{
  "id": "ai-slop",
  "title": "AI slop",
  "lang": "en",
  "source_lang": "en",
  "summary": "AI slop is a disparaging term for low-quality AI-generated content, often mass-produced and distributed with little editorial oversight.",
  "category": "computer-science",
  "subcategory": "artificial-intelligence",
  "type": "concept",
  "aliases": [
    "Slop",
    "AI-generated junk",
    "AI garbage"
  ],
  "generated_at": "2026-10-15",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/ai-slop/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/ai-slop/",
      "json": "https://aiwiki.page/zh/ai-slop.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "AI slop is a disparaging term for low-quality digital content produced using generative artificial intelligence, commonly in large quantities and with little selection, verification, or editing. It encompasses text, images, audio, and video circulated through social media, websites, and other publishing channels. The expression combines a description of how content is produced with a judgment about its value; it is not a neutral synonym for all AI-generated material. Merriam-Webster’s definition emphasizes low quality and production at scale, while other formulations emphasize unwanted distribution and inadequate human review. (globenewswire.com)"
    },
    {
      "heading": "Meaning and boundaries",
      "text": "There is no universally accepted technical definition of AI slop. In ordinary usage, the label can refer to repetitive prose, implausible imagery, misleading material, or content that contributes little beyond attracting attention. A 2025 study of AI-generated text found that binary judgments of “slop” were partly subjective, although they correlated with more specific characteristics such as coherence and relevance. Consequently, AI origin, factual accuracy, and perceived quality are distinct questions rather than interchangeable criteria. (arxiv.org)\n\nAn influential narrower definition was advanced by programmer Simon Willison: AI-generated material that is both unsolicited and unreviewed. Under this interpretation, a response deliberately requested from an AI system is not necessarily slop, nor is a publication whose AI-assisted contents have been checked by an accountable editor. The concern is indiscriminate publication rather than the mere presence of artificial intelligence in a creative or editorial workflow. (simonwillison.net)\n\nThe term overlaps with, but is not identical to, several other concepts:\n\n- Spam: emphasizes unwanted or manipulative distribution. AI slop emphasizes low-value generated material, which may circulate through ordinary recommendations rather than direct messages.\n- AI hallucination: concerns unsupported or fabricated output. Slop need not contain a factual claim at all; an image or repetitive entertainment video can receive the label.\n- Search manipulation: concerns attempts to influence search engine rankings. Low-quality generated pages can serve this purpose, but comparable abuse can also be carried out by humans. Google explicitly adopted this production-method-neutral distinction in its March 2024 spam-policy changes. (simonwillison.net)"
    },
    {
      "heading": "History and adoption",
      "text": "The expression gained visibility during public discussion of mass-produced AI content in 2024. On May 8, 2024, Willison published “Slop is the new name for unwanted AI-generated content,” advocating a concise term analogous to spam. His post did not claim sole authorship of the expression and acknowledged earlier uses. By June 2024, he was discussing it as a term beginning to receive mainstream acceptance. (simonwillison.net)\n\nOn December 15, 2025, Merriam-Webster selected slop as its Word of the Year, defining its AI-related sense in terms of low-quality digital content typically produced in quantity. The American Dialect Society subsequently selected slop as its own 2025 Word of the Year, noting that the AI context could increasingly be understood without the prefix AI. These selections document the expression’s linguistic prominence, not a measurement of the amount of such content online. (globenewswire.com)"
    },
    {
      "heading": "Forms and distribution",
      "text": "Text and search-oriented publishing\n\nTextual slop includes generated pages whose apparent purpose is to capture search traffic rather than answer readers’ questions. Google’s March 2024 announcement described large-scale publication of unoriginal, low-value pages, including pages that appear to answer popular searches but fail to provide useful information. Its category of “scaled content abuse” covers automated production, human production, and mixtures of the two. This distinction separates the abusive publishing practice from any particular model or writing tool. (blog.google)\n\nTextual quality is also multidimensional. Fluent wording does not by itself establish coherence, relevance, or informational value. Research on measuring textual slop therefore examines identifiable properties rather than treating an AI-origin label as a complete quality assessment. (arxiv.org)\n\nImages and engagement-oriented posts\n\nA prominent visual example is “Shrimp Jesus,” a family of generated images combining Christian religious imagery with shrimp or other crustaceans. Research by the Stanford Internet Observatory examined these alongside more mundane synthetic pictures of homes, household objects, and purported children’s artwork. Some posts presented generated objects or achievements without clearly disclosing their synthetic origin. (fsi.stanford.edu)\n\nThe researchers studied 120 Facebook pages that had each posted at least 50 AI-generated images. They identified spam pages directing audiences toward external content farms, scam pages promoting apparently nonexistent products or soliciting personal information, and other creator pages. This was a study of a selected set of pages, not an estimate of the prevalence of AI slop across Facebook or the entire internet. (fsi.stanford.edu)\n\nThe label also extends to generated videos and advertising imagery. Its breadth reflects a shared judgment about low quality or indiscriminate production rather than a single visual style or file format. (globenewswire.com)"
    },
    {
      "heading": "Incentives and information quality",
      "text": "The circulation of AI slop is closely connected to incentives for traffic and engagement. In the Stanford study, synthetic images helped attract audiences that could subsequently be directed toward low-quality websites or commercial schemes. Some images reached users who did not follow the publishing pages. The researchers proposed engagement-oriented feed ranking as a possible explanation for this distribution, rather than demonstrating that mechanism conclusively. (fsi.stanford.edu)\n\nThese cases illustrate why conspicuously implausible imagery can still serve a practical purpose: attracting attention does not require convincing every viewer that an image is real. At the same time, comments on more ordinary-looking images suggested that some viewers did not recognize their synthetic origin. Problems of quality, disclosure, and deception therefore overlap without being identical. (fsi.stanford.edu)"
    },
    {
      "heading": "Training-data feedback",
      "text": "A separate concern arises when generated online content enters training datasets for subsequent models. Research published in Nature in July 2024 demonstrated model collapse under recursive training conditions: successive models trained on generated outputs could lose information about the original data distribution, particularly less probable events. This is a technical phenomenon, not simply another name for low-quality internet content. (nature.com)\n\nThe result does not establish that all synthetic training data are harmful. Other experiments found that retaining original real data while accumulating synthetic examples could avoid collapse in the tested settings. The relevant distinction is therefore not merely “human versus AI,” but how data are selected, retained, combined, and evaluated. (arxiv.org)"
    },
    {
      "heading": "Identification and platform responses",
      "text": "Identifying AI-generated material and deciding whether it is slop require different judgments. Provenance records and digital watermarks can supply evidence about origin or editing history, but origin alone does not establish usefulness, truth, or artistic merit. The US National Institute of Standards and Technology has examined watermarking, provenance metadata, automated detection, and human-assisted detection as complementary approaches with different limitations and evaluation requirements. (nvlpubs.nist.gov)\n\nPlatform responses can instead target distribution practices and informational value. Google’s 2024 policy changes, for example, focused on large-scale production intended to manipulate rankings, regardless of whether the content was generated automatically. Such an approach addresses abusive publication without assuming that every AI-assisted work is low quality or that every low-quality work is AI-generated. (blog.google)"
    }
  ],
  "links": [
    "generative-artificial-intelligence",
    "social-media",
    "artificial-intelligence",
    "ai-hallucination",
    "search-engine",
    "facebook",
    "internet",
    "training-data"
  ],
  "sources": [
    {
      "title": "Merriam-Webster Announces “Slop” as the 2025 Word of the Year",
      "url": "https://www.globenewswire.com/news-release/2025/12/15/3205236/0/en/merriam-webster-announces-slop-as-the-2025-word-of-the-year.html"
    },
    {
      "title": "Slop is the new name for unwanted AI-generated content",
      "url": "https://simonwillison.net/2024/May/8/slop/"
    },
    {
      "title": "Open challenges for AI engineering",
      "url": "https://simonwillison.net/2024/jun/27/ai-worlds-fair/"
    },
    {
      "title": "Measuring AI “Slop” in Text",
      "url": "https://arxiv.org/abs/2509.19163"
    },
    {
      "title": "2025 Word of the Year Press Release",
      "url": "https://americandialect.org/wp-content/uploads/2026/01/2025-Word-of-the-Year-PRESS-RELEASE.pdf"
    },
    {
      "title": "New ways we’re tackling spammy, low-quality content on Search",
      "url": "https://blog.google/products-and-platforms/products/search/google-search-update-march-2024/"
    },
    {
      "title": "How Spammers, Scammers and Creators Leverage AI-Generated Images on Facebook for Audience Growth",
      "url": "https://fsi.stanford.edu/news/ai-spam-accounts-build-followers"
    },
    {
      "title": "AI models collapse when trained on recursively generated data",
      "url": "https://www.nature.com/articles/s41586-024-07566-y"
    },
    {
      "title": "Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data",
      "url": "https://arxiv.org/abs/2404.01413"
    },
    {
      "title": "Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency",
      "url": "https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-4.pdf"
    }
  ],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}