Do customers actually hate AI phone calls?

The honest answer is that customers do not hate AI phone calls the way the headlines suggest. They hate being trapped. People are wary of AI in the abstract, yet most are satisfied once an AI call actually resolves their problem, and modern voices are convincing enough that in controlled tests listeners identify an AI voice correctly only about 60 percent of the time. What callers reliably dislike is a phone tree or a chatbot that cannot help and will not let them reach a person. This guide walks through what the research actually says, so you can decide with evidence rather than vibes.

Do customers hate AI on the phone?

The picture is genuinely split, and it is easy to cherry-pick either half. On one side, a lot of people are uneasy about AI. Gartner found 64 percent of customers would prefer that companies did not use AI for customer service (Gartner, 2024), a figure we take seriously in our AI receptionist vs human receptionist comparison. The wider mood is wary too: 40 percent of US adults expect AI to have a negative effect on society over the next 20 years, even as about half already use AI chatbots (Pew Research Center, 2026).

On the other side, actual experiences land better than the attitudes. In global research, 74 percent of customers said they were satisfied with their most recent AI customer service interaction (COPC, 2026). That gap between what people say about AI in general and how they rate a specific interaction is the whole story. The abstract worry is real. The lived experience, when the AI works, is mostly fine.

What people actually hate

Dig into the complaints and they are rarely about the existence of AI. They are about being stuck. In a 2025 UK survey, the number one frustration when calling customer service was waiting on hold for too long, named by 44 percent, and one in four respondents said they immediately try to reach a live person the moment they hit automated options (TCN, 2025). The same survey found 78 percent are likely to abandon a brand after a single poor service experience, so the cost of trapping someone is high.

Government data tells the same story about automation done badly. In a US study, 80 percent of consumers who dealt with a chatbot came away more frustrated, and 78 percent still needed to reach a human afterwards (Consumer Financial Protection Bureau, 2023). The villain in those numbers is the doom loop: endless prompts, repeated questions, and no person at the end. It is the same feeling as an old phone tree that will not let you press zero.

This is why the fear that matters most is losing the human. Gartner found 60 percent of customers worry AI will make it harder to reach a person, and 53 percent would consider switching to a competitor if a company used AI for service (Gartner, 2024). A voice agent that answers instantly and hands off cleanly addresses the exact thing people are afraid of. One that dead-ends them confirms it.

Does telling people it is AI hurt?

Here is the most interesting tension in the research, and it is worth being honest about. There is real evidence that disclosing AI can reduce trust. Across 13 experiments, people who revealed they used AI were trusted less than those who did not (Schilke and Reimann, 2025). Taken alone, that reads like an argument for hiding it.

It is not, and the customer service data shows why. In practice, people who knew they were interacting with AI reported satisfaction 34 percentage points higher than people who were not told (COPC, 2026). The same research noted that markets leading on disclosure, including Malaysia and Singapore, also lead on overall satisfaction and comfort with AI. The two findings reconcile cleanly: a bare admission with nothing behind it can cost you, but disclosure paired with a competent experience and an obvious route to a human builds trust rather than spending it. Being upfront is also increasingly expected, and for outbound calling it sits alongside real rules on consent and identification, which we cover in PDPA and DNC compliance for outbound calling.

How human do AI voice agents sound?

Convincing, and worth being precise about rather than hyping. In a controlled study, when a clip was a real human voice, listeners were right 67.4 percent of the time; when it was an AI clone, they were right only 60.8 percent of the time, barely above a coin toss, and they judged an AI voice to be the same as its real counterpart around 80 percent of the time (Barrington and colleagues, Scientific Reports, 2025).

A second peer-reviewed study reached a similar and usefully calibrated conclusion: voice clones can sound as real as human voices, and listeners labelled AI voices as human in most trials, but the researchers found no hyperrealism effect, meaning the AI voices were not more convincing than real ones (Lavan and colleagues, PLOS ONE, 2025). So the honest phrase is convincing, not indistinguishable. Ignore any vendor claiming its voice is impossible to tell from a human; the science does not support the absolute.

What this means for your business

Put the findings together and the playbook is simple. People do not hate a competent AI that respects their time; they hate being trapped by a bad one. So three things decide whether your callers are happy:

• Say it is AI. Disclosure paired with a good experience raises satisfaction, and it is increasingly expected. Do not pretend a bot is a person.
• Make it genuinely good at the routine. The voices are convincing enough; the difference is whether the agent actually answers the question, books the slot or takes the message without a loop.
• Always leave a clean path to a human. The single biggest fear is being unable to reach a person, so the agent should transfer complex, sensitive or high-value calls by clear rules, not bury the exit.

Done that way, an AI voice agent is not the thing customers dread. It is the thing that answers on the first ring instead of leaving them on hold. For the full picture of the category, see what an AI voice agent is and what an AI receptionist is; for where a person still has the edge, our AI receptionist vs human receptionist comparison lays it out honestly.

Frequently asked questions

Do customers like talking to AI on the phone?

It is mixed, and context decides it. Many people are wary of AI in the abstract, yet most are satisfied when an AI call actually resolves their issue. In global research, 74 percent of customers were satisfied with their most recent AI interaction (COPC, 2026). What people dislike is not the AI itself so much as being stuck with automation that cannot help and will not pass them to a person.

Can you tell if a phone voice is AI?

Often not. In a controlled study, listeners correctly identified an AI voice clone only about 60 percent of the time, barely better than chance, and judged an AI voice to be the same as its real counterpart around 80 percent of the time (Barrington and colleagues, Scientific Reports, 2025). Modern voices are convincing, though researchers found they are realistic rather than more convincing than real humans.

Do AI voice agents sound human?

They sound close. Peer-reviewed work found voice clones can sound as real as human voices, with listeners labelling AI voices as human in a majority of trials, but the same study found no hyperrealism effect, meaning AI voices are not yet more convincing than real ones (Lavan and colleagues, PLOS ONE, 2025). The honest description is convincing, not indistinguishable.

Does telling customers it is an AI hurt trust?

There is a real tension. Across 13 experiments, people who disclosed using AI were trusted less than those who did not (Schilke and Reimann, 2025). Yet in customer service, people who knew they were dealing with AI reported satisfaction 34 points higher than those who were not told (COPC, 2026). The reconciliation is that disclosure plus a competent experience and a clean path to a human builds trust; hiding it, or disclosing a bad experience, does not.

Do people prefer humans over AI for support?

For complex, sensitive or high-stakes calls, yes. In one UK survey, 52 percent would rather give secure information to a live person than an automated system, and 42 percent still prefer a live agent by phone (TCN, 2025). The practical answer is not AI or humans, but AI for routine calls with an easy, obvious handoff to a person when the caller needs one.

When should a call be handed from AI to a human?

Whenever the caller asks, and whenever the call is complex, sensitive, high-value or emotional. The biggest driver of AI frustration is being unable to reach a person: 60 percent of customers worry AI makes that harder (Gartner, 2024), and a quarter of callers try to reach a human the moment they hit automation (TCN, 2025). A well-built agent answers routine calls and routes the rest to people by clear rules.