A customer who has lost trust in your chatbot rarely tells you. They do not file a complaint about trust. They just stop giving the bot a fair chance, and start working around it. By the time the loss of trust reaches a number you watch, the damage is done.
This article is about catching it earlier: the specific signals that show customers have stopped trusting the bot, and where to find them.
Why lost trust stays hidden
Michael Howlett, who writes Customer Experience Decoded, named the feeling behind this: the AI ick, the recoil a customer feels when a machine is handling something they wanted a person for. The important thing about the ick, for measurement, is that it is silent. The customer does not announce it. They act on it.
The common belief is that a satisfaction score will catch falling trust. It will not, or not in time. A customer who no longer trusts the bot often stops answering its surveys altogether, so the score stays calm while the trust drains. Lost trust is a behaviour change, and you have to look at behaviour to see it.
The signals that show it
Four behaviour signals, all in data you already have, show customers routing around the bot.
The first is the immediate escape. Customers who reach the bot and ask for a human within the first message or two, before giving it a problem to solve. A rising share means customers have decided, in advance, that the bot is not worth trying. The second is the abandoned start: customers who open the bot and leave without finishing. The third is the channel shift: customers who used to use chat moving to phone or email for the same kinds of question, voting with their feet. The fourth is the repeat avoidance: a customer who had one bad bot interaction and never returns to that channel.
Each of these is a customer deciding the bot is not for them. Together, trended over time, they are a trust signal that moves before churn does.
Read the language, not just the counts
The counts tell you trust is falling. The transcripts tell you why. Customers who have lost trust use specific language early in a conversation: asking for a human before stating a problem, naming the bot dismissively, referencing a past failure. A sample of transcripts, read for that language, turns the trend into a cause you can act on.
Why early detection is the point
Trust, once lost, is slow and expensive to rebuild, and a customer who has quietly written off the bot is most of the way to writing off the channel. The value of these signals is timing: they move while the loss is still small enough to act on.
A satisfaction score tells you customers are unhappy after they have told you. These behaviour signals tell you customers have stopped trusting the bot while they are still your customers, which is the only point at which you can do something about it.