Your brand hired a new employee this year. It works 24 hours a day, never asks for a raise, never books a vacation, and has been trained on your entire help center. It is also, by any reasonable performance review, the worst employee you have ever had. It cannot solve the problem the customer actually has, it refuses to admit when it’s stuck, and it has been instructed to apologize so warmly and so often that talking to it feels like being slowly smothered by a throw pillow that majored in customer empathy. Meet the support chatbot — generative AI’s most enthusiastically deployed and least examined hire of the decade.
The Bot That Cannot Say “I Don’t Know”
The defining trait of a bad support bot is not that it lacks information. It’s that it cannot tell you it lacks information. A human agent, faced with a question outside their knowledge, says “let me check with someone.” The bot, faced with the same question, generates a fluent, confident, structurally perfect answer that happens to be wrong — because that is what these models are built to do. They are not built to be correct. They are built to be plausible. And plausibility, deployed at the exact moment a customer is already frustrated, is not a feature. It’s a trap with a friendly avatar.
You’ve lived this. You arrive at the chat already annoyed — something broke, something charged you twice, something didn’t arrive. The bot greets you with relentless cheer. You explain. It returns a paragraph that almost addresses your issue, links you to the help article you already read, and asks “Did that solve your problem?” with two buttons, neither of which is “No, and now I’m angrier.” You loop. You type “agent.” It asks you to rephrase. You type “AGENT.” It offers a survey. This is not customer service. This is a containment strategy wearing customer service’s clothes, and customers can feel the difference, the same way they could always feel the gap between “authentic” branding and the calculated thing underneath it.
It Was Never About Helping You
Let’s be honest about why the bot exists, because the brand never will. The chatbot was not deployed to help customers faster. If speed and resolution were the goal, the budget would have gone to hiring and training more humans, which works and is boring and doesn’t appear in a quarterly innovation update. The bot was deployed to deflect — to reduce the number of conversations that reach a paid human, measured in a metric called “deflection rate” that is celebrated internally precisely because it counts the customers who gave up.
Read that again. The headline success metric for most support AI is the number of people who wanted help, didn’t get it, and left. Dressed in a dashboard, “customer abandoned the chat in frustration” becomes “issue resolved without agent escalation.” It’s one of the purest ego KPIs ever invented: a number that rises as customer experience falls, presented to leadership as a triumph. The bot isn’t failing at its job. Its job was always to make you go away cheaply, and at that job it is brilliant.
The Uncanny Valley of “How Can I Help?”
There’s a specific dread in talking to a machine engineered to sound human while being institutionally incapable of human judgment. It uses your name. It says “I completely understand how frustrating that must be.” It deploys empathy as a UI element. And the warmth makes it worse, not better, because the warmth is a promise the system can’t keep. A blunt error message at least respects you enough to be a machine. A chatbot that performs caring while delivering nothing is running the same play as a brand optimizing its language for algorithms instead of people — fluent on the surface, hollow underneath, and increasingly obvious to anyone paying attention.
This is the part the technology vendors don’t price in. Every interaction with a bad bot is a small deposit of resentment against your brand, and customers are keeping the ledger even when you aren’t. They will tell you about it. They will tell each other about it, in screenshots, with captions, in the genre of content that travels furthest precisely because it can’t be planned — the same unplannable virality brands chase in their campaigns and accidentally manufacture at their support desks. The funniest thing your brand publishes this year may be a transcript of your own chatbot, posted by a customer, with no edits required.
The Bot That’s Actually Good (And Why It’s Rare)
None of this means AI has no place in support. A well-built system is genuinely useful — when it’s designed to assist rather than deflect. The good version knows the boundary of its own competence and hands off the instant it hits it, with full context, to a human who doesn’t make the customer start over. It handles the genuinely simple, repetitive queries that humans hate, freeing those humans for the hard, emotional, judgment-heavy cases where they add the most value. It treats “escalate to a person” as a success, not a failure.
That version is rare for a simple reason: it costs more, not less. It requires keeping the human team, integrating systems properly, and choosing customer outcomes over deflection metrics. In other words, it requires the brand to deploy AI as an investment in service rather than a reduction in headcount, and most deployments are very transparently the latter wearing the former’s badge. The technology isn’t the problem. The brief is. It always is. We’ve built an entire business on that one observation.
You Are Training Your Customers to Hate You
Here’s the trend nobody’s putting on the conference slide: as bad bots proliferate, customers are learning a new default behavior — assume the brand doesn’t want to talk to you, and route around it. They go straight to “agent,” straight to social, straight to the chargeback, straight to the competitor with a phone number. Every brand that deploys a deflection bot is, collectively, teaching the entire market that contacting a company is a hostile, low-trust act. That’s a shared resource being quietly strip-mined, and the bill comes due in churn that no deflection dashboard will ever connect back to its cause.
The brands that win the next few years won’t be the ones with the most advanced chatbot. They’ll be the ones brave enough to make talking to a human easy again, and to treat that as the competitive advantage it has quietly become. Everyone else will have a 24/7 employee who works for free, never complains, and is slowly, fluently, empathetically dismantling the brand one warmly-worded non-answer at a time.
We make things for the humans still on the other end of that chat window — the support reps, the marketers, the creatives watching their company outsource its voice to a machine that can’t say “I don’t know.” The KPI Shark for the deflection-rate slide, Fuck The Brief for every “deploy AI” mandate that skipped the part about why, and a full shop of armor for anyone who still believes a conversation should help. Talk to a human. Start with the shop. No bot will greet you. That’s the point.
