Study Reveals Over 20% of YouTube Recommendations Are AI-Generated “Slop”

• Study scope, definitions, and methodology (October–December 2025)

  A report by Kapwing, a video-editing company, analysed the top 100 trending YouTube channels in every country (approximately 15,000 channels). The researchers identified 278 channels composed entirely of “AI slop”, defined as careless, low-quality content generated using automated applications and distributed to farm views/subscriptions or sway opinion. Kapwing distinguishes AI slop from the related category of “brainrot”, defined as compulsive, nonsensical, low-quality video content (often AI-generated) built around repetition and spectacle that corrodes attention and comprehension. Kapwing’s methodology includes manually sourcing top-trending lists (using playboard.co), then aggregating views, subscribers, and estimated yearly revenue via Social Blade (using midpoint estimates). Data in the report is stated as correct as of October 2025, while coverage and discussion circulated widely in late December 2025.

• Algorithmic exposure to new users

  To simulate a first-time user experience, researchers created a new YouTube account and tracked recommendations. Of the first 500 Shorts shown, 104 videos (21%) were AI-generated slop, while 165 (33%) qualified as brainrot. The first portion of the feed was relatively clean (Kapwing reports being spared slop/brainrot for the first 16 videos) but low-quality AI content appeared with increasing frequency as the feed progressed, consistent with a dynamic recommendation system that learns from minimal signals and then amplifies formats with high retention.

• Estimated prevalence across feeds

  Beyond the single-account test, Kapwing estimates that between 21% and 33% of YouTube feeds may now consist of AI slop or closely related brainrot content. The upper bound aligns with the brainrot share observed in the new-user simulation (33% of the first 500 Shorts), indicating that the broader attention-harvesting category can reach roughly one in three items in short-form consumption environments.

• Scale, reach, and revenue estimates

  The identified AI slop channels collectively amassed 63 billion views and 221 million subscribers, generating an estimated $117 million (£90 million) per year in advertising revenue. Kapwing’s analysis highlights that the most profitable channels largely mirror the most-viewed ones because Social Blade income estimates are view-driven, and many of these channels’ videos were published within the last few months, reinforcing how quickly low-cost, prompt-based production can be scaled into meaningful ad income.

• Global distribution and audience concentration

  AI slop is both globally produced and globally consumed, with strong concentration effects at the country level. Kapwing’s country aggregation finds Spain’s trending AI slop channels have a combined 20.22 million subscribers, the highest total globally, despite Spain having only eight AI slop channels in its top-100 trending set, fewer than Pakistan (20), Egypt (14), South Korea (11), or the United States (nine). Kapwing’s aggregated totals place the United States at 14.47 million combined AI slop subscribers (about 28.4% fewer than Spain) and Brazil at 12.56 million. These patterns indicate that a handful of channels can dominate national subscriber totals even when the overall number of slop channels is relatively small.

• Geographical contrasts in engagement patterns

  While Spain leads in combined subscribers, South Korea stands out in view volume. Kapwing reports South Korea’s trending AI slop channels have accumulated 8.45 billion views, compared with Pakistan (5.34B), the United States (3.39B), and Spain (2.52B). The disparity suggests different regional equilibria: Spain appears more subscriber-heavy (a smaller set of channels with very large followings), while South Korea appears more throughput-heavy (very large cumulative view counts across its trenders).

• Examples of high-performing AI slop channels

  • Bandar Apna Dost (India): Identified as the most-viewed AI slop channel in Kapwing’s report with 2.07 billion views and estimated annual earnings of $4,251,500. The channel publishes hundreds of variations on a near-identical format featuring a realistic monkey in dramatic human-style scenarios; distribution spans beyond YouTube, with around 100,000 followers reported on Instagram and Facebook attribution tied to a “digital creator” identity.
  • Cuentos Facinantes (United States; Spanish-language): Identified as the most-subscribed trending AI slop channel globally with 5.95 million subscribers, and about 1.28 billion views, centred on low-quality, franchise-themed content (including anime-style themes such as Dragon Ball). The channel was established in 2020, but the earliest video still hosted is dated January 8, 2025, signalling heavy recent activity or archival churn.
  • Imperio de jesus (Spain): Cited as a major driver of Spain’s subscriber dominance, with 5.87 million subscribers at the time of analysis—roughly 1.4% fewer than Cuentos Facinantes—using faith-based interactive quiz formats that place religious figures into either/or scenarios designed for rapid looping engagement.
  • Three Minutes Wisdom (South Korea): A central driver of South Korea’s view dominance with 2.02 billion views, nearly a quarter of the country’s trending slop views, and estimated annual ad income of about $4,036,500. Kapwing notes the channel has about 140 videos and typically uses photorealistic(ish) animal footage framed as “wild animals defeated by cute pets,” paired with apparent affiliate linking (including references to Coupang).
  • Pouty Frenchie (Singapore) and The AI World (Pakistan): Illustrative cases of child-targeting fantasy-food narratives and AI-disaster imagery paired with calming ambient soundtracks, reinforcing that the ecosystem spans multiple emotional levers (comfort, shock, cuteness, catastrophe) rather than a single genre.

• Geographic and economic drivers

  Many AI slop creators are based in middle-income countries such as India, Ukraine, Kenya, Nigeria, Brazil, Vietnam, and Pakistan, where even modest YouTube revenue can exceed local median wages. The model is enabled by low production costs (prompt-and-go tooling), rapid iteration, and the ability to scale what performs rather than what is original.

• Platform growth dynamics and structural impact

  Kapwing’s findings reinforce that short-form feeds reward volume and rapid testing. Separate analysis referenced in the overall discussion indicates that nearly one in ten of the fastest-growing YouTube channels globally publish only AI-generated content. Combined with Kapwing’s feed prevalence estimates (21–33%), this supports the conclusion that AI slop is increasingly embedded in growth mechanics, particularly for Shorts-style consumption where repetition and low-context hooks can thrive.

• Platform response, leadership posture, and enforcement limits

  YouTube states that generative AI is a neutral tool and that all content must comply with community guidelines, emphasising a focus on surfacing high-quality content regardless of production method. Kapwing highlights a strategic tension: YouTube leadership has promoted generative AI as a major step-change for creation (YouTube CEO Neal Mohan has compared AI’s impact on video to how the synthesizer changed music) while the platform also faces concerns that advertisers may feel devalued when ads appear alongside repetitive, low-quality slop. YouTube does not release comprehensive data on the share of views attributable to AI-generated content, limiting independent evaluation of policy effectiveness and enforcement consistency.

Why This Matters

The findings show that recommendation algorithms can systematically amplify low-quality, AI-generated content at global scale, particularly for new users. This has direct implications for content quality, child audiences, advertiser confidence, and the long-term health of the creator economy. As generative tools continue to reduce production costs, algorithmic optimisation, not creativity or editorial value, plays an increasingly decisive role in shaping what billions of users see. The risk is not only aesthetic degradation (“spam” video), but also attention erosion and susceptibility to repeated, emotionally tuned narratives that can normalise misinformation-like patterns through sheer volume.

This article was drafted with the assistance of generative AI. All facts and details were reviewed and confirmed by an editor prior to publication.