Teachable Machine for kids: a parent and teacher walkthrough

Teachable Machine is a free Google website where a child teaches a computer to tell things apart (two toys, three hand shapes, a clap and a whistle) by showing it examples through the webcam or microphone, then watches it guess live. It needs no account, and Google says the training runs in your browser tab. This walkthrough was written by Peabody (peabody.kids), a free reading and learning site that is free through the end of 2026. Teachable Machine isn't ours, and nobody paid to be listed. Everything below was checked on Google's own pages in October 2026.

What is Teachable Machine?

Google's own description: a web-based tool that makes creating machine learning models "fast, easy, and accessible to everyone". You give it examples sorted into groups, which it calls classes. It learns from those examples. Then it guesses which class a new example belongs to, and shows how sure it is.

It can learn from three kinds of thing:

That's the whole idea of machine learning in one screen. Nobody writes a rule like "a teddy is brown and furry". The computer finds its own pattern in the examples. If the examples are poor, so is the pattern, and a child sees that for themselves within minutes.

How to use Teachable Machine with kids: a first 10-minute project

You need a laptop or desktop with a webcam, and two objects that look clearly different. A soft toy and a shiny toy car work well. (The makers of TM2Scratch, a Scratch add-on covered below, note that Teachable Machine doesn't run on iPads, so use a computer.)

1. Open teachablemachine.withgoogle.com and click Get Started. No sign-in.
2. Choose Image Project, then Standard image model. Google marks it "Best for most uses".
3. Rename the classes. Click "Class 1" and type "teddy". Click "Class 2" and type "car". Let your child choose the names. Naming the groups is the child's first decision as the teacher.
4. Collect examples. Under "teddy", click Webcam, allow the camera, then press and hold Hold to Record while your child turns the teddy slowly in front of the camera. Aim for 50 or more pictures. Do the same for the car.
5. Click Train Model and keep the tab open. It takes a little while.
6. Test it in the Preview panel. Hold up the teddy. The bar for "teddy" should fill up. Swap to the car. Then try something it has never seen, like a cup, and ask your child what they think it will say.
7. Break it on purpose. Hold the teddy in a different corner of the room, or with the lights dimmed. Does it still work? If not, add more examples in the new spot and train again.

Step 6 is where the learning happens. The cup will be called a teddy or a car, because those are the only two answers the machine knows. That surprises most children, and it's worth a conversation: the computer can only choose from what it was taught.

Is Teachable Machine free?

Yes. As of October 2026 it costs nothing, and Google's FAQ says there's no need to make an account or log in. The only time you'd sign in is to save a project to Google Drive, which needs a Google account. That's a grown-up's job.

Where does the data go?

This is the part parents ask about most, so here is what Google's FAQ says, in plain words:

• Training happens on your computer. The FAQ says the model trains in your browser tab "without sending anything to any servers". The home page adds that you can use it entirely on-device, without webcam or microphone data leaving your computer.
• Close the tab and it's gone. If you don't save, download or upload anything, nothing is kept, in your browser or on any server.
• Saving to Google Drive keeps the pictures. It saves a .zip file of all the samples in each class to your Drive, so you can pick up later. If those samples are your child's face, they're now in a Google account.
• "Upload my model" publishes the model, not the pictures. Google hosts the trained model at a link. Anyone with the link can use it, though no one can change it. The examples used to train it aren't published.
• Downloading the model saves it to your computer, and nothing goes on a server.

Our practical advice: for a first go, save nothing. If your child wants to keep a project, train it on objects rather than faces, and download it rather than uploading. Upload only when you need the link, such as for the Scratch route below.

For a wider checklist on any AI tool, see is AI safe for kids' learning?.

What age is Teachable Machine for?

Google's pages don't give an age. They say it's for "educators, artists, students, innovators, makers of all kinds". Here's how it tends to go, as our own suggestion:

• Ages 5 to 7: a grown-up runs the mouse; the child is the teacher, choosing the classes and holding up the examples. Keep it to one project.
• Ages 8 to 10: most can run it themselves after one go together. Try audio and pose projects.
• Ages 10 and up: ready to export a model into Scratch and build a game around it.

If your child is younger than that, the explainer AI explained for 7-year-olds covers the same idea (practise, guess, get corrected) in words to read aloud.

8 Teachable Machine projects for kids

These are Teachable Machine examples that work well at home or in class. The first three follow Google's own tutorials.

1. Banana-meter (image). Ripe banana vs green banana. Google's "Bananameter" tutorial does exactly this. Then test it on a spotty one.
2. Snap, clap, whistle (audio). Google's sound tutorial. An audio project also asks for a "Background Noise" class, so record a few seconds of the quiet room first. Your child learns that silence is a sound too.
3. Head tilt (pose). Google's pose tutorial: tilt left, tilt right, straight. Google's FAQ notes that pose tracking notices where you stand in the frame, so stay in one spot.
4. Rock, paper, scissors (image). Three classes, one per hand shape. It's the classic first game to build in Scratch afterwards.
5. Simon says (pose). Arms up, arms out, hands on head. One child trains it, another tries to fool it.
6. Recycling sorter (image). Can, paper, plastic bottle. A good lead-in to AI for Oceans, which asks the same question about rubbish in the sea.
7. Thumbs up, thumbs down (image). A yes/no machine. Then test it with a different person's hand. Does it still work?
8. Fool the machine (any). Google's FAQ suggests some tricks: train with your right hand up and test with your left; train on a real object and test with a drawing of it; train a sound with your voice and test with someone else's. When it fails, ask why, and what examples would fix it.

Project 8 is the most important one. It's Google's own suggested way into bias: a model only knows the examples it was shown, so if the examples all come from one person, one room or one kind of hand, it works worse for everyone else.

How Teachable Machine connects to Scratch

The regular Scratch editor has no Teachable Machine block, so the routes we could confirm use modified versions of Scratch:

• TM2Scratch. A free, open-source extension by Junya Ishihara and Koji Yokokawa, available since 2020. It runs in Stretch3, a modified Scratch 3 at stretch3.github.io, in Chrome. It handles image and sound models; its README points to a sister extension, TMPose2Scratch, for poses. The steps from its README: train a model in Teachable Machine, click Upload my model, copy the link, open Stretch3, add the TM2Scratch extension, paste the link into the model block, then use the "when received label" blocks. The README says that once the model loads, recognising runs locally and camera footage isn't sent to a server. Remember that uploading puts the model (not the pictures) on Google's servers, at a link anyone could use.
• PoseBlocks. Teachable Machine's home page lists an MIT Media Lab activity that uses Teachable Machine with PoseBlocks, a set of Scratch blocks for body, hand and face tracking. We couldn't confirm a current public link to try it, so we're not linking one.
• Machine Learning for Kids. A different free site with its own training and its own Scratch. It doesn't use Teachable Machine at all, but it gets to the same place: train a model, then build a Scratch game with it. Our Machine Learning for Kids guide walks through one project.

For more blocks-based AI tools, see Scratch AI extensions for kids.

For teachers: a quick classroom plan

• 5 minutes: demo the teddy vs car project on the big screen. Let the class pick the two objects.
• 15 minutes: pairs build their own two-class image project. One holds, one records.
• 10 minutes: swap computers and try to fool each other's model.
• 5 minutes: talk. What made a model fail? What examples would fix it?

Google's FAQ also points to the original 2017 version at teachablemachine.withgoogle.com/v1 for a quick demo when you don't need to save anything.

Where Peabody fits

Teachable Machine shows a child how a computer learns from examples. Peabody doesn't teach that. It's a reading site: a child reads stories at their level, some of them written by an AI model that learned from a vast number of examples of writing, then checked by safety screens before anyone sees them. If your child enjoyed teaching a machine, the free AI story generator is a good place to talk about the other side: what happens when AI makes something rather than sorts it. Or just open the library and read one together.

Using Peabody in your homeschool

Teachable Machine makes a good science thread for one homeschool week rather than a weekly lesson. On Monday, read AI explained for 7-year-olds aloud together (about 10 minutes). On Wednesday, do the teddy vs car project above (20 to 30 minutes, with you beside them). On Friday, play "fool the machine" and ask your child to tell you, in their own words, why it failed. Peabody's part is small: that explainer page, and daily reading from the library around it. The honest limit: Peabody has no machine learning lessons and keeps no record of what your child trained, so a photo or a sentence in your own notes is the record. More ideas are in AI in homeschooling and on the homeschool hub.

Learn more elsewhere

• Teachable Machine: Google's tool itself, with three short tutorials on the home page.
• Teachable Machine FAQ: Google's answers on saving, uploading and who can see your samples.
• TM2Scratch: the open-source extension that brings a Teachable Machine model into Scratch.
• Machine Learning for Kids: train a model and build a Scratch game with it, with no account.

About seventy more free AI sites for kids, parents and teachers are in our guide to where to learn about AI.

Questions parents ask

Is Teachable Machine free?

Yes. As of October 2026 Google's Teachable Machine costs nothing and needs no account or login. You only need a Google account if you want to save a project to Google Drive.

Does Teachable Machine send my child's pictures to Google?

Not unless you choose to. Google's FAQ says training runs in your browser tab without sending anything to any servers. Samples leave your computer only if you save the project to Google Drive, which puts them in your own Drive. Uploading a model publishes the trained model, not the pictures.

What age is Teachable Machine for?

Google doesn't set an age. In our experience, children of about 8 and up can run it themselves after one go with a grown-up, and younger children enjoy being the "teacher" who holds up the examples while a grown-up uses the mouse.

Can you use Teachable Machine with Scratch?

Yes, through modified versions of Scratch. The free TM2Scratch extension runs in the Stretch3 editor and loads an uploaded Teachable Machine image or sound model, so a Scratch game can react to what the camera sees or the microphone hears.

What should my child learn from it?

That a computer learns from the examples it's given, and only those. If the examples are few, or all alike, its guesses get worse. That one idea explains a lot about the AI they'll meet everywhere else.