the past month, up to a daily limit, and record each capture with the date it was taken; a page that refuses our crawler, or that robots.txt closes to us, is not sent. Corrections are free and open to anyone — vendor, reader or competitor — by email, no account or payment required.
Can a vendor pay to affect ranking or inclusion?
No — nobody can pay to be listed, ranked, scored, corrected faster or removed, and no company in this section pays us anything. We earn from Amazon's affiliate program: Amazon pays us a percentage of qualifying purchases made through our Amazon links, at rates it sets per product category, so within a category a pricier purchase pays us more. Our grades and sort orders never read commission data. Default table ordering is computed by a published, versioned algorithm from the data itself.
How are corrections handled?
Anyone — vendor, reader, or competitor — can flag an error. We investigate and correct verified errors within five business days, free of charge, and every correction is recorded in our public corrections log.
How fresh is the data and what does the normalized cost mean?
Pricing rows show their last-verified date; rows not re-verified within 30 days are flagged STALE rather than silently retained. The normalized 'cost at 200 calls/mo' converts every billing model — flat, per-minute, per-call, subscription-plus-usage — to one effective monthly cost at a fixed workload. Where pricing a workload would require an overage rate the vendor never published, we omit the figure rather than invent it.