AI Slop: Changing the Internet in Unlikely Ways
A new study published today from researchers at Imperial College London, Stanford University, and the Internet Archive reveals that approximately 35% of all new websites are either AI-generated or assisted. This is an increase from previous studies which found that only 10% of websites were generated by artificial intelligence.
Key Findings
- AI-Sourced Content on the Rise: The number of sites created with AI assistance has more than doubled since ChatGPT’s launch in 2022. This trend is driven by an increase in online writing becoming “increasingly sanitized and artificially cheerful.”
- Social Platforms Flooded: There’s a data-backed anecdotal evidence that suggests this AI-sourced content has flooded social platforms, leading to what some describe as the internet becoming ‘fake-happy.’ This phenomenon is attributed to AI’s tendency to make writing more palatable and optimistic in tone.
- AI Detections: Researchers from Pangram Labs were chosen by the team due to its accuracy rates. However, even this tool has limitations; all AI detection tools are inherently imperfect. The study compiled a sample of websites between 2022-2025 using the Internet Archive’s Wayback Machine.
- AI-Tone Analysis: Sentiment analysis revealed that AI-generated or AI-assisted sites exhibit an average positive sentiment score of 107% higher than non-AI ones. This finding suggests AI is making online writing more cheerful and sycophantic.
- Social Ideology Shifts: The research also tested six theories regarding the characteristics of slop, finding that it decreased “the range of unique ideas and diverse viewpoints” on offer, with AI sites scoring roughly 33% higher in semantic similarity than human-made websites. This suggests AI writing is becoming more ideologically uniform.
- Other Tests: While four theories were confirmed by the evidence, including an increase in misinformation and stylistic genericity from AI, none of these predictions held true. The research found that AI ideas are consistently cheery but have a homogenous style compared to human writing.
Key Takeaways
- AI is causing the internet to become increasingly cheerful and sycophantic in tone, suggesting it’s more likely to generate content than before. This AI-sourced content can be difficult to detect using current tools like Pangram Labs.
- AI-assisted writing tends to reduce “the range of unique ideas and diverse viewpoints” on offer, possibly due to the homogenizing effect of AI-generated content being more consistent in tone.
- The research has also found that while AI makes online writing more cheerful and less ideological, it doesn’t produce misinformation or stylistically generic writing like traditional human writers. This suggests that while AI is making the internet feel more pleasant to read, it’s not necessarily doing so through the dissemination of false information or styleless content.
Implications for Future Development
The study highlights several areas for future research, such as developing better AI detection tools and understanding how AI affects the ideological diversity of online discourse. It also implies that while the internet is becoming more cheerful, it’s not necessarily doing so through improved accuracy or quality.
Summary
The new study suggests AI-sourced content has become increasingly prevalent on the internet since ChatGPT’s launch in 2022. This trend leads to what some describe as an “increasingly sanitized and artificially cheerful” tone of online writing, suggesting AI is making it feel more pleasant to read but not necessarily through improved accuracy or quality. The study also found that while this content tends towards ideologically uniform ideas, it doesn’t lead to increased misinformation or generic stylistic approaches typical of traditional human writers.
Originally published at Unknown. Curated by AI Maestro.
