Chatbot Maker teaching notes
For teachers, co-op leaders and parents. One period of about forty-five minutes ends with every student holding a link to a chatbot they made from rules, tested against their own messages. A second period has partners try to break each other's bots, reads the bot's real code, and compares it with the AI chatbots students already use. Each part says what to say, what students do, what their screens should show, what usually goes wrong, and a question to ask. You do not need to know how to code.
Using this with a class? Make a free class and add your students yourself: each gets a username and password, with no email or Google account needed. Or give them one join code. Using Google Classroom? Post the link there; with a free class, the link also puts each student in your class, so you see their work. mytekDev is also available from the Google Workspace Marketplace, so your school's Google admin can install it for everyone.A first lesson: Break the bot. One student builds, the other tests: the tester types real questions the bot should answer, and every message that gets the "I do not know" reply goes into the builder's test messages with + Add to my tests. Then swap. The pair that leaves the fewest misses wins.
Set up a free class Share to Google Classroom- About these lessons
- How to read this guide
- Period 1, part 1: how does the bot answer? (10 minutes)
- Period 1, part 2: make it your bot (20 minutes)
- Period 1, part 3: test it and publish it (15 minutes)
- Period 2: break the bot, read the code (45 minutes)
- Safety: what the bot does, and what it cannot do
About these lessons
Period 1 has three parts: see how the starter bot answers, make the bot your own, then test it and publish it. Period 2 breaks the bots in pairs, reads the code, and talks about how this bot is different from the AI chatbots students know.
Everything you are asked to say or point at is written here in the words on the screen. A rule card says When someone says, then the reply. The line under each reply in the chat says which rule answered. The box at the top right is called Ready to publish?: a red line in it must be fixed before publishing, and an orange line makes the bot better.
How to read this guide
- Say what to tell the class, in a sentence or two. Use your own words if you like.
- Students do the steps, in order. Read them out or put them on the board.
- Screens should show what you see when you walk round the room and the step worked.
- What goes wrong the usual problems, and the fix.
- Ask one question to check the idea, not the clicking.
Period 1, part 1: how does the bot answer? (10 minutes)
- Say This is Pip, a chatbot for a library. Ask it things, and watch the small line under each answer. We are going to find out whether Pip understands you.
- Students do Type "When are you open?" and press Send. Read the line under the answer, then click it. Type "Do you have a pet snake?". Then type "I am open to ideas".
- Screens should show The opening hours, with "Rule 2: hours, open, close" under them, and Rule 2's card on the left with a green border. "Hmm, I do not know that one" for the snake, with "No rule matched". The opening hours again for "I am open to ideas".
- What goes wrong Nothing usually does here. If a student gets a different answer, they typed a word another rule listens for: read the line under it together.
- Ask Why did Pip tell you the opening hours when you said you were open to ideas? Does Pip understand what you type?
Period 1, part 2: make it your bot (20 minutes)
- Say A rule is words to listen for and a reply. The bot tries the rules from the top, and the first rule whose words it hears answers. Make every rule yours.
- Students do Change the bot's name at the top left, and its face, color and greeting on the right. Change each rule's words and reply, or press Remove. Press Add a rule for something your bot should know. Talk to it after every change.
- Screens should show Their bot's name and face in the chat's header, rules about their topic, and answers that come from them.
- What goes wrong "Something is empty." in red: a rule has no words or no reply; fill it in or remove it. The wrong rule answers: an earlier rule hears the same word; move the right rule up, or use But not when they say. A word is not heard: the bot treats dog and dogs as the same word and forgives one wrong letter in a long word, but run and running are different words, so put both in. More than one reply on its own line: the bot picks one each time.
- Ask Two of your rules both listen for the same word. Which one answers, and how could you change that?
Period 1, part 3: test it and publish it (15 minutes)
- Say Real people will type things you did not expect. Write test messages, run them after every change, then check the bot and publish it.
- Students do Under Test messages, write five things a person might really say, one on each line, and press Run them all. Fix any that got the "I do not know" reply. Read Ready to publish? at the top right and click each line to read why. Then press Publish, then Publish my bot, and open the link in a new tab or scan the square with a phone.
- Screens should show "All 5 tests answered by a rule", "Ready to publish." at the top right, then a link and a square code in the Publish window, and the bot open in a new tab with a yellow note that it is a chatbot a student made, not a person.
- What goes wrong "Your bot must not ask for personal information.": a reply asks for a last name, an address, a school, a phone number, an email, a password, an age or a photo; reword it so it does not ask. A red line about personal information or a link: take it out; a first name is fine. Publishing says the class shares inside the class only: that is your class setting. Publishing says a full name of someone in the class is in the bot: the site checks names the app cannot see, so take it out.
- Ask Your bot is public now. Why must it never ask the people who chat with it for their address or school?
Period 2: break the bot, read the code (45 minutes)
Pairs swap bots. The tester's job is to find what the bot cannot answer, using real questions, not nonsense. Then everyone reads the bot's real code, and the class talks about how it is different from the AI chatbots they use.
- Say Your partner built a bot about something. Ask it real questions a visitor would ask. Every time it gets one wrong, tell the builder.
- Students do Ten minutes as tester on your partner's computer, then swap. As builder, click + Add to my tests under each message that missed, then fix your rules until Run them all answers every test.
- Screens should show A longer test list, with the misses turning into rule answers as the builder fixes them.
- What goes wrong Testers type gibberish to win: only real questions count. A builder adds one rule for every exact message: show them one rule with several words catches many messages.
- Ask Which question was hardest to write a rule for, and why?
- Say Press See how it works. This is the real code of your published bot. Your rules are in it.
- Students do Click one of your rule cards and find its lit line in the code. Find the bot's greeting and its "I do not know" reply in the code. Find function respond, the code that tries each rule.
- Screens should show A dark pane on the right, with the picked rule's line lit near the top, in a list called rules.
- What goes wrong The code pane covers the panel: that is how it works; press Close to get the panel back. A student wants to change the code: Open in Code Editor carries a copy of the bot into the Code Editor.
- Ask Your rules are just a list of words and replies. What part of the code decides which reply to give?
Then talk about it. In 1966 Joseph Weizenbaum at MIT published ELIZA, a program that looked for key words in what you typed and gave a reply, the same idea as their bots. It did not understand anything, yet many people felt it did. Weizenbaum wrote that his secretary asked him to leave the room so she could talk with ELIZA in private. Newer AI chatbots do not use rules a person wrote: they are trained on huge amounts of text and predict what words come next. They answer far more, but they do not understand the way people do either, and they can be confidently wrong.
Safety: what the bot does, and what it cannot do
Messages about wanting to hurt yourself, being hurt by someone, or being bullied get a fixed reply no rule can change. It says the bot is not a person, to tell a trusted adult, and gives a US number to call or text (988, or the Childhelp hotline for someone being hurt), and 911 for danger right now. It works from a word list, so it catches the common ways children say these things, not every way. Students do not need to test it, and should not.
A published bot runs only in the visitor's browser. It does not save what anyone types or send it anywhere, and the student who made it never sees the conversations.
More help
Is something on this page wrong or not clear? Write to us.
Page last reviewed November 2026.
Google, Google Classroom and Chromebook are trademarks of Google LLC.