Today, many organizations across industries are deploying increasingly human-like AI to answer questions, troubleshoot problems, and handle complaints.
Commonwealth Bank of Australia recently revealed that its AI customer-service platform handles more than 2 million conversations each month through voice and messaging channels.
Australia Post also said that 55% of customer inquiries received support from self-service AI in 2025.
The logic behind this trend seems simple: when something goes wrong, consumers are accustomed to seeking help from a human. If AI can communicate in a more natural way, having it take on some of the responsibilities traditionally handled by human customer-service agents seems logical. But research shows that the reality is far more complex.
AI customer-service systems are increasingly adopting voice interaction and conversational design, based on the assumption that consumers prefer interactions resembling communication between people.
The Australian Contact Centre Industry Benchmark Report shows that when consumers are told their problem can be resolved, 27% prefer the phone, while 12% choose web chat.
But this does not mean that making AI more human-like will necessarily improve the customer experience. A recent survey of 1,031 Australians found that 78% had used automated customer-service systems, but only 39% said they were satisfied with them.
A study published in the European Journal of Marketing conducted several experiments to examine how consumers respond to different forms of AI customer service after experiencing a service problem.
One study examined the results of interacting with a text-based chatbot rather than a voice bot after consumers encountered a service problem. It found that, in some cases, less human-like AI could actually produce more positive customer feedback.
In addition, participants who interacted with text-based chatbots showed a stronger sense of relief than those who interacted with voice bots. This sense of relief was also associated with more positive evaluations of the overall service experience.
This may be because typing is a slower and more deliberate form of communication than speaking. When consumers have to enter text using a keyboard, they have more time to shift their attention and allow some of the negative emotions caused by the problem itself to fade.
In other words, when consumers are already irritated, faster interactions that more closely resemble human conversation can sometimes take away something valuable: a little time to calm down.
However, creating some emotional distance is only one part of building effective AI customer service.
Emotional intelligence matters too
Another study, published in the Journal of Business Research, examined how consumers respond when AI customer service can recognize and respond to their emotions.
The study had participants interact with two types of bots. One explicitly recognized the consumer's anger and said, “I understand that you are very angry right now.” The other used a more general expression: “I understand that you are feeling terrible right now.”
The results showed that consumers responded more positively when the bot accurately identified their emotional state. More precise emotion recognition improved consumers' assessment of the AI customer service system's ability to understand emotions and increased customer satisfaction.
Taken together, these findings show that making AI more human-like does not necessarily mean it will be better at understanding and responding to human emotions.
Sometimes, the best customer experience comes from creating an appropriate emotional distance and giving consumers a little time to process their frustration. In other situations, accurately recognizing the consumer's current emotional state and responding appropriately is what determines whether the experience is good or bad.
AI customer service does not need to imitate humans at all costs
As AI takes on more work in customer service, the key question is not how human it sounds. Businesses need to carefully consider what consumers actually need at each stage of the customer-service process.
Future AI customer-service systems may not need to aim to imitate humans as realistically as possible. Truly effective AI customer service should be able to determine when consumers need understanding and reassurance, when they need some space and time to calm down, and how to adjust the interaction accordingly.