The Trust Gap: Most Customers Say They Don't Want AI Support. They're Not Wrong.

If you only read the headline surveys, the verdict on AI customer service is in: people hate it. A widely cited figure has roughly two-thirds of consumers saying companies should not use AI in support at all. That number gets quoted in every board deck arguing against automation. It deserves to be taken seriously — and it deserves to be read next to the other number nobody quotes in the same breath.

Because in the same body of research, a majority of those same people say they would rather deal with a bot than a human when what they want is an immediate answer. Both findings are real. They come from serious researchers, not marketing departments. And the businesses that treat the first number as a reason to avoid AI — or the second as permission to automate everything — are both misreading what customers are actually telling them.

The contradiction isn’t a contradiction

Hold those first two numbers side by side. Gartner’s 2024 consumer research found that a large majority would prefer companies not use AI in service, while roughly half of the same population would pick a bot over a person when they want an answer right now. That looks like people contradicting themselves. It isn’t. It’s people describing two different experiences and using the same word — “AI” — for both.

When a customer says they don’t want AI support, they are almost never objecting to the technology in the abstract. They are remembering the last time a bot looped them through four menus, misunderstood a plain question, and then refused to connect them to a human. When the same customer reaches for a bot for “immediate service,” they are remembering the time one gave them a tracking number in three seconds at 11 p.m. The variable that flips the answer isn’t whether AI was involved. It’s whether the thing in front of them actually resolved their problem.

People aren’t rejecting AI. They’re rejecting being trapped by something that can’t help them and won’t let them leave.

Why the distrust is earned

Here is the part the pro-automation camp skips: the skepticism is rational. By independent estimates, only a small share of support issues — on the order of 14% by Lorikeet’s 2025 analysis — are actually resolved end-to-end by self-service today. If that’s the base rate a customer has learned to expect, then reaching for the bot first is, statistically, a bad bet most of the time. Their wariness is earned experience, not technophobia.

And the failure modes are specific. Customers don’t generically “dislike bots.” They dislike a short, repeatable list of things:

• The maze. No visible way to reach a human, or one buried behind three refusals to help.
• The amnesia. Repeating your account number, your problem, and your order history to the human you finally reach, because none of it traveled with you.
• The disguise. A bot pretending to be a person, which feels like a small con the moment you figure it out.
• The dead end.A confident, wrong answer delivered with no way to flag that it didn’t work.

Notice that not one of those is a complaint about response speed, availability, or even accuracy on routine questions — the things AI is genuinely good at. Every item on the list is a design failure around the edges of the conversation: the entrance, the exit, the honesty, and the handoff.

Resolution quality is the real axis

This is why the trust question can’t be answered at the level of “AI: yes or no.” The axis that actually predicts whether a customer walks away trusting you is resolution quality — did this interaction finish the job, and if it couldn’t, did it get me to someone who could without making me start over? A majority of consumers also tell researchers they wish companies would be more careful and more transparent about how they use AI in support (per Zendesk’s 2024 CX research), which is the same signal from a different angle: the objection is to carelessness, not to the tool.

What earning the trust actually looks like

The design principles fall straight out of the complaint list. Invert each failure mode and you get the specification:

• An escape hatch that always works.The customer can reach a human whenever they ask — first request, no maze, no three strikes. Counterintuitively, making the exit obvious makes people use it less, because they stop fighting the system out of fear they’re stuck.
• Context that travels.When the conversation moves to a person, the full transcript and the customer’s intent go with it. Nobody repeats themselves. This single thing erases the most common source of “AI ruined my support experience” stories.
• Honest disclosure.The customer knows they’re talking to an AI. No costume, no ambiguity. Transparency isn’t a compliance tax here — it’s the thing that makes the fast, competent answer feel trustworthy instead of like a trick.
• Escalation on the right signal. The genuinely emotional, complex, or high-stakes moment gets handed to a human early — not after the customer has already looped three times and gotten angry.

Do this and something quietly interesting happens: the customers who told a survey they “don’t want AI support” use yours, resolve their issue, and never file it under “bad AI experience” — because it wasn’t one. You didn’t win the argument about whether AI belongs in support. You made the argument irrelevant.

The honest caveat

None of this means AI should touch everything. The trust gap is widest, and least forgiving, exactly where the stakes are highest: licensed advice, a medical worry, a claim on the worst day of someone’s year, a bill dispute where the customer is already furious. Those aren’t resolution problems to be optimized — they’re moments that belong to a human, and the AI’s only job is to recognize them fast and route them warm. A platform that can’t tell the difference between “where’s my order” and “I think I’m being scammed” hasn’t earned trust. It’s just gambling with it.