Are customers really turning against AI customer service?

Trust is falling where AI is the main channel. A 6,000-person study across the UK, US and Canada found 57% would trust a business less if it predominantly used AI for support, up from 53% six months earlier, and 69% describe chatbots as efficient but emotionless. Support for well-implemented automation remains, with 54% of UK consumers saying good chatbot service matters.

Should a small business avoid chatbots altogether?

No, but placement matters more than the technology. Automation that customers never see, such as routing, prioritisation and drafting replies for a person to send, carries little trust risk. Automation placed between a frustrated customer and a human is where the damage happens.

What should never be handled by an AI agent?

Complaints, refunds, disputes and anything where money or judgement is involved. Those interactions carry emotion, and the research consistently shows customers want a person at exactly those moments. Capture the detail automatically by all means, then route it to someone who can decide.

How do I keep automation without losing trust?

Give every automated interaction a visible one-step route to a human, and make that route commit to a time rather than disappearing outside office hours. A bot that says when a person will reply preserves trust. A bot that loops or hides the contact route costs more than having no bot at all.

Customers Are Turning Against AI Support. That Is Not An Argument For Doing Nothing

Consumer trust in AI customer service is going backwards. A 6,000-person study comparing responses from October 2025 with April 2026 across the UK, US and Canada found declining trust, rising frustration and a growing preference for speaking to a human. In that research, 57% said their trust in a business would fall if it predominantly used AI for support, up from 53% six months earlier.

That is a real finding and it deserves to be taken seriously by anyone selling automation, including us. It is also routinely misread as "do not automate support". The data says something narrower and more useful.

What the research actually says

Three numbers matter more than the headline:

69% describe AI chatbots as efficient but emotionless, missing tone, urgency and frustration.
54% of UK consumers still say good chatbot service is important to them.
Trustpilot research found that poor AI experiences put around 6% of the UK e-commerce market at risk, with chatbots the biggest offender and 40% of users reporting a bad experience.

Read together, those are not the numbers of a public that hates automation. They are the numbers of a public that has been handed automation at the exact moment it was least wanted, with no way out.

The load-bearing word in the 57% figure is "predominantly". A business whose support is mostly a bot loses trust. That is a different claim from "any automation in support loses trust".

Where automation helps and where it costs you

The distinction that holds up is whether the customer can see it, and whether they are frustrated when they meet it.

Task	Customer sees it	Risk
Routing and prioritising an incoming ticket	No	Low
Acknowledging an enquiry with a realistic time to reply	Yes, briefly	Low
Drafting a reply for a person to check and send	No	Low
Proactive updates on an order or job	Yes	Low, usually welcomed
Answering a stable factual question such as opening hours	Yes	Moderate
Handling a complaint, refund or anything with money attached	Yes	High
Standing between an angry customer and a person	Yes	Very high

The bottom two rows are where the trust damage in that research comes from. They are also, unhelpfully, the rows that get automated first, because they are the ones consuming the most staff time.

The rule worth adopting

Every automated interaction needs a visible, one-step exit to a human, and that exit must work outside office hours in the sense that it captures the request and commits to a time.

A bot that says "I cannot help with that, here is how to reach someone, and they will reply by 10am tomorrow" preserves trust. A bot that loops, deflects, or hides the contact route destroys it, and it destroys it faster than having no bot at all because the customer now knows the business chose this.

That is the practical difference between using automation to reduce your workload and using it to reduce your accessibility. Customers can tell which one they are in within about two exchanges.

We have argued the other side, and both are true

We have written before about AI agents triaging support tickets, and that position has not changed. Triage is invisible to the customer, it removes a genuinely mechanical task, and it makes the human reply faster rather than replacing it.

What this research adds is the boundary. Automation that shortens the path to a person is an improvement. Automation that lengthens it is a cost you will pay in trust, and the bill arrives late enough that most businesses never connect it to the decision.

How to check which side you are on

Four measurements, none of which need new software:

Count the proportion of automated conversations that end with the customer reaching a person when they asked to. If you cannot measure it, that is the finding.
Time how long it takes a determined customer to reach a human from your homepage. Anything over two steps is a problem.
Read fifty recent automated conversations end to end. Not a sample of the good ones.
Ask in your next post-resolution message whether the customer felt heard. It is a soft metric that moves before your review scores do.

If the answers are uncomfortable, the fix is usually not ripping the automation out. It is moving it one step earlier in the process, to the part the customer never sees.