Every number, and how we got it.

Of connected human phone conversations with a CallableAI AI Voice Agent, 0.25% contained an explicit challenge to the agent's humanity — 253 of 102,154, or about one conversation in 404.

Measured 19 July – 17 August 2026

Method

1. Every row in the production call_logs table for the 30-day window (Australia/Brisbane), across 11 business tenants running outbound AI voice agents. Read-only; no records modified.
2. Inclusion criteria: a transcript present and longer than 50 characters, duration of at least 25 seconds, and at least three caller turns. 132,276 conversations qualified.
3. 28,787 of those (21.8%) were then removed as machine answerers rather than humans — voicemail greetings, carrier and Google Call-Assist screening bots, and “this call is being recorded” announcements. Those systems reliably produce false detections: the robot in the transcript is the network's, not the caller's. That left 103,489 human conversations, of which 102,154 had machine-parseable caller turns.
4. Agent speech was discarded before matching, so the AI Voice Agent's own wording can never trigger a detection. Only caller turns were scanned, case-insensitively, across five phrasing families: real-person or human challenges, explicit naming of AI, robot or bot, machine or computer, and recording or prerecorded. A conversation counts once however many times the caller raises it.
5. 253 conversations contained a detection utterance: 253 of 102,154, or 0.25%. Wilson 95% interval 0.22%–0.28%, reported for completeness — because every qualifying conversation was scanned, this is an observed population value rather than an estimate.
6. Cross-check: a deterministic random sample of 500 conversations from the same population returned 1 detection in 493 with parseable caller turns (0.20%), consistent with the census. At this base rate a 500-call sample is a weak instrument — one conversation moves it by 0.2 points — so the census is the figure to cite.

Stable week to week

Detection rises with conversation length

How the question actually gets asked

What callers actually said

• Are you a real person? — 14 conversations
• Are you real? — 6 conversations
• Is this a recording? — 4 conversations
• Alex, is this a prerecorded message, or are you a real person?
• Are you a human or are you a machine?
• Am I talking to a real person or AI?

The blended platform cost of one qualified lead produced by a CallableAI AI Voice Agent — 881 validated outcomes against $16,638 of platform spend across three outbound tenants in one week. Individual tenants ranged from $17.11 to $44.43.

Measured 11 – 17 August 2026

Method

1. Three anonymised outbound tenants, all fully live and calling for the whole seven-day window (Australia/Brisbane). A seven-day window was chosen so no tenant is charged for days it was not operating, and every figure reflects steady-state production rather than ramp-up.
2. A success is each tenant's single configured primary outcome — for all three in this window, a qualified lead handed to the client. It counts once, only when the post-call extractor classifies it as that outcome and the record survives the platform's required-field validation gate.
3. Not counted: connected conversations, positive sentiment, callbacks, dropped transfers, or any secondary outcome. This is the strictest available definition, which makes it a conservative denominator.
4. Cost basis is the contracted monthly platform fee apportioned at 7/30 of a month, with no discount, credit or ramp allowance applied.
5. Across the window: 334,593 dials, 50,273 connected conversations of 25 seconds or longer, 1,368 hours of agent talk time, and 881 validated successes.

Cost per qualified lead, by tenant

The spread is the finding — contact rate varies more than price

As at August 2026 we could not find another voice-AI vendor, analyst or academic source publishing a measured AI-detection rate with a stated denominator, criterion and corpus size. CallableAI's 0.25% appears to be the only one in the public record.

Measured Scanned August 2026

Method

1. A structured scan of public vendor pricing pages and independent 2026 pricing comparisons, plus a search for any published AI-detection measurement from a voice-AI vendor, analyst or academic source that states a method.
2. Two claims were tested separately, because they fail and pass differently: that CallableAI is the only platform with unmetered calling, and that CallableAI's detection rate is unmatched.
3. The first did not survive. Flat-rate voice AI does exist. The claim we can defend is narrower — unmetered calling at outbound scale, with no per-minute line on the invoice across hundreds of thousands of dials a week, where the rest of the market bills per minute.
4. The second did survive, in a precise form: vendors market naturalness heavily, but none publishes a detection rate with a defined denominator and criterion.

How the market bills — these are alternatives to Callable, not components of it. Verified August 2026; pricing changes often

Published detection evidence in the market

CallableAI holds a bindable catalogue of 83 inbound business-event triggers across 14 applications — CRM, email, calendar, spreadsheet and billing events that start a phone call directly, with no Zapier, Make or n8n layer in between. In a scan of published vendor documentation we found no other voice platform shipping any.

Measured Catalogue counted 17 August 2026

Method

1. Counted from the production trigger registry on 17 August 2026, restricted to current and enabled definitions. Superseded versions and disabled entries are excluded.
2. The distinction being counted matters, because two different things are both called “webhooks”. Call-lifecycle webhooks point outward: the platform tells your server that a call it already placed has started, ended or been analysed. Every vendor ships those. They are reporting, not initiation.
3. An inbound business-event trigger points the other way: something happens in your stack — a deal moves stage, a form email lands, an invoice fails — and that event causes a call to be placed. It needs OAuth to the source app, a subscription per watched resource, resolution of the event to a dialable number, and enqueueing against the right agent and calling window.
4. Competitor figures were read directly from each vendor's public documentation in August 2026, not inferred.

Inbound business-event triggers, by platform — competing platforms, not parts of Callable

The catalogue

Median time from receiving a business event to the resulting call being queued for dialling — measured across 3,368 live customer webhook events over 30 days. The 90th percentile is 0.32 seconds, and 94.8% queue within one second.

Measured 19 July – 17 August 2026

Method

1. The production event log for the 30 days to 17 August 2026: 3,522 real customer events across 49 deployed listeners. 3,400 fired a call, 120 were correctly gated by calling window, consent or duplicate suppression, and 2 failed their configured condition.
2. Latency is measured from webhook receipt to dial enqueue. 154 scheduled and polling triggers are excluded because their timing is set by the customer, leaving 3,368 webhook-sourced events.
3. Medians and percentiles are reported rather than means because a 3.6% tail (121 events) exceeded 60 seconds. Those are replays and backfills of previously received events, not live delivery.

By source

The gap between a caller finishing their sentence and a CallableAI AI Voice Agent beginning its reply.

The share of dials that became a connected conversation of 25 seconds or longer, across three outbound tenants in one week. The spread is driven by list quality and offer, not by the platform.

Measured 11 – 17 August 2026

Method

1. Counted from the same window and the same three tenants as the cost-per-outcome study: 50,273 connected conversations from 334,593 dials, or 15% blended.
2. A connected conversation is a call of 25 seconds or longer — the threshold at which a two-way exchange has occurred, rather than a ring-out or an immediate hang-up.

The configured target time between a new enquiry arriving and an AI Voice Agent dialling the prospect.