AI Slop
AI slop is low-quality digital content mass-produced by generative AI with minimal human curation. Merriam-Webster named it the 2025 Word of the Year.
AI slop is digital content of low quality that is produced in quantity by means of generative artificial intelligence. The term carries a pejorative connotation, similar to spam. Merriam-Webster selected “slop” as its 2025 Word of the Year, and the American Dialect Society made the same choice, reflecting how the word entered mainstream vocabulary as AI-generated content flooded social media, search results, and online publishing.
Origin and definition
The word “slop” dates to the 1700s, originally meaning soft mud or something of little value. The AI-specific sense emerged in the early 2020s. A poet and technologist writing under the name “deepfates” is credited with popularizing the term in 2024, describing it as “the term for unwanted AI-generated content.”
Merriam-Webster’s official definition: “digital content of low quality that is produced usually in quantity by means of artificial intelligence.” The dictionary announced the selection on 15 December 2025, noting that the word captured “all that stuff dumped on our screens, captured in just four letters.”
What makes content “slop”
AI slop is defined by the intersection of three properties:
Low quality. The content lacks substance, accuracy, or creative merit. It may be technically fluent but reads as filler: generic, hedged, repetitive, or factually unreliable.
High volume. Slop is mass-produced. A single operator can generate thousands of articles, images, or videos per day at near-zero marginal cost.
Minimal curation. The creator invests little effort in reviewing, editing, or verifying the output before publication. The goal is attention capture or monetization, not quality.
Content that is AI-generated but carefully reviewed, edited, and verified is not typically called slop. The term targets the output of publish-first-verify-never workflows.
Where AI slop appears
- Social media. AI-generated images of fake emergencies, sentimental stories, and engagement-bait posts designed to harvest likes and shares.
- Search results. SEO-optimized articles generated at scale to capture traffic, often with thin or inaccurate information.
- E-commerce. Fake product reviews, auto-generated listings, and synthetic customer testimonials.
- Publishing. Books and articles written entirely by AI and sold without disclosure, sometimes containing hallucinated facts or plagiarized passages.
- News and information. Fabricated news sites, AI-generated misinformation, and synthetic content designed to look authoritative.
AI slop in numbers
How much content is AI-generated depends on what is measured and who measures it. Most figures come from companies that sell detection or content tools, and every detector has a false-positive rate. Read each number with its method. As of October 2026:
| Figure | What it measures | Source |
|---|---|---|
| 49.9% | New English web articles that are primarily AI-generated, Q1 2026, after a peak of 50.9% in Q4 2025. Random sample of 55,400 Common Crawl URLs (January 2020 to March 2026), averaged over three detectors (Pangram, Copyleaks, GPTZero), each with a false-positive rate under 2% | Graphite, 15 May 2026 (an SEO agency) |
| 14% | AI-generated share of articles ranking in Google search, from 31,493 keywords collected in June 2025; 18% of articles cited by ChatGPT and Perplexity | Graphite, 14 October 2025 |
| Nearly 90,000 a day | Fully AI-generated tracks delivered to Deezer, “more than 50% of total daily deliveries on peak days” in June 2026, yet only 1 to 3% of streams; Deezer says up to 85% of streams of fully AI-generated tracks in 2025 were fraudulent | Deezer, 21 July 2026 (self-reported, own detector) |
| 59% | AI slop among the first 500 videos in a new TikTok account’s For You feed; 57.4% of 2,000 featured videos in the Kids category; 10,742 videos classified by hand | Kapwing, 9 June 2026 (an AI video tool vendor; one test account) |
| 53% | Bots as a share of all web traffic | Thales 2026 Bad Bot Report, 5 May 2026 (bot-mitigation vendor) |
The article figure and the music figure describe what is produced; the search, citation and stream figures describe what people actually see. In Graphite’s studies, AI content was about half of new articles but 14% of what ranked in Google, although the two measurements were taken about a year apart; at Deezer, up to half of uploads on peak days became 1 to 3% of streams. Inside organisations, the same problem has its own name, workslop .
Why it matters
AI slop degrades the information environment. When low-quality content floods platforms, it becomes harder for users to find reliable information and harder for legitimate creators to compete for attention. The problem compounds: as AI models train on web data that increasingly contains AI-generated slop, they risk model collapse , a progressive degradation of quality and diversity.
A June 2026 convening at Columbia University’s Institute for Global Politics documented how AI slop threatens research integrity and public discourse. Participants noted that the definitional traits converge at the intersection of low quality, mass scale, and varying creator intent.
Distinguishing slop from legitimate AI use
Not all AI-generated content is slop. The term specifically targets:
| Slop | Not slop |
|---|---|
| Unreviewed, mass-published | Human-edited and verified |
| Designed for engagement farming | Designed to inform or create value |
| No disclosure of AI use | Transparent about AI assistance |
| Factually unreliable | Fact-checked before publication |
Using AI as a drafting tool, then reviewing and refining the output, is a workflow. Publishing thousands of AI-generated articles without reading them is slop.
Sources
- Merriam-Webster, “Word of the Year 2025” (15 December 2025): https://www.merriam-webster.com/wordplay/word-of-the-year
- Associated Press, “‘Slop’ is Merriam-Webster’s 2025 word of the year” (15 December 2025): https://apnews.com/article/2dffb2379cac6001aa30e148669e3393
- CBS News, “‘Slop’ chosen as Merriam-Webster’s 2025 word of the year” (15 December 2025): https://www.cbsnews.com/news/slop-merriam-webster-2025-word-of-the-year/
- PBS NewsHour, “Merriam-Webster’s word of the year for 2025 is AI ‘slop’” (15 December 2025): https://www.pbs.org/newshour/nation/merriam-websters-word-of-the-year-for-2025-is-ais-slop
- Scientific American, “AI Slop - How Every Media Revolution Breeds Rubbish and Art” (2025): https://www.scientificamerican.com/article/ai-slop-how-every-media-revolution-breeds-rubbish-and-art/
- Columbia Institute for Global Politics, “AI Slop and the Information Ecosystem: Insights from a Cross-Sector Convening” (June 2026): https://igp.sipa.columbia.edu/news/ai-slop-and-information-ecosystem-insights-crosssector-convening
- Graphite, “AI Now Writes as Many Online Articles as Humans” (15 May 2026): https://graphite.io/five-percent/research/ai-now-writes-as-many-online-articles-as-humans-do
- Graphite, AI content in search and LLM citations (14 October 2025): https://graphite.io/five-percent/ai-content-in-search-and-llms
- Deezer, newsroom release on AI music exceeding 50% of daily uploads (21 July 2026): https://newsroom-deezer.com/2026/07/ai-music-exceeds-50-percent-daily-uploads-deezer/
- Kapwing, “The TikTok AI Slop Report” (9 June 2026): https://www.kapwing.com/resources/the-tiktok-ai-slop-report/
- Thales, “The 2026 Bad Bot Report is here” (5 May 2026): https://community-cybersec.thalesgroup.com/blogs/tim-chang/2026/05/05/the-2026-bad-bot-report-is-here
Further reading
- Model collapse : what happens when AI trains on AI-generated content - progressive quality degradation.
- Dead internet theory confirmed : bots now generate 57% of web traffic, accelerating the slop problem.
- Workslop : the workplace version, AI output that looks finished and shifts the work to the receiver.
- AI washing : a related phenomenon where companies overclaim AI capabilities.
- Hallucination : when AI generates plausible but false information.
- Slopsquatting : when attackers exploit AI-hallucinated package names to distribute malware.
- Synthetic data : the controlled use of AI-generated data in training.