Most hang-ups on AI receptionists aren't about the AI

Alfredo Salkeld, a co-founder of Upfirst, published "What makes people hang up on AI receptionists? We analyzed 450,000 calls" on the Upfirst blog. Upfirst last updated the page on June 16, 2026, and we read it on September 24, 2026.

Its dataset covers 450,702 calls to 503 AI receptionists. Each receptionist had at least 200 real calls, so no single tiny account skews the result.

It counted a hang-up two ways: a call under 15 seconds, and a call where the caller never said anything. A language model then labelled each greeting for things like AI disclosure, a recording notice, and whether it opened by saying nobody was available.

Keep one caveat in view. Upfirst sells AI receptionists, and the post is a vendor analysis, not a peer-reviewed paper. The author says so himself: the numbers are associations, not proof that one setting causes fewer hang-ups.

If callers hung up because they heard an AI, you'd expect the greeting and voice to decide most of it. The study found they explain only about an eighth of the gap between businesses, roughly 13%.

Most of the gap sits with things a greeting can't change. Who calls you, why, and what kind of line you run all matter more than the words in your first sentence.

That's useful if you're weighing an AI receptionist on this fear alone. It means a hang-up rate you've heard quoted says as much about that business's callers as about its software.

Here's the finding that changes how you read a call log. About 8 in 10 of the short calls were also silent. Nobody spoke at all.

A silent five-second call isn't a customer weighing up an AI. It's a robocall dialer checking the line, a pocket dial, or a caller who never got a word out. Count those as rejections and every AI receptionist looks worse than it is.

GreetKeeper screens those out before they reach your numbers. Silent or robotic callers are dropped before the greeting, and screened calls don't count toward your plan. When you read your own call log, the short calls that remain are people.

Your instinct may be to hide the AI. This dataset points the other way. Greetings that disclosed the AI were associated with about 20% fewer hang-ups, and greetings that mentioned recording with about 30% fewer.

Two greeting habits went with more hang-ups. Opening with unavailability, such as "nobody can take your call", came with roughly 50% higher odds. Leaving out a closing question like "how can I help?" came with about 15% more.

Disclosure is also the law in places. Maine Title 10 §1500-DD has been in effect since October 1, 2025. It bars an AI from misleading a consumer into thinking it's human unless the consumer is clearly told. Utah requires regulated professions to disclose at the start of the call.

GreetKeeper's disclosure setting announces the AI in the greeting, and the recording notice is configurable alongside it. You choose the words, and the study gives you a reason to keep them.

Receptionists whose owners had added skills and business answers saw roughly 30% to 40% lower hang-up odds. Salkeld flags the obvious confounder: owners who train their receptionist also tend to do everything else well.

Either way, the direction is clear. A receptionist that can answer the caller's actual question keeps them talking. GreetKeeper answers the questions you train it on. Point it at your website for a first draft of your hours, services and FAQs, then correct it.

Open with your business name and what the caller can do right now, such as book, ask or leave details. Say it's an AI assistant, mention recording if you record, and end with a question.

Then check your call log after two weeks. With silent calls screened out, the hang-ups left are the ones a better answer can fix.