Do callers hang up when an AI answers the phone?

Not for small businesses. We looked for peer-reviewed studies that count how many callers hang up when an AI answers a business line, and found none. The hang-up percentages that circulate online usually trace back to vendor surveys with no published method, so we have left them out.

What does exist is close enough to be useful. One field study followed real inbound calls at a large telecom company after it switched to voice AI¹. Another randomized thousands of outbound sales calls between bots and people². A large survey asked customers how they feel about AI in customer service³. The table shows what each one measured, because that decides how far you can lean on it.

A large telecom company replaced its old press-1, press-2 phone menu with a voice AI system, and researchers compared calls before and after¹. At first, callers spent longer with the machine and asked for a person more often. That bump faded. What lasted was a drop in complaints¹.

The drop was not even. For simple requests, complaints fell for new and experienced callers alike. For complex requests, callers complained less once they had used the AI before. Older callers, women and heavy users of the old menu saw the biggest improvement¹. And when the AI misheard a caller, that caller was more likely to ask for a person and more likely to complain¹.

The study counted call length, requests for a person and complaints, not hang-ups, and it covers one large company¹. A small business with a different mix of calls may see a different result.

In one experiment it mattered a lot. More than 6,200 customers of one company were randomly given structured outbound sales calls from a bot or a person². Bots that did not say what they were sold as well as proficient staff, and four times better than inexperienced staff. When the bot said it was a machine before the conversation began, purchases fell by more than 79.7%, and customers were curter². Telling them later, and customers having used AI before, softened the drop².

That was a sales call the customer did not ask for. Someone calling your business already wants something, so the same effect may be smaller. Hiding the AI is not the answer either: some states require telling customers in some cases, which our disclosure law guide covers. And in a Gartner survey of 5,728 customers, 64% said they would prefer companies not use AI for customer service, 53% said they would consider switching, and the top concern was that it would get harder to reach a person³. That was a stated preference, not observed behavior³, but it points at the fix: make the AI quick and useful, and make a person easy to reach.

Angry ones. Across five studies, including one company's real service data, customers who arrived angry rated the company lower, were less satisfied and less likely to buy when the bot had a human-like name and persona. Calm customers were not affected⁵. Those studies were about text chat, but the lesson carries: a caller who is already upset needs a person, fast.

Callers the AI cannot understand come next. In the telecom study, speech-recognition failures drove up both requests for a person and complaints¹. Test your AI receptionist with accents, background noise and the words your customers actually use, and make sure "let me talk to someone" works on the first try.

The real comparison is not AI versus a perfect receptionist. It is AI versus the call that goes nowhere. In a Pew survey of 10,211 U.S. adults, 80% said they generally don't answer their cellphone when an unknown number calls, and 67% let it ring and check voicemail instead⁴. A missed caller you try to ring back may never pick up.

So judge an AI receptionist by what callers get from it: an answer, a booking or a quick handoff. Listen to your first few weeks of calls, note where callers ask for a person, and fix those spots first.