Beware of AI Slop
Have you ever seen a design on a street-side MSME food banner created using artificial intelligence (AI), where the image of nasi uduk looks more like round white beads? Or perhaps, next to the fried catfish and fresh vegetables, something looks suspiciously like a cockroach?
Or, a banner promoting a private school depicting a student, but with only four fingers on the right hand? That is ‘AI Slust’.
The internet, and the real world, are being flooded with such content. The term ‘AI Slop’ refers to the massive volume of content created using text, image, and video generators, often characterised by low quality and limited informational value. It consists of haphazardly produced images and text.
On social media platforms—such as Instagram, X, YouTube, and most notably Threads—AI-generated content with similar patterns is circulating widely. On Threads, which is often compared to the early era of Twitter, there is no shortage of AI-generated content sharing ‘personal experiences’ with easily recognisable patterns. While the ideas within the content might provide some new information, many are interspersed with AI hallucinations. Consequently, information that should be useful is being called into question: is this real, or merely an AI hallucination?
Interestingly, because there is so much of this content, netizens have even created diagrams of AI sentence patterns to recognise it. For example, content that tells a story about someone chatting with a friend or relative about a new discovery, only for the ‘sender’ or thread creator to ‘fall silent’ upon realising something. This is followed by a long, drawn-out explanation designed to drive engagement.
In longer social media articles, AI Slop is also easily identified by generic sentences, cliché structures, and repetitive phrases. The conclusions are also uniform, often ending with rhetorical questions or summaries that add nothing new.
Research from the SEO tool Ahrefs in 2025 noted that 74.2 per cent of newly published webpages contain AI-generated material. Furthermore, an analysis of the web corpus by Professor Dirk Spennemann from Charles Sturt University, Australia, estimates that 30 to 40 per cent of active text on the internet now originates from AI-generated or AI-edited sources. This provides a clear picture that AI-based content production has become a significant part of the digital information ecosystem.
The Reason
Why is AI Slop flooding our information streams? AI generators make content production far cheaper and faster. Simultaneously, many digital platforms like YouTube, Instagram, and X still operate on systems that reward accounts that upload large volumes of content. The more material produced, the greater the chance of content appearing before users. This creates a powerful incentive to prioritise quantity. Content is made to be easily produced, easily found by search engines, and just engaging enough to achieve algorithmic distribution. Depth of information becomes an afterthought.
Google has responded to this issue through several algorithm updates between 2024 and 2026. One of the primary targets is the practice categorised as large-scale content production created specifically to drive traffic. Ultimately, however, humans can recognise AI patterns. Sites heavily reliant on mass AI content production have reportedly experienced significant drops in organic traffic following these changes.
This is where the issue of AI Slop touches upon a larger problem: trust. The Khazanah Research Institute Malaysia, in its study series on AI Slop and model collapse, identifies the loss of confidence in information as a critical risk. As it becomes harder for the public to distinguish between authentic information and AI-generated content, the impact can be felt in fields that rely heavily on public trust.
Beyond design and text, the problem extends to AI-generated photos and videos that are increasingly approaching reality. When a photo or video of a famous figure is created by AI and looks remarkably real, people are left guessing: is this authentic or AI?
The Decline in Quality
The problem may even extend to the AI technology itself. Oxford AI researcher Ilia Shumailov and his colleagues published a study in the journal Nature finding that AI models trained recursively using data generated by other AI models can experience a decline in quality. This phenomenon is known as ‘model collapse’.
Simply put, a model that continuously learns from the output of previous models risks drifting further away from the diversity of original data. It is similar to photocopying a document multiple times; each copy may still be legible, but certain details gradually vanish. As the process repeats, rarer information is more easily lost, and the most common patterns become dominant.
Rare data has a higher chance of disappearing when a system relies heavily on synthetic data generated from previous systems. For instance, if a journalist makes an error in writing a news report, or even simply uses incorrect Indonesian grammar, and this is written repeatedly, the AI will capture it as truth. That data will then be used by subsequent AI to produce even more erroneous writing.