At the other end of the spectrum, fitness, music, and fashion content remained almost entirely human-made, each below two percent.

One hashtag in particular illustrates the scale of the problem. Within #CartoonKids, 97 out of 100 videos checked were AI-generated, leaving just three that appeared to be made by humans. Related tags were nearly as bad: #cartoons and #babysong both hit 83 percent, and #forkids reached 79 percent.

The formula behind these videos is recognisable to anyone who has stumbled across them. Familiar cartoon characters appear in bizarre scenarios, educational lessons are riddled with factual errors, characters speak in synthetic voices, and animations shift and morph in ways that do not quite make sense. A counting lesson that gets the numbers wrong may seem absurd to an adult, but a preschooler does not have the context to notice.

Dr Dana Suskind, a professor of paediatrics at the University of Chicago, described the phenomenon as “toddler AI misinformation at an industrial scale,” according to reporting on the study. The concern is not just that individual videos are bad, but that generative AI enables the creation of endless streams of them at a pace no human creator could match.

The problem extends well beyond content aimed at children. Educational, science, health, and history videos were among the categories most heavily saturated with AI-generated material, which is particularly damaging because those are the topics where accuracy matters most.

A poorly generated comedy skit is easy to scroll past. A history lesson filled with fabricated details or a health video presenting misleading advice is a different kind of failure.

TikTok's recommendation engine is designed to adapt quickly, using signals like watch time, likes, follows, and scrolling behaviour to personalise what each user sees. But Kapwing's research focuses on what happens before that personalisation kicks in, when a new account has provided no behavioural data and the algorithm is essentially guessing.

The result is that AI slop has become TikTok's default first impression. For a platform that built its growth on the strength of its recommendation algorithm, that is a significant problem.

TikTok is not unaware of the issue. The company introduced controls in November 2025 that allow users to increase or decrease the amount of AI-generated content in their feeds, and it has invested in AI literacy initiatives. Kapwing argues those passive controls are not enough, and the data suggests the measures have not meaningfully reduced the volume of AI slop reaching new users.

The platform also faces growing legal pressure over its handling of children's content. Florida sued TikTok earlier this month under its child social media law, alleging the platform let minors onto the app and misled parents about the content available to them. The AI slop findings add another dimension to the regulatory scrutiny: even when children are on the platform legally, the content they encounter may be overwhelmingly low-quality and machine-generated.

The comparison with YouTube is instructive. Kapwing found that roughly 21 percent of YouTube Shorts recommended to a new account were AI slop, less than half the TikTok rate. YouTube has taken a more aggressive enforcement approach, terminating 16 channels with a combined 35 million subscribers and nearly five billion lifetime views under its inauthentic content policy in January 2026.

That crackdown has drawn criticism for catching legitimate faceless creators in the crossfire, but the gap between platforms remains stark.

The broader pattern is consistent across social media. AI-generated content is flooding music streaming platforms as well, where services like Deezer now flag more than 20,000 fully AI-generated tracks per day.

The incentive structure rewards volume over quality: if a creator or bot operator can produce dozens of videos in the time it once took to make one, platforms become saturated with content that is technically watchable but offers little substance.

It is worth noting the study's limitations. Kapwing is a video editing tool company with a commercial interest in human-created content, and the classification of what counts as “AI slop” involved manual review rather than automated detection. The researchers built a seed list of 20 popular TikTok categories, selected at least three popular tags for each, and reviewed videos for obvious AI-generated visuals and scripts, a methodology that is transparent but subjective.

The study also captures a snapshot from May 2026, and TikTok's algorithm and moderation policies could change. The platform has not publicly disputed the findings.

Still, the scale of the data, more than 10,000 videos across 20 categories plus the 500-video new-account test, makes it the most comprehensive examination of AI content density on TikTok published so far. And the children's content findings are difficult to dismiss regardless of methodology: when 97 out of 100 videos in a kids' hashtag are machine-generated, the precise definition of “slop” matters less than the fact that virtually nothing in that feed was made by a human.

Social media became popular because it offered something distinctly human: creativity, personality, expertise, and connection. AI can imitate all of those things with increasing skill, but imitation is not the same as authenticity.

When nearly six out of every ten videos a new user sees are AI-generated, the question is no longer whether AI slop exists on TikTok. The question is whether it has become a defining feature of the platform.