What is an AI chatbot?
Short answer
An AI chatbot is a program that talks with people by text or voice and uses an AI model to interpret what they write and generate a reply. A scripted chatbot follows fixed menus and rules. The AI kind handles free wording, including mixed English and Filipino, but it can answer wrongly, so it needs approved sources of answers, a handover to staff and a disclosure that it is a bot.
On this page
How an AI chatbot produces an answer
IBM defines a chatbot as “a software application that communicates with people through text or voice,” and says AI chatbots “use AI models to interpret natural language, maintain context and generate responses” (IBM Think, What is a chatbot?, updated 16 July 2026). In practice the work has three parts.
- Understanding the message. The model reads the customer’s words and works out what they want, such as a price, a booking or a complaint. It does not need the customer to pick from a menu, and it can read a message that mixes English and Filipino.
- Finding information. A bot that only relies on what the model learned in training will not know your prices, hours or policies. Business bots therefore look up an approved source, such as a price list, a FAQ document or a booking system, and answer from it.
- Writing the reply. The model composes the answer in natural language, within instructions the business gave it about tone and limits.
AI chatbot and scripted chatbot compared
IBM calls the older kind “rule-based chatbots” that “follow predefined rules, decision trees or scripted responses.” They handle predictable requests well and struggle with questions their replies do not cover.
| Scripted chatbot | AI chatbot | |
|---|---|---|
| Input | Button taps, menu choices or exact keywords | Free-text messages, sometimes voice |
| Flexibility | Low: anything off the script fails | High: it can interpret varied wording |
| Setup | Someone writes every path and reply | Someone supplies approved information and instructions, then tests |
| Accuracy control | High: it can only say what was written | Lower: it generates text, so wrong answers are possible |
| Typical failure | “Sorry, I didn’t understand” | A fluent reply that is wrong |
| Best for | Fixed steps such as an order status menu | Varied questions about services, prices and policies |
Swipe the table sideways to see all columns.
Many bots combine the two. A scripted menu handles the common steps and an AI model handles the free-text questions around them.
What kind of AI a chatbot uses
Language models
Most AI chatbots now run on a large language model, a model trained on large amounts of text to predict and produce language. ChatGPT is an example of a general chat product built on such models. Whether it counts as a business chatbot depends on how it is set up: a bot that answers customers for a company is configured with that company’s instructions and information, while the general chat product answers from broad training.
Retrieval from business information
To answer from a company’s own material, many chatbots use retrieval-augmented generation (RAG). A 2020 research paper by Patrick Lewis and colleagues introduced the term for models that pair a generator with a retriever that looks up information in an index at answer time (Lewis et al., arXiv, 22 May 2020). For a business, the practical meaning is simple: the bot looks up the answer in your documents, then writes the reply. If the document is wrong or out of date, the reply will be too. The RAG entry has the short definition.
Examples by channel
These are illustrations of how the same bot appears in different places, not reports of real deployments.
- Website. A chat window on a clinic site answers opening hours and which services are offered, then offers booking links. Questions about symptoms go to staff.
- Messenger. A seller’s Facebook page bot answers “magkano po?” and delivery-area questions, and passes payment problems to the seller. The Messenger bot guide covers setup.
- WhatsApp and Viber. The rules differ by platform. WhatsApp’s business policy says a business may use automation when responding during the 24-hour window but must have “prompt, clear, and direct escalation paths” (WhatsApp Business policy, last updated 23 September 2026). Viber’s developer documentation says bots can only be created on commercial terms and can message a user only after that user has subscribed (Viber REST bot API).
- Inside apps. A booking or delivery app shows a help bot in a support screen, with a button to reach a person.
What a chatbot handles well and where a person steps in
An AI chatbot handles repeat questions well: opening hours, location, price ranges, how to book, order status. It works at any hour and applies the same wording each time.
A person should handle complaints, refunds, price negotiation, anything the bot cannot find in its approved source, and any question about health, legal or financial decisions. In a clinic, the bot can book appointments but should not answer clinical questions. A rule that sends these topics to staff is part of the design.
What to check before putting one in front of customers
Sources of answers
Decide which documents the bot may use and keep them current. Test it with real customer messages, including Taglish and misspellings, and read the replies.
Handover
The bot needs a clear route to a person, and staff need to see the conversation so far. Meta’s Messenger policy says a Human Agent tag lets a business respond manually within a 7-day period, and that standard messaging covers a 24-hour window. It also says automated bots “must respond to any and all input from the user” (Meta Messenger Platform policy, read 12 October 2026). Check the current text before building, because platform rules change.
Disclosure
The same Meta policy says automated chat experiences “must disclose that a person is interacting with an automated service” where the law requires it, and recommends telling users even when it does not. Telling customers they are talking to a bot is a low-cost habit that also supports the transparency principle in the National Privacy Commission’s AI advisory. Law-firm summaries of Advisory No. 2024-04 describe a duty to tell people the nature, purpose and extent of AI processing (Rajah & Tann Asia; advisory text at the NPC).
Data handling
Customer messages contain names, numbers and sometimes addresses or health details. Under the Data Privacy Act of 2012, the business is responsible for how the bot’s provider stores and uses them. Ask the provider where chats are stored, who can read them and how long they are kept.
Frequently asked questions
How does an AI chatbot differ from a scripted chatbot?
A scripted chatbot follows menus and written rules and can say only what someone wrote in advance. An AI chatbot uses a model to interpret free wording and generate a reply, which makes it more flexible and less predictable.
Is ChatGPT an AI chatbot?
ChatGPT is a general chat product built on a language model, so it fits the broad definition. A business chatbot is a bot set up to answer customers for that business, using the business’s own information.
Can an AI chatbot answer from a business’s own information?
Yes. Many bots use retrieval to look up answers in the business’s documents, then write the reply. The answers are only as accurate as those documents.
Do customers need to be told they are chatting with a bot?
Meta’s Messenger policy says automated chats must disclose this where the law requires it and recommends it otherwise. The WhatsApp and Viber rules differ, so read each platform’s current policy.
What type of AI is a chatbot?
Most current chatbots use a large language model, often with retrieval from the business’s documents. Older chatbots use scripted rules and no model.
Sources
All sources were read on 12 Oct 2026.
- IBM, What is a chatbot?, IBM Think, published 15 October 2021, updated 16 July 2026.
- Lewis et al., Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, arXiv, 22 May 2020.
- Meta, Messenger Platform policy overview.
- WhatsApp, WhatsApp Business policy, last updated 23 September 2026.
- Rakuten Viber, Viber REST bot API.
- National Privacy Commission, NPC Advisory No. 2024-04, 19 December 2024.
- Rajah & Tann Asia, NPC’s Advisory on AI and Data Privacy.