One channel posted more videos in seven months than Sesame Street has in twenty years. A fact-checked look at what's actually in young children's YouTube feeds, why it's cheap to make, and the three things a parent can do this week.

Right now, a toddler could be watching a video whose script, pictures and voice were all generated by AI — no writer, no animator, no voice actor. That's not a hypothetical. It's the plain description of a growing share of what autoplay serves up after a nursery-rhyme video ends. This is a fact-checked breakdown of how big that share actually is, who's making it and why, what the (thin) science says about what fast, meaningless video does to a young child's attention, and what a parent can actually do about it — starting tonight.

In early 2026, New York Times reporters spent several weeks watching YouTube the way a toddler's household actually does: starting from popular children's channels like CoComelon, Bluey and Ms. Rachel, then following wherever the Shorts feed led. In one 15-minute session, after a single CoComelon video, more than 40% of the Shorts that followed appeared to contain AI-generated visuals — some clearly marked with YouTube's "altered or synthetic content" label, others just showing the visual glitches and distortions that give the content away on close inspection.

That 40% is a single session, not a platform-wide rate, and the video says so on screen. It's also, per the Times and independent researchers who've since looked, the least alarming number in the story — because what's underneath the number is a fully industrialized way of making children's content that didn't exist three years ago.

This isn't a fringe corner of the platform. Pew Research Center's May 2025 survey of 3,054 U.S. parents found that 85% say their child watches YouTube at all, and 51% say it's a daily habit. It starts younger than most parents probably picture: 62% of parents of children under 2 say that child already watches YouTube, and 35% say it's daily. Kids' video isn't a niche on YouTube — it's one of the platform's largest products, which is exactly why an industrial supply chain has formed around feeding it.

The content itself is where the story stops being abstract. Undark's reporting catalogued specific examples circulating in kids' feeds: a scared child being chased by a T-Rex; a crawling baby biting into an apple that appears to bleed; a child swallowing whole grapes — a major choking hazard for a toddler — and eating honey, which carries a real risk of infant botulism under age one; a teacher eating raw elderberries, which are toxic uncooked. None of these are edge cases scraped from the platform's darkest corner — they're the kind of video that shows up in an ordinary recommendation chain.

The reason this matters more for a toddler than an adult is developmental, not just aesthetic. Carla Engelbrecht, a former producer at Sesame Street and PBS Kids, told Undark: "Mixed signals means you are delaying them learning the cause and effect of a thing. If you learn that red is blue and blue is red, that's a delay." A preschooler can't read an on-screen warning and can't yet distinguish "this is fake" from "this is how the world works" — every frame is treated as information.

Here's the comparison that makes the scale concrete: Sesame Street has published about 3,900 videos to YouTube across its entire 20 years on the platform. A channel called Jo Jo Funland posted more than 10,000 videos in its first seven months — an average of roughly 50 new videos a day, from what functions as a one-person operation. A language model writes the script, a video model generates the pictures, a voice model performs the narration, and scheduling software handles the upload. There is still a person pushing the button, but one person can now run what used to require a studio.

The economics explain why this happened here first. Some video-generation models now charge as little as three cents per generated second — call it roughly ninety cents of raw generation cost for a 30-second Short, before retries, writing, audio or editing. Made-for-Kids content is barred by YouTube's own policy from running personalized advertising (a COPPA compliance requirement), which historically meant lower revenue per view and, in turn, weaker incentive to flood the category with junk. Automation deleted that protection: when production is nearly free, you no longer need a valuable viewer, you need volume. Kids aren't valuable to this business model. They're cheap.

Kapwing — a video-editing software company, not a research lab, which matters for how much weight to put on this — ran its own investigation across the top 100 trending channels in every country it tracked, and found 278 channels, out of roughly 15,000 examined, publishing nothing but AI-generated video. Kapwing estimated those channels together pulled in something like 63 billion views and an estimated $117 million a year in ad revenue. Then Kapwing opened one brand-new YouTube account with no watch history and tracked its first 500 recommended Shorts: 104 of them (21%) were AI-generated, and 165 (33%) fell into what Kapwing categorized as "brainrot" content more broadly. One fresh account isn't the whole platform's recommendation graph — but a fifth of a cold-start feed being AI slop isn't noise, either.

Here's where a lot of content on this topic overstates its case, and this video (and this article) tries not to. No published study has yet isolated what AI-generated video specifically does to a young child's brain. That's not proof it's safe — it means adoption is running well ahead of the evidence, and anyone telling you otherwise is guessing.

What does exist is adjacent evidence about fast, over-stimulating video generally. A 2011 Pediatrics study by Lillard and Peterson randomly assigned 60 four-year-olds to nine minutes of drawing, a slow-paced realistic show ("Caillou"), or a fast-paced cartoon ("SpongeBob SquarePants"), then tested executive function immediately after. Drawing and the slow show produced little difference; the fast-cartoon group performed significantly worse on tests of attention, problem-solving and behavioral control. It's a small study of immediate effects on ordinary television, not AI content specifically — but AI-generated kids' Shorts are frequently faster and more visually chaotic than anything broadcast television ever put in front of a preschooler.

Rachel Barr, a developmental psychologist who directs Georgetown University's Early Learning Project, told the Times that for children under 5 — whose attention systems are still developing — these videos "move too rapidly, and usually aren't long enough to include any meaningful context or story plot." Dr. Jenny Radesky, a developmental-behavioral pediatrician at the University of Michigan, put it more bluntly: "To me, the meaninglessness of these videos is a huge problem because they're just attention capture. And then the worst case is that it's so fantastical and full of attention capture that it is going to be cognitively overloading to the child."

There's a mechanism behind why that matters, and it comes from a different line of research entirely. A 2014 Child Development study by Roseberry, Hirsh-Pasek and Golinkoff taught new verbs to toddlers aged 24–30 months three ways: a person live in the room, a real-time video call, and the identical interaction pre-recorded. Toddlers learned from the live person and from the real-time call. From the recording — same screen, same words — they did not. The one variable that changed was contingency: something on the other end paying attention back. A generated content feed, by construction, never watches back. Not once. That makes it, arguably, the most perfectly non-contingent medium ever built for a toddler's attention.

Roughly 6,700 film and sound jobs disappeared in Los Angeles in a single year, and a 2025 Animation Guild-commissioned study estimated 204,000 entertainment-industry jobs would be significantly disrupted by generative AI within three years. It's important to be precise here, because it's tempting to draw a straight line that the evidence doesn't support: you cannot cleanly attribute those layoffs to AI, and the reporting doesn't claim you can — roles most exposed to AI actually grew over the same period, even as others were cut.

What AI clearly does do is attack the price floor underneath creative work. A human studio cannot produce fifty finished, edited videos a day. The competition a small children's-content creator faces isn't necessarily better than what came before — there's simply an unlimited amount of it, made at a cost no human production can undercut.

Google's AI Futures Fund has invested in Animaj, a children's-media company that got early access to Google's own models, including Gemini and Veo, as part of the deal. Animaj — co-founded by a former Google/YouTube executive — reports 240 million unique viewers a month and roughly 22 billion annual views, and its biggest property is a relaunched version of the classic preschool show Pocoyo, now distributed across more than 100 platforms including Netflix, Disney+, Prime Video and Spotify. Worth being precise here too: those are viewer figures, not a claim that all 240 million are children.

The structural point stands regardless of any one company: the same organization builds the generative models, has funded at least one of the studios racing to use them on children's content, owns the platform those videos are served on, and sells the ads against them. That's not a scandal on its own — it's a description of exactly why nobody in the pipeline has a strong incentive to slow down.

A few tells hold up across almost all of it: read the on-screen text closely — AI gets letters almost right, and a real educational channel doesn't misspell the alphabet. Count fingers and watch eyes across cuts for anatomy that doesn't hold together. Check the upload cadence — fifty videos a day is a script running on a schedule, not a studio. Check the runtime — most of this content is under 30 seconds with no repetition and no story to follow, which is itself a tell for a format built to be cheap rather than to teach. And listen for flat, identical delivery on every line, plus lip-sync that's slightly off.

The single fastest thing a parent can do tonight: turn off autoplay. It won't remove any specific bad video, but it stops one recommendation from silently becoming the next, forever. Pair it with switching YouTube Kids to Approved Content Only, which turns an algorithmic feed into a library you actually control. And where you can, watch with them — contingency is the mechanism the research points to, and an adult who points, reacts and responds is what turns a passive video into something a toddler can actually learn from. For a curated starting point: PBS Kids programming (Daniel Tiger's Neighborhood, Sesame Street), Bluey, Ms. Rachel, The Naptime Show, and curated apps like Sensical or Meevee. A decent working test for any of it: does it send your child back to real-world play afterward, or does it just want another episode?

Turning off autoplay is real, but it's a coping mechanism for an individual household, not a fix for the pipeline. In April, more than 230 organizations and child-development experts — including Fairplay and researchers like Jonathan Haidt — sent Google and YouTube's CEOs a letter with six specific demands: label all AI-generated content, prohibit it on YouTube Kids entirely, prohibit AI "Made for Kids" videos generally, stop recommending AI content to anyone under 18, add a parental toggle that's off by default, and halt further investment in AI video aimed at children. The letter argues AI slop "harms children's development by distorting their sense of reality, overwhelming their learning processes, and hijacking their attention."

The loophole underneath all of this is that YouTube's synthetic-content disclosure requirement applies to realistic content — animated AI content made for a four-year-old generally needs no label at all, and according to Fairplay's own investigation, no such label exists on YouTube Kids at all right now. YouTube's own blog acknowledges the problem by name ("The rise of AI has raised concerns about low-quality content, aka 'AI slop'") and says it's building on existing anti-spam systems to fight it — but that's a reactive posture, not a labeling regime.

There's a precedent for what proactive enforcement looks like. In 2017, during the "Elsagate" crisis over disturbing content aimed at children, YouTube deleted more than 150,000 videos and suspended over 270 accounts within weeks, pulled ads from nearly 2 million videos and 50,000-plus channels masquerading as family-friendly, and roughly doubled its moderation staff. The platform has proven it can purge at volume when it decides the problem is serious enough. Nine years later, the exploit that made Elsagate possible — an under-moderated recommendation system feeding an audience too young to know better — is simply automated now.

The scale is real and verifiable: 85% of American kids watch YouTube, a third of that is daily, and a single automated channel can now out-publish Sesame Street's entire 20-year catalog in seven months, at a production cost measured in cents. No study has proven AI-generated video itself harms a developing brain — but the adjacent science on fast, non-contingent, meaningless video is not reassuring, and adoption is running well ahead of anyone measuring it. Tonight's version of "doing something" is small and concrete: turn off autoplay, switch to Approved Content Only, and watch one video with them. The bigger fix — labeling, and keeping this content off YouTube Kids entirely — is a policy fight, not a parenting one, and it's already underway.

This article discusses developmental research on video content generally; it does not claim a published study has isolated the effects of AI-generated video specifically, and says so above. Some scenes in the accompanying video are AI-illustrated reconstructions, labeled on screen — no footage of real children was used. Disclosure: AI by Age is run by people who work in AI and build AI-related products. We take no AI-vendor sponsorships or affiliate payments. Full disclosure on our About page.