OnClarity Insights
Will our customers accept talking to AI?
Asked by a technology leader at a contact centre outsourcing company.
6 min read
The short answer
Yes, when it fixes their problem the first time. Customers already use AI they choose, but give company bots few second chances, so acceptance follows outcomes, not a human-sounding voice. Make sure it fixes the issue, keeps a person one step away with the context passed on, and leaves nothing to come back about.
Is “will they notice it’s AI?” the right test?
It is where most teams start. Will the voice sound human enough? Will customers hang up the moment they realise? Should the bot have a name?
It is the wrong test. Customers notice quickly, and they should, because you should tell them at the start. What decides whether they accept it is what happens next. A customer who gets the refund, the new delivery date or the answer that settles it in two minutes does not care much who handled it. A customer who goes round in circles, gets a wrong answer or has to call back tomorrow will not accept it, however warm the voice sounds.
So change the question. Not “will our customers accept AI?” but “will they accept this AI, on this topic, as it works today?” You can answer that one with your own data.
What does the data say about customers and AI?
Customers are not against AI. They use it every day, on their own terms. Gartner found that customers are about three times more likely to use third-party GenAI tools than a company-provided chatbot to resolve a service issue, and that company chatbot use has not risen since 2022 (press release, 8 July 2026, survey of 3,566 B2B and B2C customers).
Many have already had AI fix something for them. In Shep Hyken’s 2026 State of Customer Service and CX study of 2,226 US adults, 50% said they had resolved a customer service issue with AI without needing a person.
The same study shows why the rest hold back: 53% said they had received wrong or incorrect information from an AI self-service bot.
And customers want a way out. In Gartner’s survey, 87% of customers said it is essential for companies using GenAI to offer an option to reach a human agent (press release, 4 August 2026).
Read together, the pattern is plain. Customers accept AI that works and avoid AI that wastes their time. The gap is not taste. It is outcomes.
Source: Shep Hyken, 2026 State of Customer Service and CX, online surveys of 2,226 US adults, January 2026.
What are the three conditions customers accept?
1. It fixes the problem, not just answers it
Good looks like this: the AI finishes the job (changes the booking, issues the credit, books the engineer) or gives the answer that settles the question, on topics it has been tested on. A polite reply and a link to a help article is not a fix.
You do not have to cover everything on day one. Start with the one or two topics where the AI can finish the job, prove it there, then widen.
Check it on your own data: take the conversations the AI marked as resolved and see whether the outcome actually happened in your systems. For keeping answers accurate, see how to stop an AI agent from making things up.
2. A person is one step away, with the context
Not everything is solved by AI, and customers know it. Good looks like this: the customer can ask for a person at any point and gets one, without telling the story twice. The AI passes on the conversation, what it tried and what it already knows. A customer who has to explain the problem again has paid for both conversations, and so have you.
The fear of being trapped is what makes customers refuse. In Hyken’s study, 71% of customers felt that companies using AI make it more difficult to reach a human. A visible exit takes that fear away.
Check it: count the steps from “I want a person” to a person, and read 20 conversations where a customer asked for one. More on the bot handoff.
3. Nothing to come back about
Good looks like this: the customer does not contact you again about the same thing, on any channel, in the days that follow. A conversation that ends as “resolved” and comes back as a phone call the next morning was not accepted. The customer just found another door.
Check it: match customers across channels and count how many contact you again about the same issue within 7 days of an AI conversation. Compare it with the same topic handled by people.
How do you measure whether customers accept it?
By what customers do next, not by how many conversations the bot handles. Four signals, from your own data:
Repeat contact within 7 days about the same issue, after AI and after people, by topic.
Requests for a person, and where in the conversation they happen.
Drop-offs: customers who leave mid-conversation and turn up on another channel.
Satisfaction, read second. A survey after a chat hears from the customers who stayed, not from the ones who gave up.
If repeats are at or below your human level and requests for a person are falling, customers are accepting it. Containment alone will not tell you, because a customer who gives up looks contained.
Should you tell customers it’s AI?
Yes, always, in the first message. It is honest and it sets expectations. But do not expect disclosure to buy acceptance. Saying “I’m an AI assistant” does not make a wrong answer acceptable, and a human-sounding voice does not either. Customers accept the result.
The honest trade-off
Some moments belong with a person first, whatever the AI can do: bereavement, suspected fraud, formal complaints and customers in vulnerable circumstances. Acceptance also varies. In Hyken’s study, older customers had resolved far fewer issues with AI than younger ones, and sector and channel matter too. And a good exit to a person costs money: more contacts reach your team and your containment rate goes down. That is the price of being accepted. Weigh it against the repeat contacts, complaints and lost customers that a closed door creates.
What to do this week
Pull the 7-day repeat contact rate after AI conversations and after human ones, by topic.
Count the steps it takes a customer to reach a person from your AI.
Read 20 conversations where a customer asked for a human, and note why they asked.
List the topics and moments that go to a person first.
Check that your AI says it is AI in its first message.
Questions people ask
Do customers prefer talking to a human or AI?
It depends on whether the AI can solve their problem. Customers happily use AI tools they choose, and many have already resolved issues with AI alone. They prefer a person when the AI gets it wrong, goes in circles or blocks the way out. Give them AI that fixes simple issues and a person one step away.
Do customers like chatbots?
Customers like chatbots that finish the job and dislike ones that only point to help articles or loop. Company chatbots have a trust problem because many customers have had wrong answers from them. Start with a few topics the bot can fully resolve, make a person easy to reach, and measure repeat contacts, not conversations handled.
Why don’t people trust AI customer service?
Mostly because of experience: wrong answers, no clear way to reach a person, and having to repeat themselves once they do. Trust comes back when the AI fixes the issue, says it is AI, hands over with the full context when it cannot help, and customers do not have to come back about the same problem.
Sources
Gartner, press release “Gartner Survey Finds Customers Are 3x More Likely to Use Third-Party GenAI Than Company-Provided Chatbots for Customer Service”, 8 July 2026 (survey of 3,566 B2B and B2C customers, February and March 2026).
Gartner, press release “Gartner Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent”, 4 August 2026 (same survey).
Shep Hyken, The State of Customer Service and CX 2026, two online surveys of 2,226 US adults aged 18 to 70, conducted 5 to 9 January 2026.
