Disclosure: Futuro Corporation publishes this page and sells Futuro AI, an AI receptionist. We did not accept payment for any mention, and no one paid for placement. Surveys commissioned by companies that sell customer-service software (Five9, Gladly, Genesys) are labeled as such.
What do surveys say about AI answering the phone?
Ask the public how it feels about AI answering the phone and the answers are grim. According to a July 2024 Gartner press release (5,728 customers, fielded December 2023), "Sixty-four percent of customers would prefer that companies didn't use artificial intelligence (AI) in their customer service," and "53% of customers would consider switching to a competitor if they found out a company was going to use AI for customer service." The top concern was that AI will make it "more difficult reaching a person." According to Keith McIntosh, Senior Principal, Research, at Gartner, "Many customers fear that GenAI will simply become another obstacle between them and an agent."
Later surveys read the same way. In a survey commissioned by Five9, which sells contact-center software (4,000 US and UK consumers, September 2024), "75% prefer talking to a real human in-person or over the phone for customer support," "nearly half (48%) of consumers say they do not trust information provided by AI-powered customer service bots," and "Over half (56%) of all consumers are often frustrated by AI customer-service chatbots." According to the Pew Research Center, "Half of U.S. adults say the increased use of AI in daily life makes them feel more concerned than excited," against 10% more excited, a concern share that has climbed from 37% when Pew first asked in 2021. And Qualtrics's 2026 Consumer Experience Trends report (20,000+ consumers, 14 countries) found "Nearly one in five consumers who have used AI for customer service saw no benefits from the experience. That's a failure rate almost four times higher than for AI use in general."
Why don't survey answers predict hang-ups?
These are stated-preference questions about AI in general, not measurements of anyone hanging up, and the same public keeps using AI: Pew finds 31% of Americans now interact with AI at least several times a day, up from 22% in February 2024. According to Gartner's September 2026 release (3,566 customers), "While 49% of customers said they would have been willing to use a chatbot if the company had provided one, only 7% used a chatbot or digital assistant during their most recent service interaction." The failure underneath is resolution: according to Gartner's August 2024 release, "Only 14% of customer service and support issues are fully resolved in self-service," and "Even for issues that customers describe as 'very simple', only 36% resolve fully in self-service." According to Eric Keller, Sr Director Analyst at Gartner, in the September 2026 release, "Customers give chatbots a short leash"; "one unsuccessful interaction can discourage future use, even as the chatbot's capabilities improve." Wording moves answers too: "would you prefer companies didn't use AI?" (64% yes) measures default bias, while "do you trust AI when a human is reachable?" (55%, Five9 2026) measures designed trust.
What does measured call data show?
Measured call data exists for the two categories on either side of the AI receptionist: phone queues staffed by people, and the phone automation that came before modern voice AI.
Do callers hang up on a human hold queue?
Yes, steadily. According to ContactBabel's US Decision-Makers' Guide 2024 (189 contact centers), US call abandonment was 7.1% in 2023, inside a band that "continues to move between a range of 5-7%," while mean speed to answer stretched from 31 to 79 seconds; the UK report recorded a mean of 8.4% with median abandonment of 6.0%, and its long-run series is the useful context: "Call abandonment rate did not show any particular upward trend, with all but three of the 14 survey results showing data between 5% and 6% between 2004 and 2019." According to Gans, Koole and Mandelbaum's call-center review (2003), "Customers who become impatient hang up, or abandon, before being served." The "Gartner 5–8% benchmark" is not Gartner's: we found no Gartner publication with it, and SQM Group's guidance says "the call center industry standard for abandonment rate is between 5% and 8%."
How often do callers bail out of automated phone menus?
The closest measured series is ContactBabel's "zeroing-out" rate: calls that enter phone self-service and leave it for an operator. In the US guide it rose from 13% in 2013 to 32% in 2023; the UK report puts it at 14%, and "71% of respondents agreed that customers abandoned self-service sessions because the self-service function simply does not offer what the customers want," and 72% agree to some extent that "the customer simply does not trust the system, preferring to have human reassurance that the request they have made has been carried out." Two asterisks: a zero-out is a request for a person, not a hang-up, and the data covers keypad menus and early speech systems, not conversational AI. Still, automation that can't do the job loses far more callers to the operator queue (32%) than a slow human queue loses to hang-ups (7.1%).
How should you read vendor-published AI numbers?
Read vendor completion figures the way you'd read a restaurant's own reviews: directionally useful, structurally flattered, and meaningless without call volume, time window, and a definition of "hang up." Vendor survey data with a stated sample is more useful. In Five9's 2026 Consumer CX Report (3,000 consumers, April 2026), the top reason people used AI at all, at 52%, "was that it was the only option offered. That isn't adoption. It's a detour."
What decides whether a caller hangs up on an AI receptionist?
Four variables explain most of the gap between an AI receptionist callers stay with and one they abandon: voice and latency, task match, disclosure, and the escape hatch.
Does a robotic-sounding voice make callers hang up?
The first test happens in the greeting. Synthetic voices are now hard to spot: in a 2025 UC Berkeley study in Scientific Reports, listeners "correctly identified a voice as AI generated only about 60% of the time." A voice that still sounds mechanical invites the reaction older research predicts: according to Dietvorst, Simmons and Massey (2015), "people more quickly lose confidence in algorithmic than human forecasters after seeing them make the same mistake." A robotic voice is that first visible mistake.
Does a slow reply make callers hang up?
Conversation runs on tight timing. A ten-language study of turn-taking led by Tanya Stivers (PNAS, 2009) found a "unimodal peak of response within 200 ms of the end of the question," with a cross-language median of about 100 ms. An AI that leaves a long silence before each answer breaks a rule callers follow without thinking. No published study yet measures hang-ups against AI reply latency, so treat this as a design principle, not a measured rate.
Which calls will people let an AI handle?
Acceptance belongs to the request more than the caller. Five9's 2026 task data, reported by CMSWire, shows 38% accept AI alone for a password reset and 35% for order status, but only 15% for refunds, where 45% want a human only. For high-stakes escalations, 60% want a human only. Routine lookups and changes sit at the high-acceptance end; disputes and anything with money or risk attached do not.
Why do people trust algorithms for some tasks but not others?
The research predicted the task split. Logg, Minson and Moore (2019) found people weigh advice more when it comes from an algorithm than from a person on estimation tasks, which they call "algorithm appreciation." Dietvorst's aversion effect appears once people watch the system make a mistake. Both pre-date voice AI, but together they describe what callers do: easy acceptance for routine answers, and a sharp pullback after the first visible error.
Does saying it's an AI make callers hang up?
Callers want to know. According to Five9's 2026 report, 71% say it is very or extremely important to know whether they're talking to a bot. So the design questions are what to say and when. Concealment is not a safe option: according to the FCC's February 2024 declaratory ruling, "the TCPA’s restrictions on the use of “artificial or prerecorded voice” encompass current AI technologies that generate human voices." That ruling covers outbound calls, not inbound reception, but it shows how regulators classify the technology.
What happens when callers can't reach a person?
They leave, often with the sale. In a Gladly-commissioned survey (1,000 US consumers; Gladly sells customer-service software), "When customers are blocked from reaching a human, 40% give up entirely or purchase elsewhere." According to Gartner's August 2026 release, "87% of customers say it is essential for companies to provide an option to reach a human agent when using GenAI," and for unwilling customers "the most common response was the ability to switch to a human agent if needed."
How much does a visible human option change trust?
More than any other lever measured so far. In the Five9-commissioned 2026 survey, trust that AI can resolve an issue rose from 26% with no human option to 55% when one was visible; without one, Silent Generation trust fell to 5% (CMSWire). Genesys's 2026 State of CX report (Genesys sells contact-center software) adds that "84% will give a virtual agent up to three attempts to resolve their issue." Gartner's September 2026 guidance is that chatbots "should serve as seamless connectors to human support, not containment traps," and its August 2026 release adds that "Service leaders should not use GenAI as a mandatory first step for every issue."
How long will callers give an AI before asking for a person?
Not long. According to Five9's 2026 report, "92% of consumers give AI fewer than 10 minutes before asking for a human." The top reason for switching is the bot repeatedly misunderstanding them (46%), and 35% often or always repeat themselves after the handoff. Misunderstanding, not automation itself, sends callers to the exit. The fixes: narrow the AI to calls it can finish, and pass the details on.
Does an escape hatch that loops back to a menu count?
No. ContactBabel's 2024 guides put it bluntly: "if businesses don't offer a live agent option to an irate and frustrated caller, they won't need to worry about providing customer service to them in the future, as they'll go elsewhere." An escape hatch that transfers to a person with the details collected turns a would-be hang-up into a completed call. One that loops back to a menu shows the caller the trap.
Where does the evidence cut the other way?
A page that claims nobody minds talking to AI would fail its own standard. Resistance is real, measured, and concentrated in four places.
Early disclosure can cost the sale. The strongest measured counterpoint is a field experiment. According to Luo, Tong, Fang and Qu's 2019 study in Marketing Science (more than 6,200 customers receiving outbound sales calls; pre-dates modern voice AI), "a disclosure of chatbot identity before the machine–customer conversation reduces purchase rates by more than 79.7%," although the effect "can be mitigated by a late disclosure timing strategy and customer prior AI experience." A behavioral experiment points the same way: according to Ishowo-Oloko and colleagues in Nature Machine Intelligence (2019), "bots do better than humans at inducing cooperation, but that disclosing their true nature negates this superior efficiency." Honesty and conversion can pull in different directions, which is why wording and timing deserve testing.
Older callers. In Five9's 2024 survey, preference for a human ran 66% for Gen Z, 76% for Gen X, and 86% for Baby Boomers and the Silent Generation. In the 2026 data, when no human option exists, trust that AI can resolve the issue falls to 16% among Boomers and 5% among the Silent Generation. If your callers skew 60+, offer the human option in the greeting and set your completion target lower.
High-stakes calls. Five9's task table shows resistance climbing with the stakes: 45% human-only for refunds, 51% for sensitive identity verification, 54% for legal compliance questions, 60% for high-stakes escalations. Let the AI capture details, then hand to a person.
Repeat failures. Gartner's 27% retry figure means most customers won't give automation a second chance after it fails once. Qualtrics's one-in-five-no-benefit finding is what those fumbles cost at population scale.
Who should not put an AI receptionist on the phone?
For some businesses the honest answer is "not yet," or "not alone."
- Lines where most calls are high-stakes. Legal intake, collections, billing disputes, and urgent medical questions are high-stakes calls; in Five9's 2026 data, 54% want a human only for legal compliance questions and 60% for high-stakes escalations. Use AI for intake and routing there, not resolution.
- Businesses with no one to take the handoff. If no person can pick up or call back promptly, the escape hatch is empty, and callers who are blocked from a person are the ones most likely to give up (40% in the Gladly-commissioned survey).
- Caller bases that skew very old, with no human fallback. Trust without a human option is 5% among the Silent Generation in Five9's 2026 data.
- Anyone planning outbound AI calls without consent. Under the FCC ruling, "calls that use such technologies require the prior express consent of the called party."
- Anyone who wants to hide that the caller is talking to AI. 71% of consumers say knowing matters to them, and the FCC already classifies AI voices as artificial.
How do you estimate your own hang-up exposure?
Three questions get you most of the way.
1. What is your call mix? Sort your last fifty calls: bookings, status and hours questions, complaints, everything else. The first two are the routine requests surveys show the highest acceptance for. If most calls are those, your exposure is lower and concentrated in the greeting.
2. Who is calling? Age shows the steepest gradient in the surveys that report it. The older your callers, the earlier the "say 'agent' anytime" line belongs.
3. Where does the exit lead? A visible escape hatch that transfers with context converts a would-be hang-up into a completed call, and the 26%→55% trust swing suggests its presence holds callers who never use it.
Then compare honestly against today: the calls a small team misses, the hold queue that loses 5–7%, and the voicemail many callers won't leave. Measure early and mid-call abandonment separately for two to four weeks. For the buying side, see our conversational AI buyer's guide and whether AI can answer your business phone.
How did we source and calculate the numbers on this page?
Sources by tier. Tier 1 is peer-reviewed research and government documents: Dietvorst et al. (2015), Logg et al. (2019), Luo et al. (2019), Ishowo-Oloko et al. (2019), Stivers et al. (2009), Gans et al. (2003), Barrington et al. (2025), and the FCC's 2024 ruling. Tier 2 is independent research organizations: Gartner, Pew Research Center, Qualtrics, and ContactBabel. Tier 3 is surveys commissioned by companies that sell customer-service software (Five9, Gladly, Genesys), labeled at every use. Each figure is quoted as published, with its sample and fieldwork date; we did no calculations beyond setting two published figures side by side.
Limits. No source measures hang-up rates on AI receptionists directly. Survey figures measure what people say, not what they do on a call. ContactBabel's zero-out figures count requests for a live agent, not hang-ups, and predate modern voice AI, as do the academic studies. We did not place test calls to any AI receptionist, we did not use any Futuro AI platform data, and we did not read the full Five9, Gladly or Genesys reports beyond their public releases and posts. Gartner's newsroom blocks automated tools, so its releases were read through a fetching service and quotes checked against that text.
Last reviewed November 15, 2026. Next scheduled review: February 2027. Corrections go through our editorial standards process.