Half your customers don't use chatbots. Keep a door that isn't one.

Tuesday evening, at your help page

Here's a composite, but you've probably met her. She's sixty-eight, she ordered a coat from your store for her granddaughter, and it's the wrong size. She wants to swap it before Christmas. She goes to your site, clicks Help, and the page is mostly a chat window with a cheerful greeting: "Hi! I'm your AI assistant. How can I help?"

She doesn't type anything. She scrolls down looking for an email address or a phone number and doesn't find one. She closes the tab. Tomorrow she'll ask her son to sort it out, or she'll call her bank about the charge, or she just won't order from you again.

Nothing on your dashboard records any of that. The bot didn't fail, because she never used it. The bot can be the fast door. It shouldn't be the only one.

What the numbers say

This isn't a hunch about older customers. In February 2026, Pew Research Center asked 5,119 US adults about AI. Fifty-one percent said they never use AI chatbots like ChatGPT, Gemini or Copilot. Forty-nine percent said they do.

The split by age is steep. Among adults 65 and older, 77% said they never use them. Among 50 to 64 year olds it was 58%. Even among 18 to 29 year olds, a third said never.

Pew also asked the people who don't use chatbots why. The most common reason was simply not being interested, at 83%. But the next two matter more for a business: 79% were concerned about how their personal information would be used, and 76% didn't trust chatbots to give accurate information. And most of them don't expect to change their minds soon. Two-thirds said they were not too likely or not at all likely to use one in the next year.

So if your help page, your ordering flow or your phone line now runs through an AI assistant only, roughly half of US adults are being asked to do something they don't do, for reasons that include not trusting it with their details or their answer.

Where the only door usually appears

We rarely see anyone decide to remove the other routes. It happens by accumulation. A few common places:

The help page. The chat widget gets bigger, the contact form moves down a page, then into a footer link, then away.

Order changes. Address changes, swaps and cancellations get routed through the assistant because it's faster to build there, and the account page loses its own buttons.

The phone line. An AI receptionist answers every call, and there's no clear way to reach a person. We wrote about calling your AI receptionist before a customer does for this reason.

Returns. The return starts with a chat, and the chat is the only way to get a label.

None of these is wrong on its own. The problem is when the bot becomes the only way to get a particular job done.

Design one door, not a maze

Keep the assistant. Many customers like it, and for simple questions it's often quicker than anything else. Just design around the people who won't use it.

Show the other route from the start. A plain line under the chat window works: "Prefer not to use chat? Email us or call, and here's when we answer." Don't make it something people only find after the bot fails.

Make both routes reach the same outcome. If the bot can swap a size, the account page or a short form should be able to as well, or a person should be able to do it the same day.

Say what it is. Label the assistant as AI. People who are worried about privacy deserve to know who, or what, they're telling their order number to.

Hand off with the conversation attached. When someone asks for a person, the person should see what was already said. Nothing makes a customer angrier than repeating everything. Our note on handing a stuck agent back to a person covers what that packet needs.

Let simple jobs skip the conversation. Order status, for example, works better as a page than a chat. See making "where is my order?" work without an account.

But the second door costs money

It does, and that's usually why it disappears. The assistant was meant to take load off a small team, and a phone line or an inbox brings some of it back. That's a fair concern, so be specific about it.

You don't have to staff a phone line all day. A callback request with a promised time works. So does an email address that's answered within one business day, if you say so on the page. What matters is that the route exists, that people can see it, and that it ends with the job done.

It also helps to look at what the second door is for. If most people who skip the bot want the same three things, like swapping a size, changing an address or getting a return label, build those as plain pages or short forms. They cost little to run, they don't need a conversation, and they work for both groups.

And count what you'd lose without it. Every customer who leaves your help page without asking anything is invisible in the bot's numbers. A support dashboard that says the assistant resolves most chats tells you nothing about the people who never opened one.

Try reaching yourself without the bot

Spend fifteen minutes this week trying to get three things done on your own site without touching the assistant: change an address, start a return, and ask a question about a product. Do it on a phone. Note where you got stuck and whether there was any human route at all.

Then check your phone line the same way. Can a caller reach a person, and do they know how?

If you're adding AI to support or ordering and want it to help the customers who like it without losing the ones who don't, that's the kind of AI applications and agents work we do. Talk to us.

Related work: Spin:Market.