Kinsta Survey: Customers Want AI Support, But Not at the Cost of Human Help

There are plenty of things AI is already good at — summarizing information, explaining a complicated term, and answering the kind of support question that’s been asked a thousand times. For hosting providers specifically, that could mean helping someone find a billing setting or understand why their SSL certificate has not updated yet.

And yet, it looks like most customers don’t seem especially interested in an AI-only support experience.

A new survey from managed WordPress host Kinsta found that consumers still very much so prefer talking to people when they need customer support. Especially when the issue is complicated, urgent, or not being resolved the first time around.

Of the 1,008 consumers surveyed, nearly 87% said they prefer interacting with a human rather than AI for customer service. More than three-quarters said humans resolve problems more accurately, and 73% said humans resolve problems faster.

That might feel a little surprising — AI is supposed to be the faster option, right? Surprise, surprise: It turns out a near-instant response is not all that helpful when it doesn’t actually do anything.

For Roger Williams, Head of Partnerships at Kinsta, the answer is not to avoid AI altogether — just to make sure it stays in its lane.

“Every customer should have an immediate and clearly visible path to a human, regardless of the issue,” Williams told HostingAdvice. “Hosting environments are complex, and customers have different levels of technical experience. AI can answer common questions quickly, but it should never become a barrier between a customer and the help they need.”

At Kinsta, Williams said, human support remains available 24/7. The managed provider uses AI to understand the general nature of a request and help route it to the appropriate team. But customers can always choose human support. The system also looks for signs that the conversation is not going anywhere, including repeated questions or customers saying things like, “I don’t understand.”

AI Is Fast (Until It Isn’t)

People do not necessarily mind automation when it works; the problem is when a company uses AI as a substitute for actual support instead of a way to make support more accessible.

Kinsta’s survey showed that customers are already running into the limits of AI support: 68% said they have experienced AI failing to understand a complex issue, while 60% said they had received inaccurate information from an AI-powered support experience. More than half — 53% — said they frequently need to escalate an issue from AI to a human representative.

Williams said Kinsta’s AI can help customers understand common hosting concepts by surfacing relevant documentation in the context of their question. That can include DNS, SSL certificates, WordPress errors, migrations, email configuration, and performance. But there are clear situations where an automated answer cannot replace a support engineer.

“What it cannot currently do at Kinsta is make changes to a customer’s environment,” Williams said. “Questions that require account specific investigation, changes outside the managed interface, Nginx customization, or interpretation of specific APM results still require a support engineer.”

Yes, AI can explain why a DNS record matters… but it can’t always tell you what’s wrong with a live site or what fix is safe to make. And even when the information is technically correct, Williams said AI may not recognize when it’s failing to properly communicate with the person asking for help.

“AI can personalize documentation, but it may not recognize when an explanation is failing to connect with the customer,” Williams said. “Our support engineers are trained to notice that, adjust how they communicate, and take ownership of reaching a resolution. Human involvement is also critical whenever a situation falls outside the scope of our documentation.”

This may seem like a no-brainer, but it looks like when someone’s site is down, they don’t just want a list of possible reasons why. They want someone who understands the urgency and can investigate the issue.

Customers Know When They’re Talking to a Bot

Not only do customers get frustrated when they’re talking to a chatbot, but they know when they are, too. Even if they cannot always explain exactly why, about 45% of respondents said they can tell whether they are talking to a human or AI in a live chat, while another 32% said they can sometimes tell.

The biggest giveaways were tone or wording, followed by the speed of responses and repetitive answers. Incorrect answers were another common sign. For anyone who’s used an LLM, that all tracks, doesn’t it? Whether it’s a lack of typos or perfect grammar, there’s something…too clean about chatbot responses that people have started to recognize.

Williams said customers can also sense the difference between AI being used to help them and AI being used to keep them away from a person. This is a problem because the survey already suggests people are already skeptical about companies’ motivations. Nearly 80% of respondents said they believe companies use AI in customer service primarily to save money rather than improve service.

“I want to believe hosting providers are trying to balance responsible cost management with better customer service,” Williams said. “History suggests, however, that cost savings can eventually become the dominant motivation, often at the expense of the customer experience.”

There is nothing wrong with a hosting company wanting to operate more efficiently, Williams is quick to clarify; support is expensive, infrastructure is expensive, and every provider has to make decisions about where to invest. But making human support hard to find is a deliberate decision, especially when a customer is dealing with a website emergency that could affect revenue or their own clients.

“The problem begins when a company deliberately makes human support difficult to find,” Williams said. “Customers can see and feel when AI is being used to help them and when it is being used primarily to keep them away from a person.”

Your Documentation Is Only as Smart as Your AI

For providers that are planning to add more AI to their support workflows, Williams’ advice is pretty simple: “Do not use AI to conceal weaknesses in your existing support operation. Start with your documentation.”

Documentation should be accurate, complete, organized, understandable, and updated as the product changes.

AI can only be as useful as the information behind it, after all. If the docs are outdated or unclear, an AI assistant is just going to deliver the same bad answer to everyone over and over again.

Williams said providers should test their AI with common support questions. Even so, internal testing is not enough: They also need feedback from real customers to determine whether the answers are genuinely useful and understandable.

“We are not looking for a quick fix,” Williams said. “The goal is to use AI where it genuinely improves the customer experience while preserving the human support our customers value.”

So, yes — AI can be a useful third option alongside documentation and human support; it can give customers a fast, personalized answer, that is if the question is straightforward and the information is reliable. When more context, judgment, action, or empathy is required, people still want a person.

“The measure of support AI should not be how many human conversations it eliminates,” Williams said. “It should be whether customers receive better answers and achieve better outcomes.”