FAQ Chatbot

General definition

An FAQ Chatbot is a conversational program scoped to a fixed body of knowledge: a help centre, product documentation, HR policies, a course syllabus or a clinic’s patient information. Its value is deflection, the share of questions resolved without a human, and its constraint is honesty: it should answer only from the knowledge set and say so when it cannot. That makes it the most common, and most measurable, kind of AI chatbot in production.

There are three ways to build one, in rising order of capability:

• Keyword and rule matching: a table of question patterns mapped to canned answers; transparent and cheap, but it only catches phrasings someone anticipated
• Intent classification: a classifier trained on example questions routes each message to one of a set of answers; better recall, still limited to predefined intents
• Retrieval-augmented generation (RAG): the documents are indexed, the most relevant passages are fetched for each question and a language model writes an answer from them, with a citation; handles new phrasings and new questions as soon as the documents change

The numbers that matter are deflection rate (questions closed without a person), containment (conversations that never asked for a human), answer accuracy checked against a golden set of real questions, and the fallback rate, where the bot admits it does not know. A good FAQ bot has a fallback rate above zero, because a bot that always answers is guessing. Escalation should be one message away: a button or a phrase that opens a ticket, adds a human to the room, or captures an email, with the transcript attached so the person does not start from scratch.

Common failures are stale content (the bot answers from last year’s pricing page), answering outside scope (a benefits bot giving legal advice), and hallucination when the retrieved passages are thin. The controls are a system prompt that restricts the bot to its sources, refreshing the knowledge base on a schedule, logging every unanswered question as a content gap, and regular review of transcripts.

In the Ethora ecosystem

On the Ethora AI SDK an FAQ chatbot is an AI agent with a knowledge base and a narrow system prompt. Point the website crawler at your help centre or upload the documents, tell the agent to answer only from those sources and to say when it cannot, choose a model (a cloud API or a self-hosted model through an OpenAI-compatible endpoint), and attach the agent to a room, a channel or the embeddable assistant widget. Per-agent knowledge with live web indexing keeps the answers current when the pages change. Most teams have a working bot in minutes rather than weeks, and the no-code agent builder needs no engineering time at all.

Escalation is built into the medium: the bot lives in a real chat room, so a human team member can be added to the conversation with the history intact, and the same room can hold several agents for different topics. Every exchange is archived server-side with an audit trail, and Trust and Safety rules can stop a customer from pasting sensitive data into the bot. The chat widget entry covers the website side; the AI SDK entry covers the platform.