Made for the feed: how AI slop reshapes what we know
by Philip Di Salvo, senior researcher and lecturer at the School of Humanities and Social Sciences, University of St. Gallen
The public conversation about artificial intelligence (AI) and information tends to treat the problem primarily as one of mis- or disinformation. The central risk is usually identified with fakery that fools people: the fabricated image, the cloned voice, or the AI-generated deepfake video circulating online. This understanding of the problem is grounded in a long-established understanding of information disorder: synthetic content can be created with the intent of deceiving audiences or, on a larger scale, polluting the information environment, for instance by pushing a political or propaganda agenda. This framework, still useful in most cases, also assumes something that human beings are the intended audience.
Yet a growing share of what circulates online suggests otherwise. The phenomenon now commonly called ‘AI slop’, a term that appeared online only in the past two years, is low-quality synthetic content, often with no connection to reality or truth whatsoever. This sort of content is now being generated on a massive scale and at impressive speed. What’s new is that it is no longer being produced to inform a human reader, or even to deceive one, but simply to satisfy the distribution algorithms of social media platforms. Its purpose, therefore, seems to be to match the patterns that recommender systems reward and to occupy space in feeds engineered only for engagement. The human encounter with it, as argued by Jason Koebler on 404 Media, is merely incidental to this function.
Content for machines, consequences for people
AI slop of this kind has emerged on social media, pushed at scale by bots, and on the web in the form of entirely synthetic websites created to earn money through spam and online advertising. This understanding allows us to reframe AI slop as an infrastructural problem before a content problem: because it is generated by AI and optimised to please recommender systems, it operates on the mechanisms of circulation themselves. It feeds the algorithms, competes for the visibility they allocate, and so alters what everyone sees – regardless of whether anyone ever engages with a single slop artefact. The harm sits less in any individual piece of content and more in the algorithmic layer that governs what becomes visible at all. Other forms are more malevolent still: their purpose is political rather than merely commercial, going beyond spam and the exploitation of the digital economy.
Recent reporting by University of Oxford’s Reuters Institute for the Study of Journalism on the information war surrounding the US–Iran conflict illustrates this infrastructural problem. Researchers found that platforms do not simply carry synthetic material from producers to audiences; they actively participate in bringing it into circulation and giving it reach, determining what users encounter more than users’ own choices.
The same reporting clarifies that much of this material is not built to pass as authentic: a great deal of AI-generated conflict slop is obviously and explicitly fake and yet circulates anyway, pushing narratives rather than particular factual claims. The question is no longer whether anyone believes an individual image – what matters more is what saturates the field of the visible, regardless of the human viewers that might be deceived. Social media has shaped that field for years under the trend of platformisation; the change emerging now is that generative AI shapes the content itself as well as the field.
The political economy of the feed
Politics has also embraced AI slop and its dynamics in the form of ‘slopaganda’ on social media, mixing classic digital propaganda with the deliberate use of AI-generated, slop-styled imagery and content. Beyond the Iranian conflict, where the Tehran regime showed a particular interest in Lego-styled synthetic war propaganda, the examples are numerous. The New Yorker recently called President Trump the ‘emperor of AI slop’, given his extensive and often outrageous use of such content even on official White House channels. Other politicians are following.
In a recent essay marking two decades of social media scholarship, danah boyd argues that platforms which began as spaces for reciprocal sociality were progressively reorganised around passive consumption of content, because socialising among friends is far less profitable than keeping users scrolling. boyd proposes that we now inhabit an era of ‘parasocial media’ defined by one-directional consumption rather than connection.AI slop is what content looks like once the feed, rather than the friend, becomes the thing it’s made to satisfy.
The consequences for journalism are structural. As Matt Pearce argued recently in Nieman Lab, in an article pointedly titled ‘You couldn’t create a more anti-news internet if you tried’, the monetary cost of content has fallen towards zero while the cognitive cost of finding reliable information has risen sharply, leaving consumers drowning in ‘content sludge’. The infrastructures on which journalism, public communication and humanitarian information all depend are extracting value from the very activities they carry.
The harm that results is a form of epistemic injustice, and an unequally distributed one. The philosopher Miranda Fricker, who coined the term, uses it to describe the wrong done to people in their capacity as knowers: when someone’s testimony is given less credibility because of who they are, or when they lack the means to make their experience understood at all. When an information environment is restructured to serve distribution algorithms, and further complicated by a layer of synthetic content, populations already marginal within these systems are pushed closer to the edge. Information in a humanitarian crisis is a form of aid and protection; degrading the information environment therefore degrades protection itself.
As reported by the Reuters Institute for the Study of Journalism, after a strike that reportedly killed more than a hundred schoolgirls in Minab, Iran, AI-generated images circulated alongside authentic photographs of the victims, casting doubt on real evidence. Mahsa Alimardani of WITNESS, a speaker at CDAC’s Community of Practice on information integrity, described verified images of civilian casualties dismissed as AI-generated on aesthetic grounds alone – the lighting looked too good, the scene resembled a performance – with the dismissals spreading faster than any verification could keep pace. This dynamic is usually called ‘the liar’s dividend’: once anything can be fabricated, anything authentic can be waved away as fabrication. In a crisis, whose images are believed is continuous with whose suffering is acknowledged.
Defending the ecosystem rather than chasing its contaminants
The question facing journalism, then, is not only how to detect synthetic content, but also to keep asking who governs the algorithmic layer that decides what people see, on what logic and in whose interest. Artefact-by-artefact debunking is crucial, but alone it cannot answer that question, least of all now that AI slop adds a further layer of synthetic material that platform algorithms reward and feed, that pollutes the ecosystem by sheer volume and makes the line between real and false harder to draw.
The more durable orientation is to defend the information ecosystem rather than chase its contaminants. Part of this means rebuilding information reward systems that nudge people toward what is civically important, not merely what is engaging. The current online environment has largely abandoned that principle. The real contest is over whether the conditions for discerning truth and falsehood survive at all, and who owns the machinery that decides that. As Jason Koebler put it, AI slop is ‘a brute force attack on the algorithms that control reality’.
Philip Di Salvo is a senior researcher and lecturer at the School of Humanities and Social Sciences, University of St. Gallen (HSG), Switzerland. His primary research interests include investigative journalism, internet surveillance, the intersection of journalism and hacking, black box technologies and information disorder. Previously, he served as a Visiting Fellow in the Department of Media and Communications at the London School of Economics and Political Science (LSE) from 2021 to 2022. From 2012 to 2021, he held various research and teaching roles at the Institute of Media and Journalism, Università della Svizzera italiana (USI). In the summer of 2024, Philip was a Visiting Fellow at Harvard University. Between 2018 and 2020, he also taught as a Lecturer at NABA – New Academy of Fine Arts in Milan, Italy.