What Is a Chatbot? From Decision Trees to Language Models

In plain English

The old generation followed decision trees: recognise a phrase, return a scripted reply, fail on anything unexpected. That is where the reputation came from.

Language models removed the scripting problem and left the design problem. A bot that cannot escalate, or that hides the contact route, is worse than no bot regardless of how well it understands the question.

What to know

Purpose-built

Scoped to a task, unlike a general assistant.

Now model-driven

Understands phrasing it was never given.

Needs grounding

Answers should come from your documentation, not from training knowledge.

Needs an exit

A visible, quick route to a human is not optional.

Why it matters

A grounded bot that answers the top twenty questions accurately and hands off cleanly is genuinely valuable. A bot that improvises answers about your refund policy is a liability, and the difference is grounding and escalation design.

Common mistakes

×No escalation path to a human.

×Answering from training knowledge rather than your own documentation.

×Deploying with no log of what it actually told people.

×Hiding contact details behind it to reduce support volume.

FAQs

Will a chatbot reduce support load?

Yes for repeated questions, if it is grounded in real documentation and can escalate.

What is the biggest risk?

Confidently stating a wrong policy. Grounding and logging are the mitigations.

WRITTEN BY TARIQ SALLAM

Marketing Consultant. Entrepreneur. Content Creator.

I'm a marketing consultant, entrepreneur and content creator. I help businesses grow through practical marketing, websites, SEO, content and AI.

More About Tariq →