Tier-1 support
FAQ, order tracking, ticket status.
What is a chatbot? Rules, NLU, and LLMs
A chatbot is a system that converses with a user in text or voice to answer, guide, or perform an action. It may be rule-based, NLU-based, or powered by a large language model. The NIST AI RMF and OWASP LLM Top 10 stress that a public-facing bot must manage risk, hallucinations, and prompt injection.
In one sentence
A chatbot automates conversation to help, qualify, or route users.
Key points
- Three families: scripts/rules, classic NLU, generative LLMs.
- Success is resolution, support deflection, and satisfaction—not “wow”.
- Clear human escalation and content guardrails are required.
- Conversation logs are personal data to protect.
Term at a glance
- Chatbot
- Conversational agent · Dialogue bot
- English term
- Chatbot
- Domain
- Artificial intelligence
- Category
- Assistants
- Level
- Beginner to intermediate
What does “chatbot” mean exactly?
Early bots followed decision trees. NLU platforms (e.g. Dialogflow) added intents and entities. LLMs enable more flexible answers at the cost of harder control.
A chatbot is not “AI” by magic: without a fresh knowledge base, tools (APIs), and refusal policies, it invents or blocks.
For an SME, the winning case is often FAQ + booking + lead qualification, with handoff to a human.
How do you deploy a useful chatbot?
- 01
Pick priority intents
Top 10 real support/sales requests.
- 02
Connect sources and tools
Validated FAQ, CRM, calendar—not only generated text.
- 03
Secure
Control prompts, filter PII, log, limit sensitive actions.
- 04
Measure and improve
Resolution rate, abandonment, CSAT, review failed chats.
A concrete chatbot example
A dental clinic in Laval deploys a site bot: emergencies vs planned care, available slots, document reminders. Complex cases escalate to reception. In two months, 35% of after-hours bookings go through the bot, without raising triage errors thanks to strict symptom rules.
What are chatbots used for?
Sales qualification
Budget, need, urgency before an advisor.
Appointment booking
Slots and automatic reminders.
Internal assistance
HR, IT, procedures for employees.
Benefits and limits of chatbots
- 24/7 availability
- Deflects repetitive tickets
- Structured data capture
- Scalable first contact
- Hallucinations if LLM is poorly framed
- Frustration if escalation is opaque
- Knowledge maintenance cost
- Security risks (prompt injection, data leaks)
How does a rule-based bot differ from an LLM bot?
| Chatbot | Generative LLM chatbot | |
|---|---|---|
| Flexibility | Low outside scripts | High across phrasings |
| Control | Strong and predictable | Needs guardrails |
| Maintenance | Trees to update | Prompts, RAG, evals |
| Risk | Missing answers | Fabrications / leaks |
Why chatbots matter for Quebec SMEs
Support and leads do not wait for office hours. A well-bounded bot cuts repetitive load and captures after-hours demand—without pretending to replace experts on high-risk topics.
Frequently asked questions
Do we need an LLM for a chatbot?
No. Many FAQs are better with controlled flows + document search.
How do we avoid hallucinations?
Ground answers (RAG), ban out-of-scope topics, test with evaluation sets.
Does the bot replace agents?
It absorbs the repetitive; complex and relational cases stay human.
What about privacy (Law 25)?
Inform users, minimize data, secure logs, define retention.
Related terms
Sources and references
Want a bot that actually resolves requests? We frame intents, sources, and guardrails for your service.
Design your chatbot