Table of Contents
- Key takeaways
- What is a support chatbot?
- How were these support chatbot examples chosen?
- FAQ chatbot examples
- 1. Emma (USCIS): a government FAQ chatbot
- 2. IRS chatbots: FAQ answers about tax notices
- 3. Scarlett (Safelite): pre-purchase FAQs and booking
- What does a good FAQ chatbot reply look like?
- Order status chatbot examples
- 4. Scout (BARK): order tracking for a subscription box
- 5. DPD: a parcel chatbot that went wrong after an update
- What does a good order status reply look like?
- Returns chatbot examples
- 6. Next: an AI agent that arranges returns
- What does a good returns reply look like?
- Troubleshooting chatbot examples
- 7. Harmony (SiriusXM): fixing a car radio in chat
- What does a good troubleshooting reply look like?
- Billing and account chatbot examples
- 8. Chime: account questions at a fintech company
- What does a good billing reply look like?
- Handoff examples: when should a support chatbot pass to a person?
- 9. Air Canada: a wrong refund answer that went to a tribunal
- What does a good handoff message look like?
- Chatbot welcome message examples
- What do these support chatbot examples have in common?
- Which support chatbot example should you copy?
- Where SiteGPT fits
- Frequently asked questions
- Basics
- What is an example of a support chatbot?
- What is an FAQ chatbot?
- What are the most common support chatbot use cases?
- Examples by job
- What is a good example of an order status chatbot?
- Which chatbot examples handle returns?
- Are there examples of chatbots that failed?
- Writing bot replies
- What should a chatbot welcome message say?
- What makes a good chatbot response?
- Getting started
- Which support job should a small business automate first?
- Can a chatbot replace a customer service team?
- Sources
Created by
Do not index
Created time
Oct 17, 2023 11:28 PM
FAQ and customer service chatbot examples fall into six jobs: answering FAQs, order status, returns, troubleshooting, billing, and handoff to a person. The nine bots below come from USCIS, the IRS, Safelite, BARK, DPD, Next, SiriusXM, Chime, and Air Canada.
TL;DR
- Nine real bots are grouped here by the job each one does. Seven are working examples and two are public failures.
- FAQ bots are the simplest to copy. USCIS Emma and the IRS chatbots answer common questions in English and Spanish.
- The IRS reports that its voice bots and chatbots have helped more than 13 million taxpayers since January 2022.
- Chime's resolution rate rose from 50% to over 70%, and BARK kept a 98% satisfaction rate. Both figures are vendor-reported.
- Air Canada had to pay C$812.02 after its chatbot gave a wrong refund answer. Policy answers need a source link and a route to a person.
Key takeaways
Support job | Example bot | What the bot does | Published detail |
FAQ answers | Emma (USCIS) | Answers immigration questions and points to the right page | English and Spanish |
FAQ answers | IRS chatbots | Answers questions about tax notices | 13 million taxpayers helped since January 2022 |
Order status | Scout (BARK) | Answers tracking and shipping questions | About a quarter of conversations (vendor-reported) |
Returns | Next AI agent | Arranges a return without an order number | Live in 6 weeks (vendor-reported) |
Troubleshooting | Harmony (SiriusXM) | Helps refresh an in-car radio | No figure published |
Billing and accounts | Chime AI agent | Answers direct deposit and password questions | Resolution up from 50% to 70%+ (vendor-reported) |
Handoff | Air Canada chatbot | Gave a wrong bereavement refund answer | C$812.02 ordered by a tribunal |
What is a support chatbot?
A support chatbot is software that answers customer questions in a chat window and passes the conversation to a person when it cannot help. Older bots follow fixed menus. Newer ones use a language model and answer from the company's own content or account data.
The basics are covered in what a customer service chatbot is. This page stays with examples.
Two sibling guides cover nearby ground. For a list of 16 large brands, see companies using chatbots for customer service. For sales and SaaS bots, see the B2B chatbot examples. If you are still deciding which job a chatbot should do beyond customer service, start with the chatbot use cases by industry and business function.
How were these support chatbot examples chosen?
These support chatbot examples were chosen on three criteria. Each entry follows the same template.
- A public page names the bot. The source is the organization's own page, a vendor case study, or a news report.
- The source says what the bot does. Each example maps to one customer service job.
- Every figure comes from that source. Figures from a vendor case study are labelled vendor-reported. An entry with no published figure says so.
Five of the nine examples come from case studies published by Sierra, the vendor that built those agents. Treat those numbers as direction. None is an independent audit.
The sample replies under each job are illustrative. They show the shape of a good answer and are not quotes from any company's bot.
FAQ chatbot examples
FAQ chatbot examples are bots that answer common questions from content the organization has already written. The three below cover immigration services, tax notices, and auto glass repair.
1. Emma (USCIS): a government FAQ chatbot
Emma is the virtual assistant on the website of US Citizenship and Immigration Services.
- What the bot does: Emma answers questions about USCIS services, guides visitors through the website, and finds information from the visitor's own search terms.
- Published detail: Emma works in English and Spanish. USCIS says visitors can ask in their own words and do not need government terms. No usage figure is published on the page.
- Handoff: The page describes none. For a personal case, USCIS points people to the online case status tool or its phone line.
- Pattern to copy: Let people ask in everyday words, then show related information under the answer.
2. IRS chatbots: FAQ answers about tax notices
The IRS runs chatbots that answer basic questions from people who receive a notice about underreported tax.
- What the bot does: The chatbot covers notices CP2000, CP2501, and CP3219A. It answers what to do after a notice arrives, how to get more time to respond, and how to check that the IRS received a response.
- Published detail: Since January 2022, IRS voice bots and chatbots have helped more than 13 million taxpayers avoid a phone wait. That includes roughly $151 million in payment agreements. The bots work in English and Spanish.
- Handoff: At the end of the conversation, the taxpayer can press a "representative" button to reach a live assistor.
- Pattern to copy: Scope the bot to one topic. The IRS built this one for three notice types.
3. Scarlett (Safelite): pre-purchase FAQs and booking
Scarlett is Safelite's AI agent for drivers with broken auto glass. Scarlett runs in chat and voice at any hour.
- What the bot does: Scarlett answers the questions a first-time buyer has: insurance coverage, glass type, repair time, and shop or mobile service. Scarlett then guides the customer through booking.
- Published detail: The vendor case study says Scarlett can identify a vehicle's model and year from a license plate. It reports high resolution rates and gives no figure.
- Handoff: The case study describes none for Scarlett.
- Pattern to copy: Answer the question, then offer the next step in the same chat. Scarlett moves from answers to booking.
What does a good FAQ chatbot reply look like?
A good FAQ chatbot reply gives the answer in the first sentence and links the page it came from. This sample is illustrative:
- Visitor: do you ship to Canada
- Bot: Yes. Standard shipping to Canada takes 6 to 9 business days. Rates are on the shipping page. Is there anything else you want to check?
To build one, follow the steps in how to build an AI FAQ chatbot. Ready-made wording is in the chatbot templates library.
Order status chatbot examples
Order status chatbot examples are bots that answer "where is my order" questions. One example below works. The other shows what happens when an update ships without testing.
4. Scout (BARK): order tracking for a subscription box
Scout is the AI agent at BARK, the company behind BarkBox.
- What the bot does: Scout takes high-volume questions about order tracking, shipping updates, and basic account details.
- Published detail: Within the first year, Scout handled roughly a quarter of all customer conversations. BARK kept a 98% customer satisfaction rate, and chat response times dropped to under two minutes. These figures are vendor-reported.
- Handoff: Emotional conversations stay with BARK's human team. The case study names the loss of a dog as one case.
- Pattern to copy: Give the bot the repetitive questions and keep people on the sensitive ones.
5. DPD: a parcel chatbot that went wrong after an update
DPD is a parcel delivery company that runs an AI element in its online chat, next to human operators.
- What the bot did: In January 2024 a customer said the chatbot could not answer his queries. He then got it to swear and to criticize DPD.
- Published detail: One post about the exchange was viewed 800,000 times in 24 hours. DPD said an error occurred after a system update and that it had disabled the AI element.
- Handoff: DPD offers human operators by phone and WhatsApp to customers with a tracking number.
- Pattern to copy: Rerun a fixed set of test questions after every update. Limit the bot to the company's own content.
- Source: BBC News, January 19, 2024
What does a good order status reply look like?
A good order status reply states the status, the expected date, and the next step. This sample is illustrative:
- Visitor: where's my order
- Bot: I can check that. What is the email address on the order?
- Visitor: (gives the email address)
- Bot: Your order shipped on Tuesday and is due on Friday. The tracking link is in your confirmation email. If it has not arrived by Monday, reply here and a team member will pick it up.
A reply like this needs the bot to read order data. More retail cases are in the guide to ecommerce chatbot examples.
Returns chatbot examples
Returns chatbot examples are bots that start a return or explain the return policy. The example below comes from a fashion retailer.
6. Next: an AI agent that arranges returns
Next is a British fashion retailer that sells in 83 countries.
- What the bot does: The agent handles arrange-return requests. It matches the customer to an order without asking for an order number and manages identity verification.
- Published detail: Next launched the agent on two use cases in six weeks. The case study gives no resolution figure. The six-week figure is vendor-reported.
- Handoff: The case study describes none.
- Pattern to copy: Do not make the customer look up an order number. Find the order from details the customer already knows.
What does a good returns reply look like?
A good returns reply quotes the policy and tells the customer the first step. This sample is illustrative:
- Visitor: can I send back the jacket I bought last week
- Bot: Yes. Items can be returned within 30 days of delivery if they are unworn. Start the return on the returns page and print the prepaid label. The full policy is linked below.
Troubleshooting chatbot examples
Troubleshooting chatbot examples are bots that walk a customer through fixing a product. The example below is a satellite radio service.
7. Harmony (SiriusXM): fixing a car radio in chat
Harmony is the AI agent at SiriusXM, an audio service with 34 million subscribers.
- What the bot does: Harmony helps subscribers refresh an in-car radio. SiriusXM says the task needs special knowledge and well-defined steps.
- Published detail: SiriusXM's earlier rule-based chatbot could not handle loose phrasing, such as a request to "rejigger" a radio. The case study publishes no resolution figure.
- Handoff: The case study describes none.
- Pattern to copy: Write the fix as numbered steps before the bot goes live. The bot follows steps that exist.
What does a good troubleshooting reply look like?
A good troubleshooting reply asks one question, gives one step, and checks the result. This sample is illustrative:
- Visitor: my speaker won't connect
- Bot: Is the light on the front blinking or solid?
- Visitor: blinking
- Bot: A blinking light means the speaker lost Wi-Fi. Hold the back button for 5 seconds until the light turns solid. Did that work?
- Visitor: no
- Bot: A team member can take it from here. Leave your email and the speaker's serial number.
For internal help desks, see the guide to an IT support chatbot.
Billing and account chatbot examples
Billing and account chatbot examples are bots that answer questions about payments, deposits, and logins. These questions involve money, so accuracy matters more than speed.
8. Chime: account questions at a fintech company
Chime is a US fintech company that uses AI agents for member questions.
- What the bot does: The agent answers questions that range from the status of a direct deposit to a password reset.
- Published detail: Chime's resolution rate rose from 50% to over 70%. Chime also reports significantly better resistance to hallucinations. Both points are vendor-reported.
- Handoff: The case study does not describe Chime's handoff in detail.
- Pattern to copy: Track resolution rate before and after launch. Chime could report a gain of at least 20 points because it had a baseline. To set up the same before-and-after measurement for your own bot, follow the baseline steps in the chatbot case studies guide.
SiriusXM's Harmony also covers two billing tasks: making a payment and updating billing information.
What does a good billing reply look like?
A good billing reply explains the charge and offers a person for a dispute. This sample is illustrative:
- Visitor: why was I charged twice
- Bot: I can see why that is worrying. A second charge is often a temporary hold that clears in 3 to 5 business days. If both charges show as completed, a team member will review it. Do you want me to pass this on?
Why bots make up answers, and how to reduce it, is covered in the guide to chatbot hallucinations.
Handoff examples: when should a support chatbot pass to a person?
A support chatbot should pass to a person when the customer asks, when the bot has no sourced answer, and when a wrong answer costs money. The example below shows the cost of a missing handoff.
9. Air Canada: a wrong refund answer that went to a tribunal
Air Canada ran a chatbot on its website that answered policy questions.
- What the bot did: In 2022 the chatbot told a customer he could claim a bereavement fare refund within 90 days of the ticket date. Air Canada's policy did not allow refunds for completed travel.
- Published detail: In February 2024 a tribunal ordered Air Canada to pay C$650.88 in fare difference, C$36.14 in interest, and C$125 in fees. That is C$812.02 in total.
- Handoff: The report describes none. The tribunal held Air Canada responsible for everything on its website, including the chatbot's answers.
- Pattern to copy: For refund and fare rules, have the bot quote the policy page and link it. Offer a person for any exception.
- Source: The Guardian, February 16, 2024
What does a good handoff message look like?
A good handoff message confirms the request, says who will reply, and asks for contact details once. This sample is illustrative:
- Visitor: I need to talk to a person
- Bot: Of course. A team member will reply here during working hours, 9 am to 5 pm. Leave your email so they can reach you if you close this window.
The IRS and BARK examples above show two working rules. The IRS bot shows a representative button. BARK routes emotional conversations to its team. Tools that pass the transcript along are compared in AI chatbots with human handoff.
Chatbot welcome message examples
Chatbot welcome message examples work best when they say the bot is a bot and name what it can do. These five are illustrative, one for each job:
- FAQ: "Hi, this is the store's AI assistant. Ask about shipping, sizing, or returns."
- Order status: "Hi. I can check an order for you. What is the email address on the order?"
- Returns: "Need to send something back? Tell me the item and I will explain the steps."
- Troubleshooting: "Hi, I am an AI assistant for setup and fixes. Which product do you need help with?"
- Billing: "I can explain a charge or a plan. For disputes, I will connect you with the billing team."
Each message tells the visitor the scope in one line. More rules for tone and scope are in the guide to chatbot best practices. When you need a full conversation from the first message to the handoff, the customer service chatbot example walks through one turn by turn, with sample dialogues for eight industries.
What do these support chatbot examples have in common?
These support chatbot examples have four things in common.
- Each bot has a narrow job. The IRS chose three notice types. Next chose two use cases. BARK started with tracking questions.
- The answer comes from the organization's own content or data. Emma answers from USCIS information. Scarlett draws on the knowledge of Safelite's technicians and service staff.
- A person stays reachable. The IRS shows a representative button. BARK keeps emotional conversations with its team. DPD keeps phone and WhatsApp lines.
- The failures came from skipped checks. DPD's error followed an update. Air Canada's bot gave an answer that its policy page contradicted.
Which support chatbot example should you copy?
You should copy the support chatbot example that matches your most common question and the cost of a wrong answer. The table maps common needs to one example.
If you need... | Copy... | Why |
Answers to repeat questions from pages you already have | USCIS Emma | Emma answers in everyday words from existing information. |
A bot for one narrow topic | IRS chatbots | The IRS limited its bot to three notice types. |
Answers that lead to a booking | Safelite's Scarlett | Scarlett answers buyer questions, then guides the booking. |
Fewer "where is my order" contacts | BARK's Scout | Scout takes tracking questions and people take sensitive ones. |
Returns without an order number | Next | The agent finds the customer from other details. |
Step-by-step fixes | SiriusXM's Harmony | Harmony follows defined steps for a radio refresh. |
Account answers with a measured result | Chime | Chime tracked resolution rate before and after. |
A rule for policy and refund questions | Air Canada, as a warning | A wrong policy answer cost C$812.02 and a public ruling. |
Most small teams should start with FAQ answers. The content already exists, and a wrong answer on shipping times costs less than a wrong answer on a refund.
Where SiteGPT fits
SiteGPT fits the first job on this page: FAQ answers from content a business already has. It is not ranked against the examples above, which are custom builds by large organizations.
- What it trains on: website URLs, sitemaps, files, raw text, Google Drive, Dropbox, OneDrive, Box, Notion, GitHub, and YouTube.
- Handoff: human support is available on any plan. By default the AI escalates when the visitor asks for a person, the situation is urgent, or the visitor is frustrated after an unhelpful answer.
- Price: Starter costs $59 a month billed monthly, or $39 a month billed yearly ($468 a year). It includes 1 chatbot and up to 4k GPT-4.1-mini messages a month.
- Limits: there is no free plan, and the 7-day trial needs a credit card. Starter has one member, manual content refresh, and no chat integrations or API access. SiteGPT answers from trained content. An order or account lookup like BARK's or Chime's needs a connection to order or account data, which a content-trained setup does not include.
Frequently asked questions
Basics
What is an example of a support chatbot?
An example of a support chatbot is Emma, the virtual assistant on the USCIS website. Emma answers immigration questions in English and Spanish and guides visitors to the right page.
What is an FAQ chatbot?
An FAQ chatbot is a bot that answers common questions from content the organization has already written. The IRS chatbots for tax notices are one example. They answer three kinds of question about three notice types.
What are the most common support chatbot use cases?
The most common support chatbot use cases are FAQ answers, order status, returns, troubleshooting, billing and account questions, and handoff to a person. The nine examples on this page cover all six. When you are planning which customer service jobs to give AI beyond the chat window, such as ticket triage, agent assist, or quality scoring, use the list of customer service AI use cases.
Examples by job
What is a good example of an order status chatbot?
A good example of an order status chatbot is Scout at BARK. Scout answers tracking and shipping questions and handled about a quarter of BARK's customer conversations in its first year, according to the vendor.
Which chatbot examples handle returns?
Next, the British fashion retailer, uses an AI agent to arrange returns. The agent matches the customer to an order without an order number.
Are there examples of chatbots that failed?
Yes. DPD disabled part of its chatbot in January 2024 after it swore at a customer. Air Canada was ordered to pay C$812.02 in February 2024 after its chatbot gave a wrong refund answer.
Writing bot replies
What should a chatbot welcome message say?
A chatbot welcome message should say that the visitor is talking to a bot and name two or three things it can do. One example: "Hi, this is the store's AI assistant. Ask about shipping, sizing, or returns."
What makes a good chatbot response?
A good chatbot response answers in the first sentence, links the source page, and offers a next step. For anything about money, it quotes the written policy and offers a person.
Getting started
Which support job should a small business automate first?
A small business should automate FAQ answers first. The content already exists on the website, and the bot needs no connection to order or account systems.
Can a chatbot replace a customer service team?
No example on this page replaced its team. BARK keeps emotional conversations with people, the IRS offers a representative button, and DPD runs phone and WhatsApp lines next to its chatbot. More company cases are in the list of conversational AI examples.
Sources
All sources were opened and checked on October 12, 2026.
Written by
