Chatbot examples in 2026: answers, order updates, bookings and hand-off to a person

Start from the job a chatbot does, not the technology

Lists of chatbot examples usually mix very different things: a greeting pop-up, a voice assistant, a shop assistant and a government helper. They are easier to understand by the job each does. One answers questions from a body of published content. One sends updates about an order or an appointment. One takes a booking by asking for a date and a name. One collects a request and passes it to a person.

This article follows those jobs and uses examples that the organisations behind them have documented themselves: the GOV.UK team, WhatsApp, Telegram and Google's Dialogflow. It does not rank products or quote results from vendors. The aim is to help a small business decide which job its own bot should do first, and what that choice means for accuracy, privacy and staff time.

A knowledge bot: GOV.UK Chat answering from published pages

GOV.UK Chat was the first generative AI experiment of the Government Digital Service. Its team describes a chatbot grounded on published GOV.UK content using retrieval augmented generation, with an OpenAI model producing the answers. The team did not train or fine-tune a model. Prompt engineering aimed to keep answers based on retrieved content rather than on the model's own knowledge.

The pilot is useful because it was measured in public. A follow-up survey of 157 users found nearly 70% said the responses were useful and just under 65% were satisfied. Human experts assessed a sample of answers, and accuracy did not reach the level GOV.UK requires. Hallucinations happened in a few cases, mostly with ambiguous or inappropriate questions, and some users trusted the answers more than they should because the brand is credible. Some questions also went unanswered when the relevant page was too long, a reminder to split long pages into short, answerable ones.

For a business the lesson is practical. A bot that answers from your own pages is only as good as those pages, and it needs a sample of real conversations read by a person every week. Fix the page when an answer is wrong, rather than patching the bot around it.

Order updates and reminders on WhatsApp

The WhatsApp Business Platform documentation lists what companies can send through its Cloud API: text, rich media and interactive messages, including order confirmations, shipping updates, appointment reminders and one-time passwords. These are the order-status and reminder jobs, and they do not need an open-ended conversation at all.

The same documentation sets rules. Templates are the only type of message that can be sent to WhatsApp users outside of a customer service window, templates generally need approval before sending, and the user must have opted in with a statement that names your business and its intent. Templates also have quality scores and messaging limits. A reminder bot is therefore as much a matter of consent and wording as of code.

Intent-based bots for bookings and leads: how Dialogflow works

Not every bot needs a language model. Google's Dialogflow documentation describes an agent that matches each user expression to the best intent, and each intent carries training phrases, parameters and responses. When a phrase resembles a training phrase, the intent is matched, and the built-in machine learning expands your list with similar phrases so that you do not have to write every variant.

This structure suits booking and lead capture, which are mostly forms in the shape of a conversation. Parameters such as a date or a name are extracted using entity types, and with fulfillment enabled, Dialogflow sends a request to your webhook service, which can perform actions and return a response. The behaviour is predictable and easy to test, but anything outside the defined intents will not be understood. That is the main difference from an AI bot grounded on content.

Telegram bots that work on behalf of a business

Telegram's bot documentation describes bots as small applications that run entirely within the app, connected to their owner's server through the HTTPS-based Bot API. Among the uses it names are payments, with digital goods sold for Telegram Stars and physical goods through third-party providers, custom tools, and Telegram Business users connecting bots to answer messages on their behalf.

The limits are as informative as the features. A bot cannot start a conversation: the user has to message it or add it to a group first. In groups, bots by default see only relevant messages, and anyone who holds a bot's token has full control of it, so the token must be stored securely. For a small business this means the first message comes from the customer, and the bot's credentials are as sensitive as a password. The same page notes that bots can host Mini Apps built with JavaScript, with built-in authorization and notifications, which suits a catalogue or a booking screen better than a long chat.

Handing the chat to a person and saying it is a bot

A bot that cannot help must pass the conversation on. WhatsApp's documentation says its Meta Business Agent can run AI-powered agents that handle conversations using your FAQs, catalogues and custom API connectors, with handoffs to human agents. Decide in advance what triggers a hand-off, such as a complaint, a payment problem or two failed attempts, and who answers.

Disclosure is a separate matter. Article 50 of the EU AI Act, as reproduced on artificialintelligenceact.eu, says providers shall ensure that AI systems intended to interact directly with natural persons inform them that they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant and circumspect person. Whether and how this applies to your bot depends on your case, so check it with a specialist; a plain first line saying that this is an automated assistant is a safe habit regardless.

Data care and how to measure a bot

The GOV.UK team lists safeguards that any business can borrow in spirit. Queries were screened for personal data, which was not stored or shared. Pages containing personal data were removed from the system so they could not reach the model. A data protection impact assessment was completed, access was limited to invited users, and red teaming found exploits that were then closed. Decide what the bot may see, what it keeps and who can read the logs before launch. The pilot itself ran in five phases of about two weeks each, from internal testing to a private pilot, which is a sensible shape for a small trial too.

For measurement, copy the structure of that pilot at a small scale: a short survey after chats, a weekly read of sampled answers for accuracy, and a count of how many chats ended with a person. WhatsApp's documentation also mentions messaging and template analytics through its Business Management API. Pick two or three numbers and review them on a fixed day each week.

What to prepare before you order a chatbot

A bot is only as clear as the content behind it. Before ordering, choose the one job it should do first, write down the twenty questions customers ask most with the answers you stand behind, decide who takes over and in what hours, and settle where the bot will live: your site, WhatsApp or Telegram. Prepare the pages or documents it may quote and a list of what it must never say.

AI chatbot at VITON13 Studio, on its service page, is a bot for your site, WhatsApp or Telegram that answers from your content, collects leads and hands the chat to a person. The price is from $73. Send your questions and the channel you prefer, and the scope and the price are confirmed in writing before work starts.