Chatbot vendors use a lot of jargon, and it is easy to nod along without knowing what a term means. This AI chatbot glossary explains the words you will meet when choosing or running a website chatbot, in plain English, grouped by topic so you can find them quickly.
What Do the Core AI Terms Mean?
The core terms describe how a chatbot generates answers and why some of them are wrong. Understanding them helps you judge vendor claims.
- Large language model (LLM): an AI system trained on large amounts of text that predicts likely words to produce a reply. It does not look facts up by default.
- Hallucination: a confident answer that is not based on any real source. See why bots make up answers.
- Grounding: limiting answers to a defined set of source material, so the bot answers from your content rather than from guesses.
- Retrieval-augmented generation (RAG): a technique where the system retrieves relevant content before generating an answer. See what RAG is.
- Prompt: the instructions and question given to the model.
- Prompt injection: an attempt to trick a chatbot into ignoring its instructions. See prompt injection explained.
What Do the Knowledge Terms Mean?
Knowledge terms describe what you give the bot to learn from. The quality of these inputs sets the quality of the answers.
- Knowledge base: the collection of content a chatbot answers from, such as Q&A pairs and website pages.
- Q&A pair: one question and its approved answer. See how to write Q&A pairs.
- Unanswered question: a question the bot could not match to its knowledge. A list of these is a to-do list for new answers.
- Category: a label such as Pricing or Delivery that keeps the knowledge base organised.
- Starter question (quick question): a tappable question on the widget's first screen.
What Do the Handoff and Inbox Terms Mean?
Handoff terms describe how conversations move from the bot to people. They matter most when the bot does not know the answer.
- Handoff (escalation): passing a conversation from the bot to a human. See when to hand over to a human.
- Shared inbox: one place where a team sees and replies to conversations.
- Assignee, status, priority, tag: fields that show who owns a conversation, where it stands, how urgent it is and what it is about.
- Internal note: a comment visible to staff only.
- Saved reply: a stored response that a teammate can insert and edit.
- Lead capture: collecting a visitor's contact details, for example through a short form. See lead capture without annoying visitors.
What Do the Website Terms Mean?
Website terms describe how a chatbot is added and shown. They are simpler than they sound.
- Widget: the chat bubble and window that visitors see. See widget placement and timing.
- Script tag: a single line of HTML that loads the widget on your site.
- API key: a secret identifier that ties the widget to your account; it can be regenerated if exposed.
- Page-aware greeting: an opening message that changes depending on the page.
- Tenant (account) isolation: keeping one customer's data separate from another's.
What Do the Measurement Terms Mean?
Measurement terms describe how you judge a chatbot. Treat them with care, because some are easy to inflate.
- Deflection rate: the share of conversations handled without a human. See deflection rate statistics.
- CSAT (customer satisfaction): a rating of how satisfied a customer was. See conversation ratings.
- ROI: return on investment. See how to measure chatbot ROI.
- Time to first reply: how long a visitor waits for the first answer from a person.
Pro Tip: When a vendor uses a term you do not know, ask for a plain-language explanation and a concrete example. A good vendor will give one.
How Should You Use This Glossary When Comparing Vendors?
Use the glossary to turn vendor claims into questions: how is the bot grounded, what happens at handoff, how is deflection counted and what is only planned. Plain questions get plain answers.
| Vendor says | Ask |
|---|---|
| "Our AI never hallucinates." | How is it grounded, and what does it do when nothing matches? |
| "High deflection rate." | How is it measured, and how do you know customers were satisfied? |
| "Enterprise-grade security." | Which certifications do you hold today, and what is planned? |
| "Powered by the latest model." | What stops it answering outside my content? |
| "Integrates with everything." | Which integrations are live today? |
Which Terms Matter Most for a Small Business?
For most small businesses, grounding, handoff and knowledge base quality matter more than model names. These three decide whether the bot helps or harms.
- Grounding: does the bot stay within your content?
- Handoff: does a real person take over, with the visitor's details?
- Knowledge base quality: are your answers accurate, current and in customer wording?
If you are earlier in the journey, start with seven signs your website needs an AI chatbot and the small business guide to AI chatbots.
What Other Terms Might You Meet?
You may also meet terms such as agent, multimodal, vertical AI and Model Context Protocol. Each is explained in its own plain-English guide.
- AI agent: a system that can take actions, not just answer. See what agents mean for small business support.
- Multimodal: handling more than text, such as images or voice. See multimodal chatbots.
- Vertical AI: tools built for one industry. See vertical AI.
- Model Context Protocol (MCP): a standard for connecting AI tools to other systems. See MCP explained.
Which Terms Are Often Confused?
Three pairs of terms are often confused: chatbot and AI agent, grounding and fine-tuning, deflection and resolution. Knowing the difference prevents buying the wrong thing.
- Chatbot vs AI agent: a chatbot answers; an agent can also take actions. See the three-way comparison.
- Grounding vs fine-tuning: grounding limits answers to your content at answer time; fine-tuning changes the model itself.
- Deflection vs resolution: a conversation that never reached a human is not necessarily a solved problem.
How Do You Keep Learning?
Pick one new term a week from vendor emails and write a one-line definition in your own words. Within a month you will follow most conversations.
Related reading: a small business guide to AI chatbots, chatbot vs live chat vs AI agent, MCP explained and GEO, AEO, AIO and SXO explained.
Conclusion
Knowing the terms for AI behaviour, knowledge, handoff, website setup and measurement makes vendor claims easier to judge. Focus on grounding, handoff and honest measurement, and ask for plain-language answers to anything unclear.
Key takeaways
- An LLM predicts words; hallucination is a confident answer with no source.
- Grounding and RAG keep answers tied to your own content.
- A knowledge base is built from Q&A pairs and website content.
- Handoff, shared inbox and saved replies describe how people take over.
- Deflection rate, ratings and ROI need careful reading.
Quick Summary
An AI chatbot glossary for small business owners covers four groups of terms: AI terms such as LLM, hallucination, grounding and RAG; knowledge terms such as knowledge base and Q&A pair; handoff terms such as shared inbox and escalation; and measurement terms such as deflection rate and CSAT. Grounding, honest handoff and careful measurement matter most when choosing a chatbot.
Want to see grounded answers on your own site?
Start free, add your Q&A pairs and paste one script tag. No credit card required.

Founder & CEO
Writes about grounded AI chatbots for websites: how they work, how to measure them and what to check before you buy.
Topics
Get new posts by email
Guides on AI chatbots for websites, straight to your inbox.
