Why your chatbot frustrates people (and how to build one that doesn't)

We've all lived the scene. You land on a website with a specific question, a cheerful chat window pops up, you type your question, and the bot answers something completely unrelated. You try again. It fails again. You end up typing 'talk to a human' three times until it leaves you alone.

That frustration has a simple cause. Many chatbots get added thinking about saving the company work more than helping the person asking. You can tell straight away when a bot is designed to make you give up before reaching a human, and it breeds more anger than having nothing at all.

Today's models are far better than those of a few years ago, so the bot already understands what you tell it. The problem has moved elsewhere: now it understands the question perfectly and still can't do anything useful, because it isn't connected to your systems and knows nothing about the person asking. It answers nicely and solves nothing.

A bot that actually helps usually has two things. It knows who you are and can see your order and your history, so it answers your case instead of a generic paragraph. And it knows when to give up: the moment things get complicated, it hands you to a person without making you beg, ideally with all the context already written down so you don't repeat yourself.

That's why we almost always suggest starting small. A bot that genuinely handles the five most common questions is worth far more than one that tries to answer everything and gets half of it wrong. People forgive a bot that says 'a person can help you better with this'. They don't forgive one that pretends to know and sends them in circles.

If you're thinking of adding one, the honest test is simple. Are you doing it so your customers get sorted faster, or so they bother you less? That answer, the sincere one, decides whether your bot will help or join the long list of bots we all hate.