Chatbot (Conversational AI)
GLOSSARY
Definition
A chatbot is software that conducts text or voice conversations with users, typically on a website, messaging app, or social platform. Rule-based chatbots follow decision trees. AI-powered chatbots use large language models to generate responses and handle questions they weren't explicitly programmed for.
Why it matters
The first wave of chatbots (2016-2019) was mostly disappointing, glorified FAQ menus that frustrated users and got turned off. The LLM wave changed the economics. A chatbot that can actually understand a question and pull the right answer from your knowledge base is useful in a way the old ones weren't. The remaining problem is trust: customers don't know if they're talking to a bot, and when the bot confidently gives a wrong answer, the brand takes the hit.
How it works
A chatbot is software that talks to visitors through text, following rules, a model, or both. Classic bot platforms route on decision trees; whenever a visitor picks an option, the flow advances. LLM-based chatbots generate their own replies and can hold open-ended conversation. In practice most products blend the two: rules handle structure, the model handles phrasing. The difference shows when the visitor asks something unexpected.
Practical uses
Chatbots handle the repetitive front line: answering FAQ questions, collecting contact details, routing to the right team, and taking basic actions like checking an order. Good ones escalate cleanly to humans when confidence drops. The metrics that matter are deflection rate, containment, and customer satisfaction after the chat. A bot that deflects 60% of simple queries pays for itself quickly if the handoff is smooth.
How to choose
Match the bot to the conversation you really get. Rule-based platforms are enough for consistent product questions and far easier to maintain. LLM platforms handle messy phrasing but need guardrails, a knowledge base, and budget for mistakes. For regulated or high-stakes products, escalation rules and human review before automated actions are non-negotiable. Start with the top ten questions, not a general-purpose assistant.
Common mistakes
The biggest failure is deploying an LLM bot without boundaries, so it confidently answers outside its knowledge and damages trust. The second is measuring chats handled instead of problems solved, which rewards bots that stall visitors. The third is neglect: bot flows rot as products change, and a stale answer quietly becomes the most-seen lie on your site. Maintenance is part of the product.
What changed with AI
LLM chatbots made the category credible for open dialogue, and agent platforms now let a bot take real actions like updating a ticket or starting a refund. The governance question is familiar: what can the bot do without approval, and what is the audit trail. For support specifically, the winning pattern is a bot that understands freely and defers on anything costly or irreversible.