Nothing is more infuriating: AI is now customer service, and it keeps sending you in circles instead of to a person.

There is nothing more infuriating than having a real problem and not being able to reach a real person.

This writer has lived it, again and again, with company after company. Different businesses, different problems, the same pattern every time.

You go to the company's help chat. An AI answers. It sends you a link to a help page. You've already read that page. It doesn't cover your problem. That's why you're asking for help.

So you push. You ask for a manager, someone higher up. You get what seems like the next level, but it's just another AI, and it sends you to the same page. Around and around.

Sometimes the links don't even go where they say they go. The page is titled one thing and covers another. Or the button says "contact us" and drops you right back into the same chat.

Then, if you keep pushing, an email arrives that looks like it's from a real person. It says your issue is being handled. It includes... the same link. And after that, nothing. No more replies, as if you've been quietly marked "difficult."

It happens at the very top, too. Even writing to a big tech CEO's office got this writer a reply promising a special team would solve it, followed by an email with the same help-page links, and then silence. And when replies may finally come from a real person, they can arrive after midnight and at 2 a.m., pointing back to the same help articles that didn't solve the problem in the first place.

This isn't one bad company. It's everywhere.

If your problem were simple, the help page would have solved it. You need a human because it isn't. And those are often the urgent ones: a locked account, a wrong charge, a service you depend on that stopped working.

Here's why this is happening, and what's being done about it.

A "doom loop": the government even has a name for it

When you get stuck going in circles with a chatbot, experts call it a doom loop. The U.S. Consumer Financial Protection Bureau warned about it back in 2023: "These 'doom loops' are often caused when a customer's issue falls outside the chatbot's limited capabilities" (CFPB).

In plain words: chatbots are good at easy questions. Your hard question is exactly the kind they can't handle, so they keep handing you the easy answers.

And they're everywhere. In 2022 alone, about 98 million Americans, roughly 37% of the country, used a bank's chatbot, the CFPB found.

It's often by design

Customer-service software companies measure something called "deflection." Zendesk, one of the biggest, defines it plainly: "Ticket deflection routes customers away from human support queues" (Zendesk).

Think of it like a store that counts it as a win every time a customer walks out without talking to an employee, whether or not they got what they came for. Zendesk itself warns that a high deflection number can hide problems that never got fixed and people who just gave up.

Most people don't want this

• 64% of customers would prefer companies didn't use AI for customer service, a Gartner survey of almost 6,000 people found. Their top worry: that AI will make it harder to reach a human. More than half said they'd consider switching to a competitor over it (Gartner, via Insurance-Canada).
• Other surveys, mostly by customer-service companies, find even bigger numbers, with large majorities saying they'd rather talk to a person (results vary a lot depending on how the question is asked).

Companies tried replacing people with AI. Some had to back up.

• Klarna, the buy-now-pay-later company, bragged in 2024 that its AI did the work of 700 customer-service agents. By 2025 it was hiring humans again. Its CEO admitted AI was cheaper but "lower quality," and promised "there will always be a human if you want" (Entrepreneur).
• Salesforce cut about 4,000 customer-support jobs in 2025, saying AI now handles much of the work (TechRadar).

When the AI support bot just makes things up

AI doesn't only loop you. Sometimes it invents answers, and companies have paid for it:

• Air Canada: its website chatbot told a grieving customer he could get a bereavement discount after his trip. That wasn't the policy. Air Canada argued the chatbot was responsible for its own words. The tribunal called that a "remarkable submission," said the bot "is still just a part of Air Canada's website," and made the airline pay (McCarthy Tétrault, on Moffatt v. Air Canada).
• Cursor: in 2025, the software company's AI support agent told users about a "one device per subscription" rule that didn't exist. Customers cancelled over a policy the bot made up (AI Incident Database).
• New York City: the city's own business-help chatbot told business owners they could take workers' tips and turn away tenants with housing vouchers, both illegal. In 2026, Mayor Zohran Mamdani shut it down, calling it "functionally unusable" (The Markup).
• DPD: a frustrated customer got the delivery company's chatbot to swear and call DPD "the worst delivery firm in the world." The company switched off the AI part (Yahoo News).

Even the AI companies do it

The companies building AI use it for their own support, too, and some say so openly. Anthropic, the maker of the Claude AI, publishes how its support works: you start with "Fin," which its help page calls "our AI support bot." Paying subscribers can have the bot pass their case to the human support team. Free users get the bot and help articles only, and there's no phone support (Anthropic Help Center).

That's a common setup across the industry: a bot first, help articles second, and a human only for some customers, only in writing, and only after the bot agrees to pass you along.

Help is coming: new laws

• California passed the Right to Human Customer Service Act, signed September 28, 2026. Starting January 1, 2027, big companies (over $500 million in revenue) must connect you to a real person, or schedule a callback, when you ask. A chatbot can answer first, but it can't pretend to be human and can't be the only option. Fines can reach $10,000 per violation (PYMNTS).
• New York City proposed a ban on AI customer-service "doom loops" on October 20, 2026. Businesses would have to tell you when you're dealing with an automated system and give you an easy way to reach a human. "Customer service cost-cutting can't come at New Yorkers' expense," the mayor said. A public hearing is set for November 19 (NYC Department of Consumer and Worker Protection).
• In Congress, a bipartisan bill, the Keep Call Centers in America Act, would require companies to tell you when you're talking to AI and transfer you to a human in the U.S. if you ask. It hasn't passed (Sen. Gallego).
• Europe: under the EU AI Act, starting August 2026, companies must tell people when they're talking to an AI, unless it's obvious (EU AI Act, Article 50).

How to reach a real person

1. Say or type "agent," "representative" or "human," over and over if you have to. On phone menus, try pressing 0, or "0#" a few times. AARP found that trick worked at 5 of 8 big companies it tested (AARP).
2. Pick "cancel service" or "new service" in the menu. Those lines are often answered by people, because companies don't want to lose you or miss a sale (AARP).
3. Say clearly: "This is not answered by your help pages. I need a person." Write down every link they send and why it didn't help, so you can show you've already tried.
4. Go public, politely. Post to the company's social media account. Public complaints often get a human faster.
5. For banks, loans and credit cards, file a complaint with the CFPB. Companies are expected to respond.
6. For other companies, file with your state attorney general's consumer office or the Better Business Bureau.
7. Save everything: dates, chat transcripts, emails, links. If the bot told you something wrong, the Air Canada case shows a company can be held to what its chatbot says.

The bottom line

AI can answer simple questions fast, and that's fine. But the people trying hardest to reach a human are exactly the people with problems a help page can't fix. Sending them around in circles isn't customer service. It's a wall. And when the bot gets it wrong, makes something up, or the "special team" email just resends the same link, there's nobody left to fix it.

You shouldn't need to be famous, or to tag a CEO, to talk to a person. Lawmakers are finally starting to agree.

A real person, when it matters

Support-O was built on the same idea: when something might be wrong, a real person should find out. For someone you love who lives alone, Support-O does a simple morning check-in, and if they don't answer, the people they chose get a text. More below.