Some customers really don't want to talk to an AI

If you've been told everyone will happily talk to an AI by now, the data says otherwise. Preference for a real human in customer service is going up, and the share of people who'd hang up on a bot is going up with it. This post takes that seriously: why it's happening, what it means if you're thinking about putting an AI on your phone line, and the handful of things that actually earn a caller's trust back.

What the numbers actually say

AnswerConnect ran a 6,000-person survey through OnePoll across the US, UK and Canada, comparing responses in October 2025 with responses in May 2026. Over those months, every figure moved in the same direction — away from AI.

Separately, the share of people who say they feel excited about AI in 2026 sits at 19%, down from 50% two years earlier. The word being used for that is AI fatigue — not fear, not outrage, just being worn out by the sheer amount of it.

These aren't enormous swings. A two-point rise in a survey isn't a collapse in confidence. But the direction matters more than the size. Plenty of people assumed familiarity would soften attitudes: the more people used AI, the more relaxed they'd get. Over this period, the opposite happened. Exposure went up and enthusiasm went down.

There's one more finding worth sitting with, because it's behavioural rather than an opinion: people repeatedly say or type "human" or "agent" to try to get past a bot. That's not a survey response. That's what people do with their hands and their voice when they're stuck. It tells you the bot is being treated as an obstacle between the customer and the thing they came for.

Why people are fed up — and it isn't really about AI

The easy reading is "people don't like robots". I don't think that's it. Most of the frustration traces back to a specific, older experience that has nothing to do with modern voice AI at all.

Fifteen years of IVR menus

Press 1 for accounts. Press 2 for billing. Please listen carefully as our options have changed. Anyone who has tried to sort out a broadband fault or a mis-sold insurance policy has spent time in a phone tree that had no branch for their actual problem. The menu wasn't built to help you. It was built to sort you cheaply, and if your issue didn't match a category, tough.

That experience trained a generation of callers to treat any automated voice as a gate. So when a new AI voice answers, people aren't evaluating it fresh. They're reacting to the last ten times something automated wasted four minutes of their evening.

Chatbots that can't do anything

The website chat widget made it worse. Many of them can only search a help centre and paste back an article you'd already read. Ask for something specific — cancel this, refund that, change the date — and you get a loop. Hence typing "agent" five times in a row.

The lesson customers drew wasn't "this chatbot is bad". It was "the company put this here so it wouldn't have to deal with me". Once someone believes that, every automated interaction reads as a brush-off. We've written about a similar credibility problem on the outbound side in why AI sales outreach is getting a rough reception — the pattern is the same. The tool is fine. The way it's used tells people they don't matter.

Being lied to about who you're talking to

This is the one that does real damage. A caller spends ninety seconds explaining a leak under the sink in detail, in the voice you use for another person, then realises halfway through that nobody is there. The embarrassment turns into resentment fast. People don't mind a machine. They mind feeling daft.

Sheer volume

AI is now in the search results, the email client, the supermarket app and the TV advert. The 19% excitement figure isn't a verdict on any one product. It's the sound of people wanting a break. If you put an AI in front of your customers in 2026, you are arriving into that mood whether you like it or not.

What this means if you run a small business

Here's where I'd push back on how the stat usually gets used. "85% prefer a human" is true and also not quite the question a plumber or a salon owner is facing.

The survey compares AI to a person. Your caller at 7:40pm on a Tuesday isn't choosing between AI and a person. They're choosing between an AI and voicemail — or an AI and eleven rings and nothing. That's a different comparison, and the ranking may well flip. Most people would rather tell a machine about a burst pipe and get a call back within the hour than leave a message into a void.

The other thing worth noticing: the survey asks about "customer service". For most people that phrase means a bank, a telecoms provider, an airline — big companies, complaints, hold music. It's a fair bet that a short call to book a haircut, confirm you're free Thursday, or take a name and number for a boiler that's packed in, doesn't feel like the same event. Nobody is emotionally invested in giving their postcode.

The gap between frustration and hanging up

59% report frustration. 31% say they'd hang up. That gap is roughly the practical space you're working in — a large group of people who are irritated by AI in general but will still go through with the call if it gets them what they need.

So the goal isn't to win over the 5% who prefer AI. It's to not lose the middle group. And the middle group's tolerance is set almost entirely by whether the call is useful. Frustration doesn't come from the voice being synthetic. It comes from the interaction being a dead end.

What actually earns trust back

None of this is complicated, but it does require you to give up the fantasy that nobody will notice.

1. Say it's an AI, early. In the greeting, not after the caller has told you their life story. "Hi, you're through to Meadow Plumbing — I'm the AI assistant, I can take your details and get Dave to ring you back." People are far more forgiving of a machine that admits it's a machine. The ones who'd rather wait for a human can decide immediately, before they've wasted their own time. We went through greeting wording in what callers actually hear when an AI voice answers.
2. Give a real route to a person, and make it fast. Not buried, not conditional. If someone says "can I speak to a human", that should work the first time they ask — not the fifth. The people typing 'agent' over and over are telling you exactly what the failure mode looks like.
3. Be specific about what happens next. "Dave will call you back this evening" beats "someone will be in touch" by a mile. A vague promise sounds like the brush-off people have learned to expect.
4. Then actually do it. The trust is won or lost at the callback, not on the call. An AI that takes a perfect message and then nobody rings back is worse than voicemail, because it raised expectations first.
5. Keep it short. Nobody wants a conversation with your phone system. Name, number, what's wrong, when they're free. Done in under a minute is a good outcome, not a failure to engage.
6. Let the caller correct it. If the AI mishears a postcode and there's no way to fix it, the caller is trapped in exactly the IVR experience they've learned to hate.

The common thread: the things that annoy people about AI are mostly design choices, not properties of the technology. Hidden identity, no escape hatch, no follow-through. You can avoid all three.

Where an AI on the phone isn't the right answer

There are businesses where I'd tell you not to bother, and it's worth being straight about them.

If your customers are mostly elderly or vulnerable. Domiciliary care, some healthcare, bereavement services, anything where the person calling may be distressed or confused. The hang-up risk isn't 31% there, and the cost of getting it wrong isn't a lost booking.

If the call itself is the sale. Some work is won by talking — high-value consultative jobs where the customer is buying your judgement and wants to hear it on the phone. A message-taker in the middle of that is a speed bump. You'd rather ring back late than intercept badly.

If you answer almost everything already. If you've got a receptionist and calls rarely go unanswered, you're adding friction for no gain. The case for AI rests on the calls you currently miss, not the ones you handle.**

If your callers need answers, not messages. Complex account queries, technical support, anything where taking a name and number doesn't advance the problem. A comparison of the options — including human answering services — is in AI receptionist vs answering service vs voicemail.

If you'd be tempted to hide it. If your instinct is to give the AI a human name and hope nobody clocks it, don't do it at all. That's the route that produces the resentment in the survey. Somebody will work it out, and they'll tell people.

The trend probably isn't going to reverse

It's tempting to assume this softens as the technology improves. Maybe. But the frustration numbers rose over a period when voice AI got noticeably better, which suggests quality isn't the only driver. Fatigue doesn't respond to better fatigue-causing things.

What that means practically: plan for a world where a meaningful slice of your customers will always prefer a person, and build for them rather than around them. That's not a reason to leave the phone ringing out. It's a reason to be honest about what's answering it, and to make sure the human is never more than one sentence away.

The businesses that come out of this well won't be the ones with the most convincing synthetic voice. They'll be the ones where the caller finishes the call knowing exactly what happens next — and then it happens.