AI customer service is becoming easier to find, but a survey of consumers suggests that the decisive feature may be the option to speak to a human. In research released in October 2026, people were much more willing to trust an AI interaction when a person remained within reach. The finding challenges the assumption that the best automated support system is simply the one that answers the most questions without ever transferring a customer.
The 2026 Consumer CX Report announced by Five9 draws on an online survey of 3,000 adults in the United States, United Kingdom and Germany. Five9, which sells customer-service technology, commissioned the research with Hanover Research. Its commercial interest is relevant, and the figures describe reported attitudes rather than observed behaviour across every customer-service system.
Why AI customer service still needs a human option
The headline finding is striking. According to Five9, the share of respondents who trusted AI to resolve their issue rose from 26% when no person was available to 55% when a human option was visible. The comparison does not prove that merely adding a button causes that increase in every context. It suggests that the prospect of being trapped with a machine may be a significant part of people’s hesitation.
Think about a straightforward request, such as checking the progress of a delivery. A chatbot that can retrieve the correct information quickly may be welcome. But consider an expensive item that has been marked as delivered when it has not arrived. The customer may need to explain a complicated situation, request an exception or challenge an earlier answer. In that setting, knowing a person can take over may change how comfortable the customer feels using the automated first step.
The difference is not necessarily a rejection of AI. People may appreciate a convenient assistant while remaining wary of being forced to accept an incorrect response. A visible handover offers a form of control. It tells the customer the machine is one route through a service rather than the only authority available to make a decision.
There is a limit to how long people will keep trying
Five9 reports that 92% of respondents would give an AI system less than ten minutes before escalating to a person. That does not mean every chatbot conversation actually lasts ten minutes, or that all customers are impatient. It is a response to a survey question about willingness to persist. Nevertheless, it provides a useful warning about experiences that make users repeat their problem without progress.
The report also says that repeated misunderstanding was the leading reason for abandoning an AI interaction for 46% of respondents, and that roughly a third reported having to repeat information after escalation. A company can therefore frustrate customers twice: first through a bot that fails to grasp the problem, and then through a human handover that discards the conversation.
A good transfer needs context. If the customer has already explained their order number, circumstances and attempted solutions, the human agent should ideally receive enough accurate information to continue. That still requires careful treatment of personal data. Passing all chat history without limits is not automatically sensible, particularly when the original conversation contains irrelevant or sensitive details.
Five9’s accompanying analysis of the findings makes a related point about the gap between making AI available and earning trust. Only one in five respondents reported using AI ‘most of the time’ or ‘always’ when seeking customer service, even though many could see some value in it. Availability is not the same as voluntary preference.
What the country differences reveal
Attitudes were not identical across the three countries surveyed. Five9 says 52% of German respondents agreed that AI was useful in customer service, compared with 46% in the United Kingdom and 44% in the United States. It also reports that UK respondents placed particularly high importance on being told when they were talking to AI.
Those differences should not be turned into stereotypes about national character. The survey was conducted online using a recruited panel in April 2026, and its results reflect the wording, design and sample of that study. The right conclusion is that businesses operating in different markets should test how their own customers respond rather than importing a single universal assumption about trust.
The responses also varied by age. Younger participants were generally more open to seeing a benefit in automated support than some older respondents. That may reflect familiarity with conversational technology, different expectations or experiences of service. The study does not by itself prove which explanation is correct or show that any generation will always prefer AI.
Transparency is more than announcing that a bot is present
Five9 reports that 71% of respondents considered knowing whether they were speaking to AI very or extremely important. There is a simple reason to make that distinction visible: people may change what they share or how much they rely on a response if they understand whether a human is making the judgement.
A customer might be content to ask an automated assistant about shop opening hours, but may want a person to discuss a complex disputed charge. Some tasks involve empathy, discretion and exceptions to routine rules. Others involve private data that a customer might hesitate to disclose in a conversational system. Transparent identification helps people choose the right level of trust for the situation.
The obligation to disclose AI interactions also has regulatory dimensions, which LiveAIWire has covered in its guide to EU transparency rules for chatbots. Those rules are not identical to a company’s customer-experience preferences, but they reinforce the importance of making the nature of the interaction understandable rather than hiding it in confusing interface language.
Why companies still want the automation
Customer service involves many repetitive requests, including requests for routine information, account details or next steps. A well-designed assistant may answer some of them quickly and leave human staff more time for difficult cases. That is the business promise behind the technology. But it only works when the system has accurate information, understands its limits and knows when to stop.
One risk is optimising for the wrong target. A dashboard may reward a low transfer rate because each conversation handled by AI appears cheaper. If customers cannot get their problem resolved, the organisation might pay for repeated contacts, complaints, refunds or lost loyalty instead. Measuring successful resolution and customer effort can be more revealing than counting how often a bot prevented a handover.
The question was already visible before this survey. LiveAIWire previously asked whether AI call centres are becoming better or merely cheaper. The new figures add a more specific consumer preference: people may value AI convenience more when human help is not removed. That does not automatically settle the economics, but it challenges the idea that every human transfer represents a system failure.
A handover has to work in practice
An obvious ‘Speak to a person’ option is useful only if it leads to someone who can meaningfully help. A service that offers a button but repeatedly returns the customer to another automated menu has not solved the problem. Nor is a transfer successful when the human agent lacks the information or authority needed to resolve the issue. The experience needs to be designed from the customer’s point of view.
Some service failures begin long before a customer meets the bot. If records disagree, a parcel has been misrouted or a payment system has made an error, conversational fluency cannot repair the underlying process. An assistant can help collect the right information, but the organisation still needs someone with access and authority to correct the problem. A reliable handover therefore depends on operational design as much as on how politely the first automated response is phrased.
Businesses also need to be realistic about hours, staffing and support channels. A customer may accept an AI response at midnight if the system explains that a person will review the request in the morning. They may be less satisfied if the assistant pretends to offer immediate escalation when none is possible. Clear expectations can build trust more effectively than a vague promise of always-available help.
LiveAIWire has also explored the debate about customers being able to reach a human. The ethical and policy argument concerns meaningful choice, while Five9’s survey suggests there is also a practical business reason to offer that choice. Those are related but not identical propositions.
For customers, the simplest useful test remains whether a service can solve the issue without wasting time, explain its limits and provide a credible alternative when it cannot. For companies, the survey suggests the best deployment may not be the loudest or most automated one. AI can be a helpful first contact while the presence of a capable person remains part of what makes the whole service trustworthy.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.