Yes. The available evidence strongly suggests that AI-generated content on YouTube is increasing rapidly. What is harder to determine is exactly how much of YouTube is AI-generated and, more importantly, how much of that material is low-quality, misleading, or simply fabricated.
YouTube does not publish a comprehensive platform-wide census of AI-generated videos, and automated detection remains imperfect. As a result, there is no defensible statistic such as “25% of all YouTube videos are AI-generated.”
There is, however, enough independent evidence to conclude that synthetic video has moved well beyond the margins of the platform and has become a significant part of what some viewers encounter.
The strongest evidence
One of the most useful attempts to measure the phenomenon comes from video software company Kapwing. Its researchers examined 15,000 prominent YouTube channels—100 from each of 150 countries—and also created a new YouTube account to study what the recommendation algorithm served to an untrained user.
Of the first 500 recommended videos, 104—approximately 21%—were classified as AI slop.
Depending on whether a broader category of extremely low-effort “brainrot” videos is included, Kapwing estimated that roughly 21% to 33% of the initial recommendation feed could consist of AI-generated or similarly low-quality material.
That does not mean that 21% of everything on YouTube is AI slop. A recommendation feed is not a random sample of YouTube’s enormous library.
But it may be a more meaningful measurement from the user’s perspective: it estimates what someone actually encounters while browsing YouTube.
Kapwing also identified 278 prominent channels consisting entirely of AI slop. Collectively, those channels had accumulated approximately 63 billion views and 221 million subscribers.
Those figures demonstrate that synthetic content is not simply an obscure corner of the platform. Fully automated or heavily automated channels can attract audiences measured in billions of views.
Separate reporting by The Guardian found that at one point in 2025 nearly 10% of YouTube’s fastest-growing channels were entirely AI-generated, further suggesting that synthetic content is growing faster than the platform overall.
Academic research finds a smaller but still significant number
More rigorous academic research has produced lower prevalence estimates in specialized categories.
A 2025 study examining 1,082 biomedical educational videos, including 814 YouTube videos, classified approximately 5.3% as probably AI-generated, low-quality content.
That figure cannot be generalized to all of YouTube. Biomedical education is quite different from entertainment Shorts, celebrity videos, history documentaries, or sensational news.
But the contrast between the academic finding and the recommendation-feed study is instructive.
A reasonable interpretation is:
- Specialized informational searches may contain AI slop in the mid-single digits.
- Algorithmically driven recommendation feeds may expose users to substantially more, in at least some circumstances reaching around one video in five.
The prevalence of AI content therefore seems highly dependent on how a person uses YouTube.
YouTube itself is expanding extraordinarily rapidly
The scale of YouTube makes the issue more complicated.
Research firm Omdia estimated that the platform contained approximately 29 billion videos by December 2025, with the total expected to pass 30 billion shortly thereafter. Shorts account for an overwhelming share of new uploads, while Omdia also identified AI-generated material as one of the factors contributing to growth.
Even a relatively small percentage of 29 billion videos represents an enormous quantity of content.
If only 1% were synthetic, for example, that would represent hundreds of millions of videos. But because YouTube does not disclose a reliable percentage, calculations of that sort should be treated as illustrations rather than estimates.
Be skeptical of some of the bigger numbers
Several striking statistics frequently appear in discussions of AI-generated media but should not be used as measurements of YouTube.
One widely repeated prediction attributed to Europol suggested that 90% of online content might be synthetically generated by 2026.
That number does appear in European institutional discussions, but it was a forecast about the entire internet, not a measurement of YouTube. It should not be interpreted to mean that anything approaching 90% of YouTube—or even the current internet—is synthetic.
Similarly, claims that AI-generated content already represents more than half of everything on the public internet are difficult to substantiate using transparent, representative methodologies.
Another frequently quoted statistic says that roughly 42% of YouTube creators use AI tools. Even if broadly correct, that measures something very different. A creator using AI to generate captions, clean up audio, brainstorm titles, or help edit a video is not equivalent to an automated channel producing synthetic documentaries.
That distinction is essential.
There are really three different phenomena:
AI-assisted content: A human creator uses AI as one tool in a conventional production process.
AI-generated content: Significant portions of the script, narration, imagery or video are generated synthetically.
AI slop: Content is produced largely or entirely through automated processes, usually at high volume and with minimal attention to originality, accuracy or quality.
Only the third category is inherently associated with the problem viewers are increasingly noticing.
The more serious problem is videos created “out of whole cloth”
The most concerning development isn’t simply that AI is being used to make videos.
It is that generative AI makes it extraordinarily cheap to produce the appearance of researched factual content without actually conducting the research.
A modern AI production pipeline can generate virtually every component of a video:
topic selection
→ script
→ synthetic narrator
→ images or video
→ editing
→ music
→ thumbnail
→ title and description
→ upload.
The marginal cost of creating another video can become extremely small.
That changes the economics of misinformation.
Historically, producing a convincing 15-minute documentary required writing, recording, editing, acquiring photographs or video, and substantial production time. Even an inaccurate documentary usually represented a meaningful investment of labor.
Today a producer can manufacture dozens—or potentially hundreds—of documentary-looking videos with relatively little human involvement.
More importantly, generative AI can supply missing information rather than merely summarize existing information.
When a model does not know an answer, it may generate a plausible one.
The resulting video can therefore contain invented quotations, imaginary historical incidents, nonexistent experts, fabricated statistics, false biographies or wholly fictitious events—and present them with complete confidence.
The appearance of authority has become cheap
This may ultimately be the most consequential change.
For much of media history, production quality functioned as an imperfect signal of credibility.
A documentary containing professional narration, archival photographs, maps, animated graphics and sophisticated editing was expensive enough to produce that viewers could reasonably assume some organization or group of people stood behind it.
Generative AI breaks that relationship.
A fabricated story can now include:
professional narration,
convincing historical imagery,
animated maps,
apparent archival photographs,
authoritative-looking quotations,
precise dates and statistics,
documentary music,
and polished editing.
All of the traditional visual and auditory signals of authority can be reproduced without there being a reporter, historian, researcher or expert anywhere in the production process.
The problem is therefore not merely more misinformation.
It is a dramatic increase in the ratio of:
apparent authority to actual research.
There is an economic incentive to produce enormous quantities of it
Research into the creator economy also shows an extensive ecosystem teaching people how to monetize generative AI.
Creators share techniques for automating scripting, synthetic narration, imagery and editing, sometimes operating multiple channels simultaneously. Advertising revenue, affiliate marketing and other monetization systems reward successful videos regardless of how much—or how little—research went into producing them.
Traditional media production contains a natural constraint: people cost money.
When most of the production process becomes automated, that constraint largely disappears.
The rational strategy for some operators therefore becomes producing a very large number of inexpensive videos, observing which ones attract algorithmic attention, and producing more variations of whatever succeeds.
Quality becomes secondary to volume.
YouTube’s response is revealing
YouTube’s own actions strongly suggest that the company believes the problem has become significant.
Beginning in 2024, YouTube began requiring creators to disclose realistic altered or synthetic content in circumstances where viewers could mistake it for actual people, places or events.
On July 15, 2025, YouTube revised its monetization policies to clarify that repetitive or mass-produced material was covered by its rules and renamed its “repetitious content” category “inauthentic content.”
In early 2026, YouTube CEO Neal Mohan explicitly identified AI slop as something the platform intended to address.
YouTube subsequently continued refining its policies covering mass-produced and synthetic content while developing automatic systems intended to identify AI-generated material.
The distinction YouTube is attempting to draw is important. The company is not trying to prohibit AI creation—in fact, YouTube itself is introducing AI creation tools.
Instead, it is trying to distinguish between AI used as a creative tool and industrial-scale synthetic production designed primarily to capture clicks, views and advertising revenue.
How large is the problem?
The best available measurements can be summarized this way:
| Measurement | What it actually tells us |
|---|---|
| 21% of 500 recommendations | AI slop encountered in one controlled fresh-account test |
| 21–33% AI slop/brainrot | Broader estimate of low-quality content in that recommendation environment |
| Nearly 10% of fastest-growing channels | AI-only channels achieved significant momentum in a 2025 snapshot |
| 278 AI-slop channels | Large synthetic channels clearly exist at meaningful scale |
| 63 billion views | Such channels can attract enormous audiences |
| 5.3% of biomedical educational sample | AI slop has penetrated substantive informational categories |
| 29 billion YouTube videos overall | Even a small percentage represents enormous absolute volume |
| Percentage of all YouTube that is AI-generated | Unknown |
| Percentage containing fabricated factual information | Unknown |
The most defensible conclusion is therefore not that “one-third of YouTube is AI garbage.”
It is this:
AI-generated video has moved from the margins of YouTube into its mainstream recommendation ecosystem. In at least one controlled test, more than one in five videos recommended to a new user qualified as low-quality AI-generated content. Fully AI-generated channels have accumulated tens of billions of views, and academic research confirms that synthetic low-quality material is appearing even in serious informational categories.
The deeper issue is provenance
The phrase “AI slop” is useful, but it may actually understate the problem.
Bad animation, bizarre AI animals and repetitive Shorts are relatively easy to recognize and ignore.
Much harder to identify is synthetic material that looks credible.
The central challenge is increasingly one of provenance:
Who made this?
What sources did they use?
Did a human being verify it?
Does the person narrating it exist?
Are the photographs authentic?
Did the quoted expert actually say those words?
Did the historical event described in the video ever happen?
Generative AI has made it possible to manufacture the appearance of authority at extraordinarily low cost.
That is why the growth of synthetic content on YouTube matters beyond simple questions of quality. YouTube increasingly contains videos that possess all the visual characteristics viewers traditionally associated with factual media while providing few clues about whether any genuine research exists behind them.
There is not yet enough evidence to say precisely what percentage of YouTube falls into this category.
There is more than enough evidence, however, to say that the problem is real, rapidly growing and already operating at enormous scale.