Do customers want AI or a human? What the 2026 surveys actually say

Search for whether customers want AI in customer service and you will find confident statistics pointing both ways. Eight in ten are willing to use it. Two-thirds would rather have a person. Nearly nine in ten insist on a route to a human. All three come from credible 2026 surveys, and all three are true at once.

This article sets the main studies side by side, notes who paid for each, and draws out what they mean if you are deciding how much of your customer contact to hand to AI.

The headline figures, side by side

A note on reliability. Gartner is an analyst firm and Ofcom a regulator. Five9 and Qualtrics sell customer service software, so read their framing with that in mind, though both used large samples and independent fieldwork. We have left out several smaller UK polls reporting that 84 to 90% prefer a person, because their methods are thin.

Willing to use it, and still preferring a person

There is no contradiction between being willing to use something and preferring something else. Most people are willing to use a self-checkout. Fewer would choose it for a trolley full of shopping with an age-restricted item in it.

The same pattern shows up by task. Ofcom's 2026 research found that telecoms customers are using AI to help manage their services, but that human contact is still valued, particularly for complex or sensitive issues. YouGov's study of British preferences found email and phone were the preferred channels for about a third of people each, and that preference for the phone rose steeply with age, from 16% of Gen Z to 50% of the oldest group.

So the realistic picture is a split by job, not by person. The same customer who is happy for an AI to confirm tomorrow's appointment wants a human when the boiler has failed in January.

The figure that matters: 26% against 55%

If you take one number from 2026, take this one. Five9 asked consumers how far they trusted AI to resolve their issue. With no human option, 26% did. With a human reachable, 55% did.

Nothing about the AI changed between those two answers. What changed was the customer's sense of risk. If the AI is the only door, every failure is a dead end. If there is a person behind it, trying the AI first costs nothing.

This is why "AI or human?" is the wrong question for anyone designing a service. Putting people within reach is what makes customers willing to use the AI at all. The same survey found 92% give the AI less than ten minutes before they want to escalate, and that the most common reason for giving up is being repeatedly misunderstood.

One bad experience is expensive

Gartner's September 2026 finding is the cautionary one: after a negative experience with a chatbot, only 27% of customers would try it again. In Five9's data, four in ten say they are less likely to use a company at all if it uses AI for service.

The bad experiences have a consistent shape. Citizens Advice reported that, among UK adults who could not reach a person, 49% felt stressed or frustrated, 44% said the problem took longer to resolve and 14% gave up on it altogether. Qualtrics found that half of consumers worry AI will stop them reaching a human.

Notice what these complaints are about. Hardly anyone objects to an AI that answers the question. They object to being stuck.

The handover is where it goes wrong

Even where a route to a person exists, it often fails at the join. Five9 found nearly all contact centre leaders believe their handoffs preserve context, yet 83% of consumers say they have to repeat themselves at least sometimes after being transferred from AI, and over a third say it happens often or always.

That gap, between what managers believe and what customers experience, is the single most fixable problem in AI customer service. The person taking over should see the whole conversation before they say a word. We set out how in AI-to-human handoff, done properly.

Where AI on its own does well

It would be dishonest to leave out the evidence that customers are sometimes happier with AI.

• Octopus Energy's AI-written replies to routine emails scored 76% satisfaction against 72% for human replies to similar queries.
• In a large field experiment on job interviews conducted by AI voice agents, 78% of applicants chose the AI interviewer when offered the choice, and outcomes were better.
• Around half of customers in Gartner's survey said interactions are easier when companies use generative AI.

What these have in common is a narrow, well-defined task, honest labelling, and a person available for anything else. Octopus sends complex and vulnerable cases straight to staff. That is the design the surveys are describing.

What regulators are saying

Public opinion is turning into rules. Ofcom said in September 2026 that it should be very clear to telecoms customers when they are dealing with generative AI, and that it would be concerned if AI became a gatekeeper to support. Citizens Advice is campaigning for a right to talk to a human at essential service providers. Spain's customer service law requires a person on request from 28 December 2026, and the European Commission's proposed Digital Fairness Act, published in November 2026, includes a right to request a human in online customer service.

None of that is yet a general legal duty in the UK. It is, though, the direction of travel, and a service built with a route to a person is already compliant with wherever it ends up.

What to do with this

1. Let AI go first on simple things. Customers are willing, and it is faster.
2. Make the way out obvious. Say early that a person is available, on every channel. It raises trust in the AI itself.
3. Send some topics straight to people. Complaints, vulnerability, money disputes and anything urgent.
4. Carry the context across. The customer should never have to say it twice.
5. Offer the customer's channel. A third prefer email, a third prefer the phone, and younger customers lean to messaging. See choosing between WhatsApp, SMS, email and a call.

The surveys do not say customers reject AI. They say customers reject being left alone with it. For the other side of the question, what the AI itself cannot yet do, see what AI agents can't do yet. And if you want to see AI and people sharing one inbox and one call queue, that is what Unibound is built for.

Questions people ask

Sources

1. Gartner, 87% of customers say companies using GenAI for customer service must provide access to a human agent (August 2026). gartner.com
2. Gartner, only 27% of customers would try a chatbot again after a negative experience (September 2026). gartner.com
3. Five9, Business Leaders CX Report and Consumer CX Report (2026). five9.com
4. Qualtrics, AI-powered customer service fails at four times the rate of other tasks (October 2025). qualtrics.com
5. Ofcom, consumers using AI to help manage telecoms services, but human contact still valued (September 2026). ofcom.org.uk
6. The Guardian, UK service providers urged to stop relying on chatbots (Citizens Advice, September 2026). theguardian.com
7. YouGov, a deep dive into British customer service preferences (April 2025). yougov.com
8. CMSWire, customers say the AI handoff is broken (Five9 data, 2026). cmswire.com
9. Octopus Energy, AI trial wins customer approval (July 2026). octopus.energy