In December 2025, the Macquarie Dictionary, Merriam-Webster, and the American Dialect Society each independently named "slop" their word of the year — a rare three-way consensus reflecting how widely a single complaint had spread: the internet is filling up with low-effort AI-generated content, and people have noticed.

Confirmed

Multiple independent studies converge on the same basic pattern even though their exact numbers differ. A June 2026 Kapwing investigation analyzing 10,742 TikTok videos across 20 categories found 59% of videos served to a brand-new account's For You page qualified as "AI slop." A separate analysis using Ahrefs and Originality.ai data found 74% of nearly one million newly published web pages tested positive for detectable AI generation, alongside 79% of new images posted to Instagram, TikTok, and Pinterest. A Sprout Social Q1 2026 survey found 56% of consumers report seeing AI slop "often or very often." The trust cost is measurable: content audiences identify as AI-generated sees roughly a 12% engagement penalty on average, half of surveyed Gen Z respondents report unfollowing or blocking an account over suspected AI content, and 88% of consumers say AI video tools have decreased their trust in social media as a news source.

Platforms are responding, unevenly. YouTube's disclosure policy, in force since early 2025, requires creators to manually flag "realistic altered or synthetic content," and CEO Neal Mohan named managing AI slop a top 2026 priority. Meta reported in March 2026 that prioritizing original content in Facebook's Feed and Reels roughly doubled views and watch time on original Reels in the second half of 2025 compared to the same period a year earlier. TikTok has moved toward automated detection and labeling using C2PA content credentials rather than relying solely on creator self-disclosure.

Genuinely uncertain

The headline "how much of the internet is AI now" number is genuinely contested, and readers deserve to know that rather than be handed one confident figure. Depending on the source and measurement method, estimates range from roughly 52% of new web content to as high as 90%, and these studies measure different things — some count any detectable AI assistance, others count wholesale AI generation, and detection methodologies themselves vary in reliability. Treat any single "X% of the internet is AI" statistic with real skepticism; the direction of the trend is well documented, the precise magnitude is not.

It's also unresolved whether AI disclosure labels actually help. Separate research (covered in ATD's companion piece on AI-generated history) found labeling AI images reduced belief in AI-illustrated false claims, but made people more susceptible to false claims illustrated with unlabeled human-made images — meaning the "fix" doesn't uniformly make audiences better at judging truth, it can just redirect their skepticism unevenly.

Worth examining honestly

GHOST PROTOCOL would tell you this was always the business logic: when production cost approaches zero, even minimal engagement generates positive returns through advertising, so flooding the feed is rational even when the content adds nothing. AURORA would push back that the same pressure is producing a real market correction — Meta's original-content boost and the emerging premium advertisers now pay for verified, provenance-tracked content suggest audiences and platforms are already adapting, just more slowly than the flood itself is growing.

The Door This Opens

The choice this door presents isn't whether to use AI tools in content creation — plenty of thoughtful, original work uses them. It's whether the content adds something a viewer couldn't get from a thousand near-identical alternatives generated in the same afternoon. The flood doesn't discriminate between the two. Audiences, increasingly, are starting to.