How do we handle YouTube?

The videos are the problem

Start with the content itself, because for young children it is worse than most parents assume.

When researchers looked at more than 1,600 videos that children eight and under were actually watching at home, one in four was made for older audiences, and one in five carried ads, many of them violent. That was 2020, and it described the human-made version of the problem: an ocean of videos dressed as content for kids, cheap to make, profitable at volume, and mixed in with the good stuff closely enough that the platform’s own defenses could not reliably tell them apart. A peer-reviewed analysis the same year found that a young child browsing from perfectly benign videos was likely to wander into disturbing ones, and that the platform’s countermeasures caught them too slowly to matter.

Then generative AI made the junk close to free. One channel aimed at toddlers posted more than 10,000 videos in seven months, roughly 50 a day. In contrast, Sesame Street has posted about 3,900 in 20 years. Journalists documented AI videos showing infants fed whole grapes and honey, children walking in traffic, and a vowel lesson that displays consonants. Nobody made these in any meaningful sense, and nobody checked them, and YouTube took channels down only after reporters asked.

So no, the fix is not only about what plays next. What plays now deserves your attention, and the younger the child, the more it deserves it.

The machine is the other problem

When a video ends, a system decides what follows, and that system is measured on how long people keep watching. No malice, and no person choosing: an optimizer doing exactly what it was built to do. What it discovers, over billions of attempts, is that more extreme content is stickier than moderate content, louder is stickier than quiet, and a thing that makes you slightly anxious is stickier than a thing that leaves you happy.

Nobody designed that outcome – it happened organically. It is also the delivery system for the junk above: the slop channels exist because the recommendation and search machinery reliably serves their output to children. The two problems feed each other, which is why the answer needs both halves.

What actually works

Choose the sources rather than each video. You cannot preview everything, and you do not have to. Subscribe together to channels you trust, build playlists from them, and make that the whole app as far as the child is concerned. The rule that holds is “we watch channels we choose,” and for a child under seven or so, you preview what goes on the list. The open home page and the search box are where the junk lives.

Autoplay off. This one switch converts watching from something that continues by default into something that requires a decision. It is the difference between a child choosing to watch a fourth video and a child simply not stopping.

Notifications off. There is no version of a YouTube notification that serves your child. Every one of them exists to restart a session that had ended.

Do not rely on the kids’ version of the app to solve this. The disturbing-video research and the AI-slop reporting both cover content that reached children through the supposedly safe front doors. It is better than nothing, and it is still an open catalog with the same incentives, so the choose-the-sources rule applies inside it too.

Where these switches live on each device, from iPad to Roku, is on the device settings page.

What this looks like in daily life

Watch on the biggest screen in the house instead of the smallest. A television in the living room is a social object; a phone under a duvet is not. Same content, but with entirely different behavior.

Sit through some of it. Not all, and not as supervision. This is the richest possible ground for the conversations, because the mechanism is happening live in front of you and you can just point at it. A slop video is a teaching opportunity of its own: older kids can learn to spot them.

When something slips through, and it will, skip the panic. Trace it, name it, and move the rule up a notch if the rule failed.

What to say

When you turn autoplay off:

“I turned off the part that picks the next video. You can still pick

anything from our list. I’d just rather you were the one making the choice.“

When a video feels wrong:

“That felt off, right? It’s possible nobody made that video, and a computer glued it together so you would keep watching. Or it’s possible someone made it feel wrong on purpose. You noticed, which is great because we can make sure that doesn’t happen again.”

When they land somewhere strange:

“How did you end up here? Let’s trace it back.”

Do that without any annoyance in your voice and you get two things: an unexpectedly good archaeology lesson, and a kid who tells you next time.

When they say it is their favorite thing:

“I know. I’m not taking it away; I’m taking away the part that decides for

you.“

When this doesn’t apply

Video is a real interest for a lot of kids, and some of them are learning to make it. A child studying editing, or animation, or how a particular creator builds a story, is doing something worth encouraging. Or a child might be deep-diving on a topic of interest and is truly learning from the material. In these cases, the feed is still the feed, but the watching has a point. Treating it as the same activity as idle scrolling will land as unfair because it is, but be careful that scrolling is supervised and managed.

Households where a screen buys a parent the twenty minutes needed to cook dinner are not failing at anything. The twenty minutes should just come from the playlist you built, with autoplay off.