- A Guardian reader survey on US customer service in 2026 found people describing support as "debilitating, depressing, and enraging"... and AI chatbots as endless doom loops that block real resolution.
- Most of the pain isn't new. Bad phone trees and runarounds existed before ChatGPT. AI can make it faster or worse depending on how you build it.
- Rule 1: Feed the bot real data. A support agent without onboarding is useless. A bot without your policies, edge cases, and escalation triggers is the same.
- Rule 2: Design human handoff from day one... visible escape hatch, auto-escalate after a few failed turns, or both.
- Rule 3: Aim for ~80% automation, not 100%. Hiccups are chances to show you care. That's part of your brand, not a bug to eliminate.
The Guardian didn't run a scientific poll. They opened a call for stories and hundreds of Americans wrote in about battles with big companies... telecom fraud claims, FedEx reroutes, printer hell, insurance phone trees. The headline stat that stuck: people really, really don't like AI customer service. About one in ten responses called out chatbots specifically as doom loops... massive time sinks that never get you to a human who can actually fix the thing.
If you're a small business owner eyeing an AI support widget because it looks cheap and modern, read that twice. The complaint isn't usually "it's impersonal." It's "it doesn't work for anything beyond check my balance or update my address."
Customer service was already broken
Jackson and I went through the Guardian piece on a recent Infacto Daily episode. A lot of the stories aren't uniquely AI problems.
A healthcare executive spent a full day trying to reach a human at AT&T over a $600 fraud charge. Someone else burned days rerouting a FedEx package across phone calls and outsourced support layers. A 35-year-old software engineer wrote that products "need an app to use" and that fixing anything means "countless time researching, arguing with reps." That's the annoyance economy in plain English... and researchers have quantified the waste at well over $165 billion a year in hidden time and hassle costs.
So no, AI didn't invent bad customer service. Companies have been training customers to give up for decades. AI just scales the loop.
The flip side is also true: when support is good, you barely notice. Amazon returns. Smooth checkout fixes. You don't rave about them because they never became a problem in your brain. When it's bad, you tell everyone. That asymmetry is an opportunity if you're small enough to still care.
When AI support actually works
Not every conversational interface is a trap.
I signed up for a HubSpot account recently for a project. I don't use HubSpot day to day, but their onboarding is fully conversational... no wall of form fields, no twelve-step wizard. It feels like ChatGPT: you tell it about your business, it asks follow-ups, it pre-selects company type from what you said. On the back end it's still filling out a form. The interface is what changed.
Was it less exhausting than a traditional signup? Yes. I'd rather text-style my way through setup than stare at a big empty form and bail. HubSpot has also been quick on the customer-service side with website chat that works more like search than a hostage situation. Different job than "my package is lost" or "someone frauded my bill," but proof that conversational UX can reduce friction when the scope is narrow and the data behind it is solid.
That's the pattern: narrow task + good data + clear exit beats "we bolted a bot on the contact page and called it innovation."
The doom loop (and why people hate it)
Readers in the Guardian survey used language you can't un-hear: "stupid, useless, brain-dead bots," "infuriating, exhausting, debilitating, depressing, enraging."
The mechanics are simple. Customer has a non-routine problem. Bot only handles routine intents. Bot restates the FAQ. Customer asks for a human. Bot offers another FAQ. Time passes. Problem unchanged. Trust gone.
Many respondents were in their 60s and 70s, which matters for how you design defaults. "Just go online" stopped being universal advice when phone trees became voice bots. If your audience skews older, assume lower patience for hidden gestures and assume they'll want a big obvious Talk to a person button.
And some problems AI won't fix soon. Printer support is its own circle of hell. If you're HP, you probably still need humans who've seen a jam error code at 2 a.m. For everyone else: know which tickets are bot-eligible and which aren't before you automate the front door.
Three rules before you turn on the bot
We turned the episode into a short list. Two hard rules and one mindset shift.
1. Data first
You wouldn't hire a barista who never went through orientation, then swap in a new barista for every single order. That's what a support bot looks like without your docs, policies, refund rules, shipping exceptions, and "when in doubt, escalate" examples.
Before you deploy:
- Feed it real FAQs, not marketing copy.
- Include edge cases your team already handles in Slack.
- Test with the angry emails you actually get, not the happy-path demo script.
If you're curating which tools to wire up and what to automate first, the AI tools checklist is a sane place to start... pick one lane, prove it, then expand.
2. Human handoff is a feature, not a failure
Build escalation into the product, not into the customer's frustration.
Practical options:
- Visible button: "Talk to a human" always on screen. No scavenger hunt.
- Auto-escalate: After two or three failed turns, route to a person (or at least offer the handoff without making them type "representative" seventeen times).
- AI-detected escalation: Train the bot to recognize fraud, billing disputes, safety issues, and repeated "that didn't help" signals... then stop looping.
Jackson's bias: let the human control the escape hatch even if the AI is smart. Mine: the AI should know when it's out of its depth and the button should exist anyway. Belt and suspenders.
Either way, "we'll add human support later" is how doom loops get shipped.
3. Expect ~80%, and plan for the other 20%
You don't want 100% automation. Part of what makes your business yours is how you show up when something breaks.
There's a culture story from high-end hospitality: staff who treat a messed-up reservation as a chance to demonstrate who they are. Not celebrating failure... celebrating the recovery. Customer service isn't only deflection. It's reputation under stress.
Boy Scout version: be prepared. When I was a kid my dad hunted, so I took a hunter safety course. You pack extra water and an emergency snack because the trail doesn't care about your plan. Your business solves problems for people. Problems will pop up. Build the support model like you're packing for the worst afternoon, not like sunny weather is guaranteed.
80% handled fast by AI can be a win if the 20% gets white-glove human treatment. "A little better than before" is a reasonable first milestone. Perfection on day one isn't.
What to do this week
If you already have a chat widget or you're shopping for one, run this audit:
- Pick three real tickets from last month (angry ones). Try to resolve them with the bot. Time how long until a human appears.
- Find the human button as a new customer would. If it takes more than five seconds, fix that before you tune prompts.
- List what the bot must never touch (fraud, medical billing, custom quotes, anything legal). Hard-code escalation for those intents.
AI should shrink wait times and handle the boring stuff. It shouldn't be a moat between you and someone who already trusts you with their money.
When you're ready to go deeper on prompts for everyday business tasks, we keep a free library here: infacto.digital/go/ai-prompt-library-yt.
Conclusion
The Guardian headline is emotional because the stories are real. But the lesson for small business isn't "never use AI for support." It's don't ship the loop.
Give the bot your data. Give the customer a human. Plan for the hiccup instead of pretending automation replaces caring. Do that and you're not competing with AT&T's phone tree. You're competing with the rare company people actually rave about... because you made the hard moment easy.