What is AI slop? Meaning, examples and how to spot it
AI slop is low-quality AI content made in bulk. What the word means, where it came from, examples in text, images and video, and the signs that give it away.
Edited by Troy · Published · Updated
AI slop is low-quality content made with generative AI, usually in bulk and with little human care. It looks competent at a glance, costs almost nothing to make and arrives in floods: stock-phrase text, engagement-bait images, cheap video and spam pages. The giveaways are sameness, padding and errors nobody fixed. Not all AI content is slop. Slop is the output nobody bothered to check.
AI slop is low-quality content made with generative AI, usually in large amounts and with little human care. Merriam-Webster, which made "slop" its 2025 Word of the Year, defines it as "digital content of low quality that is produced usually in quantity by means of artificial intelligence." Wikipedia's AI slop article adds that it is usually made "to gain advantage, to earn money, or to deceive people", and that the word carries the same sneer as spam.
Using AI does not make something slop on its own. Simon Willison, the programmer credited with bringing the word into the mainstream, drew the line at care. Not all AI-generated content is slop, he wrote in May 2024, "but if it's mindlessly generated and thrust upon someone who didn't ask for it, slop is the perfect term for it."
What makes something slop
Nobody has a test that sorts slop from everything else. A January 2026 paper by Cody Kommers and colleagues says AI slop "has so far resisted formal definition", then names three properties most of it shares.
- Superficial competence. It looks skilled at a glance, with no substance, craft or intent underneath. The paper's examples are AI work memos with good grammar, and the Shrimp Jesus images that flooded Facebook in 2024, which are photorealistic.
- Asymmetric effort. It comes from a quick prompt and takes a fraction of the work the same thing would take without AI. A paper on slop in software adds the other half: the reviewer always does more work than the generator.
- Mass producibility. It is built for digital production and distribution, to reach a mass audience.
A separate team tried to pin down slop in text. Chantal Shaib and colleagues built a taxonomy from interviews with experts in NLP, writing and philosophy, then had annotators mark slop in news articles and short answers. The judgments were "(somewhat) subjective". Still, text that drifted off topic, carried little information, got facts wrong or leaned on biased language was labelled slop consistently, whatever the source. The authors also say no fully automated way to measure it exists yet.
Slop is cheap to make and costly to read, and there is a lot of it.
Where the word "slop" comes from
Slop is an old insult. Merriam-Webster traces it to "soft mud" in the 1700s and "food waste", as in pig slop, in the 1800s, and from there to "rubbish" or "a product of little or no value."
The AI sense started as in-group slang. Wikipedia traces early uses to 4chan, Hacker News and YouTube commenters reacting to the AI image generators released in 2022. In May 2024 Simon Willison quoted a post by the user deepfates predicting that the way "spam" became the word for unwanted email, "slop" would become the word for unwanted AI-generated content. Wikipedia credits Willison's blog with pushing the term into the mainstream, though he has said it was in use long before him.
The word spread fast from there. Wikipedia says it gained ground in the second quarter of 2024, partly because Google began answering searches with its Gemini model. Merriam-Webster and the American Dialect Society both went on to pick "slop" as their Word of the Year for 2025.
It has since grown spin-offs. "Workslop" is AI-made work that looks finished and isn't. "Microslop" is aimed at Microsoft for building its Copilot chatbot into Windows.
AI slop examples
Slop turns up in every format generative AI can produce. Merriam-Webster's list for 2025 runs from absurd videos and off-kilter advertising images to fake news, junky AI-written books, "workslop" reports and "lots of talking cats." These are the kinds you are most likely to meet.
Text
- Stock phrases. Chatbots lean on the same words and phrases, and each one has its own favourites. Wikipedia's guide to AI writing says one or two of these words may be coincidence, but lots of them, lots of times, is one of the strongest tells. Our AI slop words list sorts them by model, with a receipt for each.
- Leftover chatbot lines. In June 2026 Bill Oliver, a politician in New Brunswick, read a speech aloud that included the sentence "Here's a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points." The video went viral and was described as AI slop.
- Workslop. BetterUp defines it as "AI-generated content that looks good, but lacks substance". In a Harvard Business Review study run with Stanford and BetterUp, 40% of the employees taking part had received some, and each case took about two hours to sort out.
- Fake reports. In January 2026 the curl project closed its bug bounty after a flood of AI-generated vulnerability reports that looked plausible and contained no real findings.
- Books. Amazon sells many AI-generated mushroom foraging books. The New York Mycological Society warned that a wrong identification from one could "literally mean life or death". When author Jane Friedman had to report 29 AI novels using her name in a single week, Kindle limited each author to three new novels a day.
- Travel guides. Willison's own example of slop is a Microsoft AI-generated travel guide that listed the Ottawa Food Bank under "Here's what you shoudn't miss!"
Names
Ask a model to invent a character and it reaches for the same few names. Wikipedia notes a lighthouse keeper named Elias Thorne who shows up in a disproportionate number of stories from several different models. A Cornell study covered by 404 Media sampled 20,000 stories from ChatGPT, Claude, Gemini and Ai2's chatbot, and found the same 11 words, including the names Elias, Mara and Elara and jobs like lighthouse keeper and clockmaker, in more than 88% of them, with little difference between models.
The names leak into the real world. In August 2025 Max Read found no person named Elara Voss anywhere in the United States, and 62 books credited to an "Elara Voss" on Amazon. Our guide to AI slop names covers where these names come from, which model favours which, and how to avoid them.
Images
- Engagement bait. AI image slop spread on social media partly because it earns its makers money on Facebook and TikTok. Wikipedia quotes a Kenyan creator who asked ChatGPT to "WRITE ME 10 PROMPT picture OF JESUS WHICH WILLING BRING HIGH ENGAGEMENT ON FACEBOOK" and fed the results into an image model such as Midjourney.
- Fake news photos. After Hurricane Helene in 2024, influencers shared an AI-generated image of a young girl holding a puppy in a flood as supposed evidence that President Joe Biden had failed to respond.
- Fake events. Willy's Chocolate Experience in Glasgow sold its tickets with AI-generated Facebook ads and no real photos of the venue. The event was held in a sparsely decorated warehouse.
For the tells that still work on AI images, and how the apps we review score on art, read how to spot AI art slop.
Video
- When the video-editing company Kapwing set up a new YouTube account, 104 of the first 500 videos recommended to it were AI slop, the Guardian reported. Kapwing also found 278 channels among the world's most popular that post nothing but AI slop, with more than 63bn views between them and an estimated $117m a year in revenue.
- Kapwing ran the same test on TikTok in May 2026. 59% of the videos a new account saw on its For You page were AI slop.
- A March 2026 New York Times investigation found that around 40% of the videos YouTube recommended to children appeared to be AI slop.
Feeds, search and the open web
- Search spam. Google's spam policy calls it scaled content abuse when many pages are generated mainly to manipulate search rankings. Its first example is using generative AI to make many pages without adding value for users.
- Content farms. In March 2026 the firm DoubleVerify found a network of more than 200 websites, called AutoBait, publishing AI-generated articles and images made mainly to earn ad money.
- Fake listings. In October 2024 thousands of people gathered in Dublin for a Halloween parade that did not exist, after a listings site that used AI-generated content advertised it.
- Music. In July 2025 the Velvet Sundown, a band that had gained over 850,000 Spotify listeners in a few weeks, was accused of being AI-generated. Its Spotify bio later called it "a synthetic music project" made "with the support of artificial intelligence".
Signs of AI slop
No single sign proves anything, so look for several together. Wikipedia's guide to AI writing warns that not all text with these signs is AI-generated, since people wrote the text the models learned from. It also says detection tools beat chance but have "non-trivial error rates", and cites a 2025 study that found people no better than chance at telling AI text from human text.
So judge the effort, not the tool. The question is whether anyone cared about the result.
Signs in text
- Lots of words, little information. The text pads, wanders off the question, or gets facts wrong. In the Shaib study, annotators consistently labelled text like that as slop.
- The same stock words and phrases. One is coincidence. Lots of them, lots of times, is a tell. Paste the text into our AI slop detector for a slop score with the stock phrases highlighted, or check it against the words list.
- Puffed-up importance. LLM writing often ties small details to big trends, so an ordinary fact becomes a turning point.
- Everything in threes. LLMs overuse the rule of three, from three adjectives in a row to three short phrases.
- Lines meant for someone else. Chat leftovers like "I hope this helps" or "Certainly!", or a fill-in-the-blank template that nobody filled in.
- Sources that don't hold up. References whose DOIs don't resolve can be hallucinated. Look up one quote or study and see whether it exists.
Signs in images and video
- Background text. Zoom into distant storefronts, book spines and small print. Reviewers who test image generators report that background text is still where models usually fail.
- Hands under stress. Hands gripping something at an odd angle, overlapping or half hidden still come out with extra or fused fingers more often than not.
- Patterns that drift. Follow a brick wall or tiled floor across the frame. AI patterns that start straight often curve by the far edge.
- Reflections that don't match. A puddle or window that reflects the wrong objects, or the right ones at the wrong angle, is still a strong signal.
- The account behind it. Scroll the page that posted it. Dozens of near-identical images or clips is mass production in action. For video, Kapwing counted clips with obvious AI visuals, and compilation videos with clearly AI-written scripts and AI voiceovers, as slop.
The same reviewers stress that no single check is conclusive. The full checklist, with what no longer works, is in how to spot AI art slop.
Signs in an AI chat or story app
If you are asking whether an app is AI slop, or whether your own app is, run this five-minute test. It touches the same five things our reviews score in text.
- Ask the app to invent a new character and tell you their name. Do it three times.
- Tell it one specific, odd detail about yourself. Ask about it in a new session a day later.
- Read the first line of its last ten replies and count how many open the same way.
- Push the plot once, then stop steering. See whether anything moves.
- Before you pay, find out what one more hour of chatting costs.
Paste a few replies into the AI slop detector for the prose check.
What slop looks like in AI chat and story apps
In a companion, roleplay or story app, slop comes in five common forms. They are the five text axes of our Slop Report, and apps with art get a sixth.
- Prose. Stock phrases that turn up in every app: a voice "barely above a whisper", "a shiver down your spine", eyes that sparkle "with mischief". One of them is fine. One per reply is slop.
- Memory. You mention your sister's wedding on Tuesday. By Friday the character has never heard of her, or remembers a wedding that never happened.
- Characters. Every character shares one voice, and when asked to invent someone new, the model reaches for the same few names, Elias Thorne the lighthouse keeper among them.
- Story. Nothing happens unless you make it happen, or the scene keeps circling back to the same beat.
- Pricing. A message cap that lands mid-scene, or a currency that hides what one more conversation costs.
- Art. Avatars and scene images straight out of the generator, with garbled lettering, melted hands or one face on every character.
How PromptPlay scores it
Every app review carries a Slop Report: a 0 to 100 score for prose, memory, characters, story and pricing, plus art when the app has any, where lower is better. A Fun Meter from 0 to 10 sits next to it, because an app can be sloppy and still fun. The Slop Index ranks every app we have reviewed, and How we test prints the rubric behind every band. If you are leaving Character.AI over any of these, start with our Character.AI alternatives ranking.
Sources
The source text behind the claims above, and the day we read it.
- Receipt 1Reviewer
AI slop is AI-generated content seen as lacking effort, quality or meaning, usually made in high volume.
"digital content made with generative artificial intelligence that is perceived as lacking in effort, quality, or meaning, and usually produced in high volume"
en.wikipedia.org, retrieved
- Receipt 2Reviewer
Wikipedia says slop is usually made to gain advantage, to earn money or to deceive people.
"usually produced in high volume to gain advantage, to earn money, or to deceive people"
en.wikipedia.org, retrieved
- Receipt 3Reviewer
The term carries a pejorative tone similar to spam.
"the term has a pejorative connotation similar to spam"
en.wikipedia.org, retrieved
- Receipt 4Reviewer
Merriam-Webster and the American Dialect Society both picked slop as their 2025 Word of the Year.
"was selected as the 2025 Word of the Year by both Merriam-Webster and the American Dialect Society"
en.wikipedia.org, retrieved
- Receipt 5Reviewer
Merriam-Webster defines slop as "digital content of low quality that is produced usually in quantity by means of artificial intelligence."
"We define slop as "digital content of low quality that is produced usually in quantity by means of artificial intelligence.""
www.merriam-webster.com, retrieved
- Receipt 6Reviewer
Merriam-Webster says slop meant soft mud in the 1700s, food waste in the 1800s, then rubbish or a product of little value.
"The original sense of the word, in the 1700s, was "soft mud." In the 1800s it came to mean "food waste" (as in "pig slop"), and then more generally, "rubbish" or "a product of little or no value.""
www.merriam-webster.com, retrieved
- Receipt 7Reviewer
Merriam-Webster's list of 2025 slop includes absurd videos, off-kilter ad images, fake news, junky AI-written books, workslop reports and talking cats.
"The flood of slop in 2025 included absurd videos, off-kilter advertising images, cheesy propaganda, fake news that looks pretty real, junky AI-written books, "workslop" reports that waste coworkers' time... and lots of talking cats."
www.merriam-webster.com, retrieved
- Receipt 8Reviewer
Early uses of slop for AI material came in reaction to the AI image generators released in 2022.
"Early uses of the term "slop" as a descriptor for low-grade AI material apparently came in reaction to the release of AI image generators in 2022."
en.wikipedia.org, retrieved
- Receipt 9Reviewer
Early uses of slop came from 4chan, Hacker News and YouTube commenters.
"Its early use has been noted among 4chan, Hacker News, and YouTube commentators as a form of in-group slang."
en.wikipedia.org, retrieved
- Receipt 10Reviewer
Wikipedia credits Simon Willison as an early champion of the term in the mainstream.
"The British computer programmer Simon Willison is credited with being an early champion of the term "slop" in the mainstream"
en.wikipedia.org, retrieved
- Receipt 11Reviewer
Simon Willison used the term on his blog in May 2024.
"having used it on his personal blog in May 2024"
en.wikipedia.org, retrieved
- Receipt 12Reviewer
Willison has said the term was in use long before he began pushing it.
"However, he has said it was in use long before he began pushing for the term."
en.wikipedia.org, retrieved
- Receipt 13Reviewer
Willison's May 2024 post quotes deepfates predicting that slop would become the term for unwanted AI-generated content, the way spam did for unwanted email.
"the way that "spam" became the term for unwanted emails, "slop" is going in the dictionary as the term for unwanted AI generated content"
simonwillison.net, retrieved
- Receipt 14Reviewer
Willison wrote that not all AI-generated content is slop, only content mindlessly generated and thrust on someone who didn't ask for it.
"Not all promotional content is spam, and not all AI-generated content is slop. But if it's mindlessly generated and thrust upon someone who didn't ask for it, slop is the perfect term for it."
simonwillison.net, retrieved
- Receipt 15Reviewer
Willison's example of slop is a Microsoft AI-generated travel guide that listed the Ottawa Food Bank.
"Remember that time Microsoft listed the Ottawa Food Bank on an AI-generated "Here's what you shoudn't miss!" travel guide? Perfect example of slop."
simonwillison.net, retrieved
- Receipt 16Reviewer
The term gained popularity in the second quarter of 2024 partly because Google began answering searches with its Gemini model.
"The term gained increased popularity in the second quarter of 2024 in part because of Google's use of its Gemini AI model to generate responses to search queries"
en.wikipedia.org, retrieved
- Receipt 17Reviewer
Microslop is a compound term aimed at Microsoft putting its Copilot chatbot into Windows.
"one such example is "Microslop", used in response to increasing integration of the Microsoft Copilot chatbot into the Windows operating system"
en.wikipedia.org, retrieved
- Receipt 18Reviewer
The Kommers paper was published in January 2026.
"According to an academic article by Cody Kommers and five other scholars that was published in January 2026"
en.wikipedia.org, retrieved
- Receipt 19Reviewer
Kommers and colleagues write that AI slop has so far resisted formal definition.
"AI Slop has so far resisted formal definition."
arxiv.org, retrieved
- Receipt 20Reviewer
The paper names three properties slop tends to share.
"superficial competence, asymmetric effort and mass producibility"
en.wikipedia.org, retrieved
- Receipt 21Reviewer
In the paper, superficial competence means skill on the surface without substance, craft or intent underneath.
"this is belied by a lack of underlying substance, craft, or communicative intent"
arxiv.org, retrieved
- Receipt 22Reviewer
The paper's examples are AI work memos with good grammar and the photorealistic Shrimp Jesus.
"For example, AI-generated work memos have good grammar; even Shrimp Jesus is photorealistic."
arxiv.org, retrieved
- Receipt 23Reviewer
AI-generated Shrimp Jesus images spread on Facebook in 2024.
"AI-generated images of "Shrimp Jesus" proliferated on Facebook in 2024."
en.wikipedia.org, retrieved
- Receipt 24Reviewer
In the paper, asymmetric effort means slop comes from a quick prompt and takes far less work than making it without AI.
"AI Slop is generated with a facile prompt and requires little or no effort of the kind that would be necessary to create such an output without AI."
arxiv.org, retrieved
- Receipt 25Reviewer
In the paper, mass producibility means slop is made to reach a mass audience through digital production and distribution.
"AI Slop is designed to work within a digital ecosystem of production and distribution to reach a mass audience."
arxiv.org, retrieved
- Receipt 26Reviewer
A paper on slop in software says asymmetric effort means the reviewer always does more work than the generator.
"Asymmetric effort means the reviewer always does more work than the generator."
arxiv.org, retrieved
- Receipt 27Reviewer
Shaib and colleagues built a taxonomy of text slop from interviews with experts in NLP, writing and philosophy.
"we develop a taxonomy of "slop" through interviews with experts in NLP, writing, and philosophy"
arxiv.org, retrieved
- Receipt 28Reviewer
The researchers' annotations covered news articles and question-answering passages.
"We now present results from all annotations collected across the news and QA datasets."
arxiv.org, retrieved
- Receipt 29Reviewer
The researchers found slop judgments somewhat subjective, though they track coherence and relevance.
"binary "slop" judgments are (somewhat) subjective, but such determinations nonetheless correlate with latent dimensions such as coherence and relevance"
arxiv.org, retrieved
- Receipt 30Reviewer
Text lacking relevance and information, or with factual errors or biased language, was consistently labelled slop.
"Text lacking relevance and information, or containing factual errors or biased language, is consistently labelled as "slop" across domains."
arxiv.org, retrieved
- Receipt 31Reviewer
The researchers say fully automated, scalable ways to measure slop remain an open problem.
"We highlight that fully automated and scalable methods remain an open challenge."
arxiv.org, retrieved
- Receipt 32Reviewer
Wikipedia's AI writing guide says one or two AI vocabulary words may be coincidence, but many of them, many times, is one of the strongest tells.
"One or two of these words appearing in an edit may be coincidental, but an edit (post-2022) introducing lots of them, lots of times, is one of the strongest tells for AI use."
en.wikipedia.org, retrieved
- Receipt 33Reviewer
The overused AI vocabulary differs by chatbot.
"The distribution of "AI vocabulary" is also somewhat different depending on the chatbot or LLM used."
en.wikipedia.org, retrieved
- Receipt 34Reviewer
Text with these signs is not always AI-generated, because the models are trained on human writing.
"Not all text featuring these indicators is AI-generated, as the large language models that power AI chatbots are trained on human writing, including Wikipedia."
en.wikipedia.org, retrieved
- Receipt 35Reviewer
AI detection tools beat chance but have non-trivial error rates.
"While they perform better than random chance, these tools have non-trivial error rates."
en.wikipedia.org, retrieved
- Receipt 36Reviewer
A 2025 study found people no better than chance at telling AI text from human text.
"a 2025 study has shown that human ability to distinguish LLM text from human is no better than random chance"
en.wikipedia.org, retrieved
- Receipt 37Reviewer
LLM writing often puffs up the importance of its subject by tying it to broader topics.
"LLM writing often puffs up the importance of the subject matter by adding statements about how arbitrary aspects of the topic represent or contribute to a broader topic."
en.wikipedia.org, retrieved
- Receipt 38Reviewer
LLMs overuse the rule of three.
"LLMs overuse the rule of three."
en.wikipedia.org, retrieved
- Receipt 39Reviewer
People sometimes paste chatbot text meant as correspondence or advice into finished writing.
"Editors sometimes paste text from an AI chatbot that was meant as correspondence, prewriting or advice, rather than article content."
en.wikipedia.org, retrieved
- Receipt 40Reviewer
Wikipedia's guide lists "I hope this helps", "Certainly!" and "You're absolutely right!" as words to watch.
"I hope this helps, Of course!, Certainly!, You're absolutely right!"
en.wikipedia.org, retrieved
- Receipt 41Reviewer
Chatbots can produce fill-in-the-blank templates, and some users forget to fill in the blanks.
"However, some LLM users forget to fill in those blanks."
en.wikipedia.org, retrieved
- Receipt 42Reviewer
DOIs that don't resolve and invalid ISBNs can point to hallucinated references.
"Unresolvable DOIs and invalid ISBNs can be indicators of hallucinated references."
en.wikipedia.org, retrieved
- Receipt 43Reviewer
In June 2026 New Brunswick politician Bill Oliver read aloud a chatbot line in a speech.
"On June 9, 2026, New Brunswick politician Bill Oliver gave a speech that included the sentence, "Here's a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points.""
en.wikipedia.org, retrieved
- Receipt 44Reviewer
The video of the speech went viral and was described as AI slop.
"A video of the incident went viral on social media, was described as AI slop, and sparked controversy and calls for him to resign."
en.wikipedia.org, retrieved
- Receipt 45Reviewer
BetterUp defines workslop as AI-generated content that looks good but lacks substance.
"BetterUp defines workslop as "AI-generated content that looks good, but lacks substance"."
en.wikipedia.org, retrieved
- Receipt 46Reviewer
A Harvard Business Review study, run with Stanford and BetterUp, found employees using AI to make low-effort workslop.
"A Harvard Business Review study, done in conjunction with Stanford University and BetterUp, found that employees were using AI tools to create low-effort "workslop" that created more work for their colleagues."
en.wikipedia.org, retrieved
- Receipt 47Reviewer
In that study, 40% of employees had received workslop, and each case took about two hours to resolve.
"it was found that 40% of participating employees received some form of "workslop", with each incident taking an average of two hours to resolve."
en.wikipedia.org, retrieved
- Receipt 48Reviewer
In January 2026 the curl project shut down its HackerOne bug bounty program.
"In January 2026, the curl project shut down its HackerOne bug-bounty program."
arxiv.org, retrieved
- Receipt 49Reviewer
The cause was a flood of AI-generated vulnerability reports that looked plausible and contained no real findings.
"It was a flood of AI-generated vulnerability reports - AI slop - that looked plausible, contained no real findings, and consumed maintainer time until the program became untenable."
arxiv.org, retrieved
- Receipt 50Reviewer
Many AI-generated mushroom foraging books are sold on Amazon.
"On Amazon, there are many AI-generated mushroom foraging books being sold."
en.wikipedia.org, retrieved
- Receipt 51Reviewer
The New York Mycological Society warned that misidentification from these books could mean life or death.
"In 2023, the New York Mycological Society issued a statement about AI-generated mushroom books on social media, stating that misidentification could "literally mean life or death"."
en.wikipedia.org, retrieved
- Receipt 52Reviewer
Author Jane Friedman reported 29 AI novels using her name in one week, and Kindle limited authors to three novels a day.
"Author Jane Friedman stated she had to report 29 novels in one week which were created by AI but used her name and likeness. Kindle created a limit of three novels per day from a single author to combat the issue."
en.wikipedia.org, retrieved
- Receipt 53Reviewer
Several models keep inventing a lighthouse keeper named Elias Thorne.
"a lighthouse keeper named Elias Thorne, who shows up in a disproportionate number of stories generated by multiple LLMs"
en.wikipedia.org, retrieved
- Receipt 54Reviewer
404 Media reports that Cornell researchers Sil Hamilton and David Mimno published the study in late May 2026.
"In late May, researchers Sil Hamilton and David Mimno at Cornell University's Department of Information Science published their paper"
www.404media.co, retrieved
- Receipt 55Reviewer
404 Media reports a Cornell study sampled 20,000 stories from ChatGPT, Claude, Gemini and Ai2's chatbot.
"They sampled 20,000 total stories from OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini, and the Allen Institute for AI's chatbot using five prompts"
www.404media.co, retrieved
- Receipt 56Reviewer
The study found the same 11 words, including Elias, Mara, Elara, lighthouse keeper and clockmaker, in more than 88% of stories, with little difference between models.
"found that the same 11 words-names like Elias, Mara, and Elara, and occupations like lighthouse keeper, clockmaker, and librarian-appear in more than 88% of generated stories, with little difference between models"
www.404media.co, retrieved
- Receipt 57Reviewer
Max Read found no person named Elara Voss in the United States.
"There is no person named Elara Voss in the United States."
maxread.substack.com, retrieved
- Receipt 58Reviewer
Max Read counted 62 books credited to an Elara Voss on Amazon in August 2025.
"There are, as of this writing, 62 books credited to an "Elara Voss" available on Amazon"
maxread.substack.com, retrieved
- Receipt 59Reviewer
AI image and video slop spread on social media partly because it earns money for its creators on Facebook and TikTok.
"AI image and video slop have proliferated on social media in part because it can generate revenue for its creators on Facebook and TikTok"
en.wikipedia.org, retrieved
- Receipt 60Reviewer
A Kenyan slop creator described asking ChatGPT for Jesus image prompts built for Facebook engagement, then feeding them to an image model such as Midjourney.
"a Kenyan creator of slop images described giving ChatGPT prompts such as "WRITE ME 10 PROMPT picture OF JESUS WHICH WILLING BRING HIGH ENGAGEMENT ON FACEBOOK [sic]", and then feeding those created prompts into a text-to-image AI model such as Midjourney"
en.wikipedia.org, retrieved
- Receipt 61Reviewer
After Hurricane Helene in 2024, influencers circulated an AI image of a girl holding a puppy in a flood as supposed evidence that President Biden failed to respond.
"circulated an AI-generated image of a young girl holding a puppy in a flood, and used it as evidence of the purported failure of President Joe Biden to respond to the disaster"
en.wikipedia.org, retrieved
- Receipt 62Reviewer
Willy's Chocolate Experience in Glasgow was sold with AI imagery and no real photos, and was held in a sparsely decorated warehouse.
"misled audiences into attending an event that was held in a sparsely decorated warehouse. Tickets were marketed through Facebook advertisements showing AI-generated imagery, with no genuine photographs of the venue."
en.wikipedia.org, retrieved
- Receipt 63Reviewer
The Guardian reports that Kapwing's new YouTube account was shown 104 AI slop videos among its first 500.
"The researchers also made a new YouTube account and found that 104 of the first 500 videos recommended to its feed were AI slop."
www.theguardian.com, retrieved
- Receipt 64Reviewer
Kapwing found 278 of the world's top YouTube channels contain only AI slop.
"found that 278 of them contain only AI slop"
www.theguardian.com, retrieved
- Receipt 65Reviewer
Those channels had more than 63bn views and an estimated $117m a year in revenue.
"Together, these AI slop channels have amassed more than 63bn views and 221 million subscribers, generating about $117m"
www.theguardian.com, retrieved
- Receipt 66Reviewer
Kapwing found 59% of the videos served to a new TikTok account's For You page were AI slop.
"Some 59% of videos served to a new TikTok account's "For You" page are AI slop"
www.kapwing.com, retrieved
- Receipt 67Reviewer
Kapwing counted videos with obvious AI visuals, and compilation-style videos with clearly AI-generated scripts and voiceovers, as AI slop.
"AI slop videos were defined as those with obvious use of AI-generated visuals, as well as low-quality clip/compilation-style videos with clearly AI-generated scripts and voiceovers."
www.kapwing.com, retrieved
- Receipt 68Reviewer
Kapwing's TikTok data is correct as of May 2026.
"Data is correct as of May 2026."
www.kapwing.com, retrieved
- Receipt 69Reviewer
A March 2026 New York Times investigation found around 40% of the videos YouTube recommended to children appeared to be AI slop.
"A March 2026 investigation by The New York Times found that around 40% of videos recommended to children, both on the main platform and on YouTube Kids, appear to be AI slop"
en.wikipedia.org, retrieved
- Receipt 70Official site
Google defines scaled content abuse as generating many pages mainly to manipulate search rankings rather than help users.
"Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users."
developers.google.com, retrieved
- Receipt 71Official site
Google's first example of scaled content abuse is using generative AI to make many pages without adding value for users.
"Using generative AI tools or other similar tools to generate many pages without adding value for users"
developers.google.com, retrieved
- Receipt 72Reviewer
In March 2026 DoubleVerify found a network of more than 200 websites, called AutoBait, publishing AI content to earn ad revenue.
"In March 2026, researchers at the cybersecurity firm DoubleVerify identified a network of more than 200 websites publishing AI-generated content. The operation, called "AutoBait", used templated prompts in a large language model to produce articles and images designed primarily to generate advertising revenue."
en.wikipedia.org, retrieved
- Receipt 73Reviewer
In October 2024 thousands of people gathered for a Halloween parade in Dublin that did not exist, after a listing on a site that used AI-generated content.
"In October 2024, thousands of people were reported to have assembled for a non-existent Halloween parade in Dublin as a result of a listing on an aggregation listings website, MySpiritHalloween.com, which used AI-generated content."
en.wikipedia.org, retrieved
- Receipt 74Reviewer
In July 2025 media reported accusations that the Velvet Sundown, which gained over 850,000 Spotify listeners in a few weeks, was AI-generated.
"In July 2025, numerous media outlets reported on accusations that an indie band named the Velvet Sundown, which in only a few weeks had amassed over 850,000 listeners on Spotify, was AI-generated."
en.wikipedia.org, retrieved
- Receipt 75Reviewer
The band's Spotify bio was later updated to call it a synthetic music project made with AI.
""a synthetic music project guided by human creative direction, and composed, voiced, and visualized with the support of artificial intelligence""
en.wikipedia.org, retrieved
- Receipt 76Reviewer
Reviewers report background text, such as distant storefronts, book spines and small print, is still where image models usually fail.
"text sitting in the background, on a distant storefront, a license plate, a book spine on a shelf, small print on packaging, is usually still where models fail"
www.opticflux.com, retrieved
- Receipt 77Reviewer
Reviewers report hands gripping objects at odd angles, overlapping or partly hidden still produce extra or fused fingers more often than not.
"Hands gripping an object at an unusual angle, hands overlapping other hands, or hands partially obscured by hair or clothing still produce extra or fused fingers more often than not."
www.opticflux.com, retrieved
- Receipt 78Reviewer
Reviewers report AI-generated repeating patterns like brick and tile frequently drift across the frame.
"AI-generated versions of these patterns frequently drift, a brick course that's straight on the left of the image curves by the right edge"
www.opticflux.com, retrieved
- Receipt 79Reviewer
Reviewers report reflections and shadows that don't match their source are still a strong signal.
"a reflection in a puddle or window that shows different objects, or objects at the wrong angle, than what's actually in frame is still a strong signal"
www.opticflux.com, retrieved
- Receipt 80Reviewer
Reviewers say no single image check is conclusive on its own.
"No single check above is conclusive on its own."
www.opticflux.com, retrieved
Questions people ask
What does AI slop mean?
AI slop means low-quality content made with generative AI, usually in large amounts and with little human effort. Merriam-Webster defines slop as digital content of low quality produced usually in quantity by means of artificial intelligence. Wikipedia adds that it is usually made to gain advantage, earn money or deceive people, and that the word carries a pejorative tone similar to spam.
What are some examples of AI slop?
Chatbot text full of stock phrases, AI-written work reports that look finished and say nothing, AI mushroom foraging books on Amazon, engagement-bait images like Shrimp Jesus on Facebook, YouTube and TikTok channels that post nothing but AI video, and networks of sites that publish AI articles for ad money. Names count too. A lighthouse keeper called Elias Thorne turns up in stories from several different models.
How do you spot AI slop?
Look for several signs at once. In text, check for lots of words with little information, the same stock phrases, puffed-up importance, everything in threes, leftover chatbot lines and sources that don't resolve. In images, zoom into background text, hands gripping things and repeating patterns. No single sign proves anything, and detection tools make mistakes too.
Is all AI-generated content slop?
No. Simon Willison, who helped make the word mainstream, wrote that not all AI-generated content is slop, only content mindlessly generated and pushed on someone who didn't ask for it. Researchers describe slop by its effort and scale. An app where people write or edit what you read can land in our Artisanal or Home-cooked tiers.
Where does the word slop come from?
Slop meant soft mud in the 1700s and food waste in the 1800s, then rubbish. Wikipedia traces the AI sense to 4chan, Hacker News and YouTube commenters reacting to the image generators of 2022, and credits programmer Simon Willison with pushing it into the mainstream on his blog in May 2024. Merriam-Webster and the American Dialect Society both made slop their 2025 Word of the Year.
How can you tell if an AI app is slop?
Test it. Ask it to invent a character three times and see whether the names repeat. Tell it one odd detail and ask about it a day later. Count how many of its last ten replies open the same way, push the plot once and see whether anything moves, and find out what an extra hour costs before you pay.
Slop check: passedThis page went through our slop counter before publishing. How we check our own copy