Customers distrust AI customer service bots because the bots have been measurably worse than humans on the contacts the customer most needs help with (complex, emotional, edge cases), and because customers have learned to recognise that being put on a bot often means being further from a resolution, not closer to one.
A customer calls a support line, hears "I'm here to help," realises they are talking to an AI, and feels their shoulders drop. They have been here before. They suspect they will be made to explain their problem twice, sent through a loop, and finally connected to a human who will ask them to explain it a third time. That suspicion is mostly correct, and it is why "talk to a human" is the most common first sentence customers say to bots.
What people in the field are saying
CX Decoded has a piece called "The AI ick factor," arguing that customer distrust of AI in service is not irrational. It is the rational response to repeated past experience of AI being worse than the human alternative.
Where does the distrust come from?
From experience. Customers have spent ten years with scripted chatbots that could not understand them, then five years with phone trees that pretended to be helpful, and increasingly with fluent AI agents that sound capable and still fail. The distrust is learned. New AI deployments inherit it whether they deserve to or not.
Why is it not going away with better AI?
Because the cost of getting it wrong is asymmetric. A customer who gets a good answer from an AI saves a few minutes. A customer who gets a bad answer, or has to explain themselves three times, loses an afternoon and feels disrespected. The bad case is so much worse than the good case is good that customers default to expecting the bad one.
What do customers actually want?
A clear option to reach a person. The most-cited improvement in customer satisfaction with AI contact centres is not better AI; it is a faster, cleaner route to a human when the customer asks for one. AI that fights the customer's request to escalate compounds the distrust.
What rebuilds trust?
Three things, in this order. First, the AI handles the contact and the customer actually gets what they came for. Second, when the AI cannot help, it hands over to a human cleanly, with context, and quickly. Third, the human visibly knows what the AI already tried. Repeat across enough contacts, and individual customers start to trust the system. It is slow. It is also the only way that works.
Related: the field note on the AI ick, detecting lost chatbot trust, and the glossary explainer on escalation.