Guide

Live Chat Terms Explained

4 minute read · Updated August 14, 2026

Why the vocabulary is worth ten minutes

Live chat has accumulated a vocabulary that sounds precise and frequently is not. The same word means different things in two vendors' dashboards, and a comparison built on those numbers can be confidently wrong. Knowing what each term measures — and where the definitions differ — is most of what you need to read a dashboard or a sales deck critically.

The terms below are grouped the way you will meet them: the conversation itself, the people, the measurements, and the automation.

The conversation

Widget is the chat interface on your site as a whole. Launcher is the small button that opens it — the bubble in the corner. Vendors often use the two interchangeably, which matters when a contract mentions customising one and not the other.

Pre-chat form is the set of fields a visitor completes before the conversation starts, typically name and email. Offline message or offline form is what the widget shows when nobody is available, capturing a message for later reply.

Proactive invitation is a message the widget sends first, usually triggered by a page, a time on page or a visitor's behaviour, rather than waiting to be opened. Transcript is the stored record of a conversation. Tag or label is a category applied to a conversation so it can be counted later.

Canned response, sometimes called a macro or shortcut, is a saved reply an agent can insert rather than retyping. Co-browsing is watching or sharing a visitor's view of a page to help them through something, distinct from full screen sharing.

The people and the routing

Agent is a person who answers chats; seat or licence is the billing unit that lets one agent sign in, which is not always the same thing — some plans price by named agent and some by concurrent sign-in, and that difference can double a quote.

Department or team is a group chats can be directed to. Routing is the rule deciding who receives a conversation. Queue is what forms when chats arrive faster than they are answered; queue depth is how many are waiting.

Concurrency is how many conversations one agent handles at once, and it is the single most important number in chat economics — it is why chat costs less per conversation than phone. Transfer moves a conversation to another agent; escalation moves it to someone more senior or to another channel; handover usually means passing from automation to a human.

The measurements

First response time is how long a visitor waits for a first reply. Check whether a given dashboard counts an automated greeting as the first response, because that single choice can make the number look several times better than the experience.

Resolution means the conversation ended with the problem solved, but almost nobody measures it directly — most tools infer it from a proxy such as the chat closing without a follow-up. Missed chat is a conversation nobody answered, and definitions vary on whether an unanswered chat that later got an offline reply counts.

Deflection is a contact that did not happen because the answer was available elsewhere; it is always an estimate. Containment is the share of conversations that finished without a human, which is the automation-specific cousin of deflection.

CSAT is a satisfaction rating collected after a conversation, usually from a small and unrepresentative fraction of visitors. SLA is a stated commitment about response time and coverage. Abandonment is a visitor leaving before being served.

Automation and knowledge

Chatbot traditionally means a rule-based flow with buttons and scripted branches; AI chat generally means a model answering in natural language. The distinction matters because they fail differently — a rule-based bot fails visibly by not understanding, and an AI answer can fail invisibly by being confidently wrong.

Knowledge base is your library of help content, which is also the source material an AI answers from. Grounding or training on your content means the answers are drawn from that material rather than from the model's general knowledge. Hallucination is a fluent, plausible, incorrect answer.

Guardrails are the limits on what the automation is allowed to discuss or promise. Fallback is what happens when it cannot answer — ideally a clean handover to a person rather than a loop.

Where the definitions bite

Three of these terms cause most of the confusion in vendor comparisons. First response time, because of the automated-greeting question. Resolution, because everyone measures a different proxy. And containment, because a conversation where the visitor gave up counts as contained in most systems, which makes a frustrating automation look successful.

When comparing two numbers from two sources, the useful question is not which is higher but which definition each one used. If you cannot find out, the comparison is not usable, and the honest thing is to say so rather than to publish it with a caveat nobody will read.

Put it into practice

MyLiveChat is free forever for one agent, with unlimited chats and the embed code ready in about a minute.

Free forever for 1 agent

Give every visitor an instant way to reach you.

Launch live chat, connect your knowledge base, and add AI answers when you are ready. No credit card, no trial clock.