The Internet Is Being Flooded With AI Slop

To understand why AI slop will proliferate despite its obvious low quality, you need to understand the specific economic logic that drives its production. The web's commercial infrastructure rewards content that captures search traffic, generates page views, and produces engagement signals, not content that is genuinely valuable to the reader. A ten-thousand-word article produced by an expert after weeks of research that is read by a specialist audience has less commercial value to most content producers than a series of thin AI-generated pieces that collectively capture the same search queries from a much larger audience. The quality difference between these two categories of content is enormous. The revenue difference is negligible or negative from the perspective of the producer.

The specific dynamics of search engine optimisation, the practice of producing content designed to rank highly in search results, have always created incentives for thin content production. The AI tools that can produce such content at near-zero cost make those incentives dramatically more powerful. A content farm that previously needed human writers to produce a hundred articles a month can now produce ten thousand with the same investment. The law of large numbers means that some of those ten thousand articles will rank for valuable search terms, capturing traffic that justifies the investment even if the vast majority captures nothing. The economics of AI content production make this not just possible but rational, and the commercial actors who do not pursue it will be outcompeted by those who do.

The most serious consequence of AI slop is not that bad content exists, which has always been true. It is what the flood of slop does to the findability and visibility of genuinely good content. If every search query returns AI-generated content that looks superficially similar to expert content, the signal-to-noise ratio that allows people to identify reliable information from unreliable information collapses. The good article and the AI-generated filler occupy the same format, share the same visual presentation, and the automated production of the filler means there are vastly more instances of it. The reader without the expertise to evaluate the content on its merits, which is the position most readers are in when they are searching for information on topics they do not already know well, has no reliable way to distinguish them.

There is a specific harm to expertise that the slop flood produces that I want to name clearly. When AI-generated content mimics the surface features of expert content without containing expert knowledge, it devalues the signal that those surface features provided. The long-form article, the well-structured argument, the specific citation to a primary source: these used to be signals that correlated with expertise. AI can replicate all of them without the underlying knowledge. The signals that allowed people to roughly distinguish reliable from unreliable content are being rendered uninformative by the ease with which they can be mimicked.

The harm to specific categories of quality content is already becoming visible. Journalism that requires original reporting is threatened not just by the economic pressures that have always threatened it but by competition from AI content that can produce news-adjacent text at scale without incurring the costs of original reporting. Creative writing that reflects genuine human experience is in competition with AI-generated fiction that is competent and generic. Specialist analysis that reflects deep domain knowledge is surrounded by AI-generated analysis that sounds similar and is substantively thinner. In each case, the quality content continues to exist but is progressively harder to find, harder to fund, and harder to sustain economically in a market where the competition is producing at near-zero marginal cost.

The most durable forms of content in the AI slop environment are those that cannot be generated without genuine human investment and that carry verification mechanisms that AI slop cannot replicate. Original investigative journalism that produces facts that did not previously exist. Creative work that reflects a specific human perspective in ways that are rooted in particular experience. Specialist content that requires genuine expertise to produce and to evaluate, and that is produced in communities where that evaluation happens. Content that is embedded in trusted relational contexts, where the trust in the relationship carries over to the trust in the content. These forms are not invulnerable, but they have properties that make them more valuable in a context of abundant undifferentiated content.

What individuals can do in a slop-flooded information environment is, unfortunately, more demanding and less effective than we would like. Developing the ability to evaluate content on its substantive merits rather than its surface features. Cultivating relationships with specific creators and publications whose quality you can verify over time. Relying on communities of practice where content is evaluated by people with the expertise to do so. Using AI tools themselves to synthesise and evaluate the content you encounter, with appropriate scepticism about the outputs. None of these is a complete solution, because the fundamental problem is structural: the production economics of AI content are misaligned with the information ecosystem we want. Fixing that misalignment requires either changing the economics, which requires platform and regulatory intervention, or accepting that the internet will increasingly be a place where finding genuinely good information requires substantially more effort than it did before.