Teach a Teen Free Local AI Without the Scam App Maze

By Austin Little

Search "free AI" on a phone and you get a wall of look-alike apps, most wanting a subscription after a three-day trial. There is a better first lesson for a teenager: AI you run yourself, on the family computer, for free, with no account. Here is how to set it up together, and the habits that keep you both out of the scam app maze.

AI disclosure. This page was drafted with AI assistance and edited for Apiary. We don't invent quotes, stats, people, or events.

Why start with local AI

Teens are going to use AI. Most will meet it through whatever app is advertised hardest. That is the worst possible introduction, because it teaches three bad lessons at once: that AI lives inside a slick app you rent, that free means "free trial," and that you should type your life into a box owned by someone you have never heard of.

Running a model locally teaches the opposite lessons:

AI is software, not magic. You download a program and a model file. It runs on your computer. You can see it use memory and disk. It has limits you can observe.
Free can actually mean free. Free local tools such as Ollama do not need a subscription to run models on your own machine.
Your words can stay on your machine. The Ollama FAQ says Ollama does not see your prompts or data when you run locally.
You are the operator. You choose the model, the settings, and when to turn it off.

This is not about avoiding cloud AI forever. A teenager who understands local AI will be much better at judging cloud apps later, because they know what is underneath.

What you need before you start

Be realistic about the computer. Local AI is heavier than a web browser.

A computer, not a phone. The tools in this guide are desktop software. According to Ollama's own pages today:

macOS: the download page and macOS docs say it requires macOS 14 Sonoma or later. The docs mention Apple M-series Macs get CPU and GPU support, and Intel (x86) Macs run on CPU only.
Windows: the Windows docs list Windows 10 22H2 or newer, Home or Pro.
Linux: the download page gives a one-line install command for Linux.

Disk space. The Ollama macOS docs warn that models can be tens to hundreds of gigabytes in size. You do not need to download big ones, but you need room for at least one small model plus some breathing space.

Memory. More memory lets you run larger models.

An hour together. The setup itself is quick. The conversation around it is the point.

If the family computer is old or short on space, that is fine. Skip to the section on safer cloud options, or use the time to learn the scam-spotting habits, which matter regardless.

Part one: the scam app maze, explained honestly

Before installing anything, talk about why you are doing it this way. Teens are sharp, and they respect being told how things actually work.

What the maze looks like

When you search an app store or a search engine for AI, you meet several kinds of products mixed together:

Official apps and sites from the companies that make the models. Real, but often with paid tiers and data collection you should read about.
Wrapper apps. A third party puts its own interface on top of someone else's model, often charges more than the original, and may not say clearly which model it uses or what it does with your data.
Look-alike apps. Names and logos designed to be confused with famous AI brands.
Outright malware. Fake "AI" downloads that install something harmful.

The fourth category is not hypothetical. The FTC published a consumer alert in 2023 titled "Ads for fake AI and other software spread malicious software," describing criminals running bogus ads for AI tools on social media and search engines.

The FTC's simple rule

The same FTC alert gives advice that is easy to teach:

Don't click on ads to download software. If an ad interests you, go to the website by typing the address yourself.
If you don't know the address, search for it, but skip results labeled "Ad" or "Sponsored," because scammers buy search ads too.
Keep your operating system, browser, security software, and phone up to date, with automatic updates on.

Write that first rule on a sticky note. It stops a lot of trouble.

The newer trick: "paste this to prove you're human"

In June 2026, the FTC warned about fake CAPTCHA pages. Per that alert, instead of the usual picture puzzle, the page tells you to press keys like Windows + R, then Ctrl + V, then Enter, which actually pastes and runs hidden malware. The FTC's point is simple: real CAPTCHAs won't ask you to run commands on your device.

This one matters for teens interested in AI, because installing local AI does sometimes involve a terminal. That makes it a perfect teaching moment: you only paste commands that you copied from the official documentation, for software you chose to install. A website demanding that you paste something to "verify" you is a red flag, every time.

Red flags to teach

Here is a list you can go through together. Any one of these means stop and ask:

An app or site that asks for payment details for a "free" trial, especially one that auto-renews.
A download that comes from an ad, a pop-up, a link in a comment, or a direct message.
A site that asks you to paste commands to prove you're human.
An app that wants access to contacts, photos, microphone, or location for no clear reason.
An app name that is almost but not quite a famous brand.
Reviews that all sound the same, or are suspiciously all five stars.
Pressure: "Offer ends in 10 minutes."
Any request to send money, gift cards, or crypto.

And report scams: the FTC's consumer pages point people to ReportFraud.ftc.gov.

Part two: installing Ollama together

Now the hands-on part. Sit side by side. Let the teen drive the keyboard and mouse while you watch and ask questions.

Step 1: type the address yourself

Open a browser and type ollama.com into the address bar. Do not search for it and click an ad. This is the FTC rule in practice. Point out the address bar and check the spelling together.

Step 2: download from the official page

Go to the Download page. Pick your operating system. On macOS and Windows you download an installer. On Linux, the official page shows a command that starts with curl -fsSL https://ollama.com/install.sh.

Here is a good conversation: "We're about to paste a command into a terminal. When is that okay?" The answer: when you copied it yourself from the official site of software you chose, and you understand roughly what it does (this one downloads and runs Ollama's install script). Compare that to the fake CAPTCHA scam. Same action, opposite context.

Step 3: install

Follow the installer. On macOS, the docs describe dragging the app to the Applications folder. On Windows, run the installer. The Ollama docs say both macOS and Windows versions register to start at login during installation, and the FAQ explains how to turn that off if you prefer.

Step 4: pick a small model

Open the Ollama library together. This is a good place to talk about what a "model" is: a big file of learned patterns, made by a company or research group, released for people to download. The library pages list different model families and sizes.

Pick a small one to start. Small models download faster, run better on ordinary computers, and are plenty for learning. Read the model's page together: who made it, what license it uses, what it is designed for.

Step 5: run it

Ollama's quickstart lets you chat through the app, or from a terminal with ollama run followed by the model name. The first run downloads the model, which may take a while. After that, it loads from disk.

Step 6: notice it is local

This is a great moment. Once the model is downloaded, try turning off Wi-Fi and asking a question. If you are running a local model, it still answers. That one demonstration teaches more about how AI works than any article.

Step 7: decide about cloud features

Ollama now also offers cloud-hosted models and web search, which require signing in. The FAQ explains that when using cloud models, Ollama processes your prompts to provide the service, and it documents a way to turn cloud features off entirely, using a setting in a config file or an environment variable, so Ollama runs in local-only mode.

For a teen's first setup, I would keep it local-only. It is simpler to explain, and it keeps the "your words stay on this machine" lesson true. You can always turn cloud features on later as a deliberate choice.

If the terminal feels like too much

LM Studio is another desktop app for downloading and running local models, with a more visual interface. Its homepage today promotes an agent product called Bionic.

Part three: what to actually do with it

A local model sitting unused teaches nothing. Give it real jobs right away.

Good first jobs for a teen

Explain a concept from class in a different way. "Explain photosynthesis like I already understand chemistry but missed this lesson."
Quiz me. "Here are my biology notes. Ask me ten questions, one at a time, and tell me if I'm right."
Check my understanding. "Here's my explanation of the causes of the war. What did I leave out?"
Practice a language. "Have a simple conversation with me in Spanish. Correct my mistakes gently."
Brainstorm. "Give me ten ideas for a science fair project using stuff from a kitchen."
Learn to code. "Explain what this line of Python does."

Jobs to talk about before doing

Writing the essay for you. Have this conversation directly, because it is going to come up. A good family rule: AI can help you understand, plan, and check, but the words you hand in are yours unless your teacher says otherwise. And be honest with teachers about how you used it.

Asking for facts. This is the most important lesson in the whole setup.

Part four: the "it can be confidently wrong" lesson

Local models, especially small ones, make things up. They do not search the internet by default. When they do not know something, they often produce a plausible-sounding answer anyway.

Do this demonstration together:

Ask the model about something you know well. A local place, a family fact, a niche hobby.
Watch it get something wrong while sounding certain.
Talk about it. "It didn't say 'I don't know.' It made something up and said it nicely."

Then set the rule: A model's answer is a draft or a lead. It is not a source. For anything that matters, check a real source: a textbook, a teacher, an official website, a library. This lesson transfers to every AI tool, including the big paid ones.

Topics where the rule is non-negotiable:

Health, medication, mental health crises.
Anything legal.
Money and anything involving payment.
Physical safety: electricity, chemicals, vehicles, tools.
Claims about real people.

And for a mental health crisis specifically, a chatbot is not the right place.

Part five: privacy habits that last

Local AI is a great place to learn privacy habits, because the stakes are lower. Practice them here and they carry over to cloud apps.

Never type these into any AI

Passwords or one-time codes.
Home address, school name plus schedule, or other details that locate you.
Other people's private information: friends' secrets, screenshots of group chats.
Photos of ID cards, bank cards, or documents.
API keys or account recovery codes.

Even on a local model, these habits matter. Once your teen moves to a cloud app, everything typed may be stored, reviewed, or used for training, depending on that service's settings.

Read the settings in any cloud app

When your teen does use a cloud AI service, sit together and look for:

Whether conversations are used for training, and how to opt out.
How to delete history.
Age requirements in the terms of service.

Keep accounts separate

If your teen creates any AI account, use their own email, a strong unique password, and two-factor authentication where offered. Do not share a parent's account, especially one connected to a payment card.

Part six: avoiding the subscription trap

A lot of the "scam app maze" is not illegal. It is just designed to get a card on file.

Teach these:

"Free trial" is not free. It is a countdown to a charge. If you start one, set a reminder to cancel before it renews.
App store subscriptions are managed in the store settings, not in the app. Show your teen where to see and cancel subscriptions on your phone platform.
Do not save a card in a teen's app store account without a spending limit or approval step.
Ask "what does this do that the free local model can't?" Often the answer is "nothing I need."

Paying for an AI service is a reasonable choice for some people. It is never required to learn, and it should be a deliberate decision, not something that happened by tapping "Continue."

Part seven: house rules you can write together

Kids follow rules better when they help write them. Here is a starting draft. Edit it as a family.

We download software only from official sites we typed ourselves.
We never click ads or pop-ups to download anything.
We never paste commands a website asks us to paste to "prove we're human."
We only paste terminal commands copied from official docs for software we chose.
AI answers are drafts and leads, not facts. Important facts get checked.
We never type passwords, codes, addresses, or other people's private info into AI.
Schoolwork follows the teacher's AI rules, and we're honest about how we used it.
No free trials or subscriptions without asking first.
If something feels weird, we stop and ask. No one gets in trouble for asking.

That last rule is the important one. Teens hide mistakes when they expect punishment. If clicking a bad link leads to "thanks for telling me, let's fix it," you will hear about problems early, when they are easy to fix.

Part eight: when local isn't enough

The family computer may not run local models well, or your teen may need web search for research. That is fine. Local AI is the classroom, not the whole school.

When you move to cloud tools:

Type the official address yourself, or install the official app from the platform's store after checking the developer name matches the company.
Start on the free tier. Many services offer free plans. Read the limits on the company's own pricing page, not an ad.
Turn off training data use if the service offers that setting.
Keep the same rules. Everything from the house rules applies.

The difference is that now your teen knows what is under the hood. They have run a model. They have seen it be wrong. They know that "AI" does not mean "a paid app," and that free and private options exist. That knowledge is the real protection against the maze.

A one-page summary for the fridge

Install: type ollama.com yourself, download for your system, install, pick a small model from the library, run it.
Prove it's local: download the model, turn off Wi-Fi, ask a question.
Keep it local: turn off cloud features if you want local-only mode.
Use it well: explain, quiz, brainstorm, practice, learn to code.
Never trust blindly: models can be confidently wrong. Check important facts.
Never paste secrets: passwords, codes, addresses, other people's info.
Never click ads to download. Never paste commands to prove you're human.
Ask first: before any trial, subscription, or weird request.

The bottom line

The safest way into AI for a teenager is not a better app. It is understanding. Install a free local model together, from an address you typed yourself, and use it for real schoolwork and curiosity. Show them it can be wrong. Write house rules together. By the time they meet the scam apps, they will already know what the real thing looks like, and that it never needed their card number.

References

Ollama download: https://ollama.com/download (fetched 2026-10-01)
Ollama macOS docs: https://docs.ollama.com/macos (fetched 2026-10-01)
Ollama Windows docs: https://docs.ollama.com/windows (fetched 2026-10-01)
Ollama FAQ: https://docs.ollama.com/faq (fetched 2026-10-01)
Ollama model library: https://ollama.com/library
FTC, "Ads for fake AI and other software spread malicious software" (April 13, 2023): https://consumer.ftc.gov/consumer-alerts/2023/04/ads-fake-ai-other-software-spread-malicious-software (via search index 2026-10-01)
FTC, "How to spot a CAPTCHA scam" (June 8, 2026): https://consumer.ftc.gov/consumer-alerts/2026/06/how-spot-captcha-scam (via search index 2026-10-01)
LM Studio: https://lmstudio.ai/ (fetched 2026-10-01)