ChatGPT has an ad pixel now. Here is the kid version.

Last weekend a write-up about ChatGPT's advertising setup collected 646 points and more than 330 comments on Hacker News in twelve hours. That is a lot of attention for a cookie.

Here is what the independent network analysis reported. OpenAI serves advertising measurement code from an ad-collection domain, and it sets a cookie called __obi that is scoped to openai.com, marked to travel across sites, and able to persist for a year. Any business buying ads through ChatGPT can install the measurement code on its own website. The researcher observed 936 advertiser pixels across 1,029 hostnames over several months.

If that sounds familiar, it should. It is the same deployment model as the Meta Pixel and Google's conversion tags, which have been on most of the commercial internet for over a decade.

OpenAI began testing ads inside ChatGPT earlier this year. This is the plumbing that makes those ads measurable. None of it is secret, exactly, and none of it is unusual. That is precisely why it is such a good teaching moment.

Try this version.

Imagine a shop that gives you a free sticker at the door. The sticker has a number on it. You do not think about it again.

Then you walk into a different shop across the street, and that shop has a little scanner by the entrance that reads sticker numbers and sends a message back to the first shop: number 4471 just came in here and bought trainers.

The first shop never followed you. It did not need to. It just handed out a sticker and made a deal with the second shop.

That is a tracking pixel. The cookie is the sticker. The measurement code on the advertiser's website is the scanner. And the deal is that the advertiser gets to find out whether its ad worked, which is the thing it is actually paying for.

Once a kid has the sticker picture, the rest of the conversation is easy, because they now have a mental model that is accurate rather than spooky.

Ask your kid to list every app on their phone they have never paid a cent for. It will be a long list.

Then ask the follow-up: all of these cost money to build and run. Somebody is paying engineers and servers. Where is that money coming from?

There are only about four answers in practice. Someone else pays, usually advertisers. You pay later, once a free tier converts. Somebody rich is funding it for now and will need one of the first two eventually. Or it is a real charity or public service, which is rarer than kids expect.

This is not a lecture about evil companies. Advertising funds an enormous amount of things people like, and a kid who leaves this conversation thinking all ads are sinister has learned the wrong thing. What you want them holding is simpler and more durable: if you cannot see what the product costs, find out who is paying, because somebody is.

That question works on apps, on games with free currency, on influencer posts, and eventually on jobs and contracts.

Here is the part that is actually new, and worth slowing down for.

A search engine sees what you typed in a box. An AI assistant sees your reasoning. Kids use these tools to work through homework they do not understand, to draft messages to friends they are nervous about, to ask questions they would not ask a person. The content is a different category from a search query, and it is often more revealing than anything they would post publicly.

None of the reporting suggests conversation content is being sold. What is being connected is identity and purchase behaviour across sites, which is standard ad measurement. But the underlying point stands and is worth saying out loud to a teenager: the more a tool knows about how you think, the more careful you should be about what you type into it when it is free.

The practical rule we use at home is short. If you would not want it read aloud in class, do not type it into something you did not pay for. It is not a perfect rule. It is a rule a thirteen-year-old will actually remember.

Skip the lecture. Do these instead.

Look at the ad settings together. Most major platforms have a page showing what they think you are interested in. Open one with your kid. The list is usually weird, occasionally accurate, and always more interesting than any explanation you could give. Turn off what you want to turn off while you are in there.

Install a content blocker on the family browser and watch a page load with it on and off. The visible difference is the lesson. Kids who see fourteen trackers blocked on a recipe site do not need the concept explained again.

Price one thing. Pick a free app your kid uses and look up whether it has a paid tier. Ask whether the paid version is worth it to them. This is the one that turns privacy from an abstraction into a purchase decision, which is where it belongs.

None of this requires you to be technical. All three take under fifteen minutes.

The tracking is the part that made Hacker News angry. The ads are the part your kid will actually see, and they deserve their own thirty seconds.

An ad inside a conversation is different from an ad in a sidebar. When a tool has spent ten minutes helping you think through a problem and then suggests a product, the suggestion arrives wearing the same voice as the help. That is a harder thing for a twelve-year-old to filter than a banner.

So teach the tell. Ask your kid to look for the label. Sponsored, promoted, paid partnership, any of those words. Then ask the question that follows: would this tool have suggested this if nobody paid?

They will not always be able to answer. The habit of asking is the point.

If your kid ever sells anything online, they will be on the other side of this within a week of starting.

The first question any ad platform asks a small seller is whether they want to install a pixel so they can see which ads led to sales. Without it you are guessing. With it you know that $12 of ads produced $40 of orders, which is the difference between a business and a hobby with receipts.

So the lesson is not that tracking is bad. It is that measurement is valuable, valuable enough that companies build enormous infrastructure to get it, and that value has to come from somewhere.

A kid who understands both sides of that trade, as a user and as a future seller, is in much better shape than one who has only ever been told to be careful online.

• Independent analysis this month reported an OpenAI advertising cookie, __obi, observed across 1,029 hostnames with 936 advertiser pixels, working like the Meta Pixel.
• The sticker-and-scanner analogy gives kids an accurate mental model in about thirty seconds.
• Teach the question, not the fear: if you cannot see what it costs, find out who is paying.
• AI assistants see reasoning rather than just queries, which raises the stakes on what kids type into free tools.
• Do three concrete things: review ad settings together, watch a page load with a blocker on and off, and price one free app's paid tier.
• Anyone who sells online will install a pixel themselves. Understanding both sides is the actual literacy.

Has OpenAI confirmed this? The findings come from independent network-traffic analysis rather than a company announcement. OpenAI has publicly been testing ads in ChatGPT since earlier this year, so ad measurement infrastructure is expected. Treat the specific technical details as reported rather than confirmed.

Should I stop my kid using ChatGPT over this? That would be an overreaction to something nearly every free service already does. The better response is a conversation about what goes into a free tool, plus a look at the account's privacy settings.

Does a paid subscription remove tracking? Not automatically, and it varies by service. Paying usually removes ads and sometimes reduces data used for ad targeting. Read the specific policy rather than assuming, which is itself a good habit to model.

What age should we have this talk? Around eight for the sticker analogy, and again around twelve or thirteen when they start using AI tools for schoolwork without supervision. The second conversation matters more.

Are private or incognito windows enough? They help with cookies on one device but do not make someone anonymous. Worth telling kids directly, because many of them believe incognito does far more than it does.

1. ChatGPT Ad Collector Tracks User Activity Across Third-Party Advertiser Websites, Cyberpress
2. ChatGPT now knows what you do on other websites via ad collector
3. Hacker News discussion, 646 points, September 20, 2026
4. Why OpenAI Is Putting Ads in ChatGPT and What Comes Next, September 16, 2026
5. ChatGPT, overview and history