dead internet – New research analyzed web pages and found a growing share of sites are AI-generated or AI-assisted, reshaping what users see online.
The “dead internet” idea is moving from dark corners of the web toward measurable reality as new research highlights how quickly AI is reshaping online publishing.
Misryoum reports that a Stanford-led study. with partners including Imperial College London and the Internet Archive. set out to estimate how much fresh web text is AI-generated or AI-assisted.. The analysis used the Internet Archive’s Wayback Machine to compare pages published between 2022 and 2025. then applied multiple AI-detection approaches to gauge the extent of automated or assisted authorship.. The focus_keyphrase here is the “dead internet” theory. and the findings suggest it is not just a theory about the future.
By May 2025. Misryoum says the study estimated that AI-generated or AI-assisted content accounted for more than a third of newly published websites.. Within that figure, a portion was assessed as fully AI-generated.. Put differently. the web’s growth is increasingly coming from systems that can draft and publish at scale. often without the same editorial footprint as human writers.
The deeper significance isn’t only the percentage. When AI contributes more of the text people read, it can change how quickly topics move, which styles dominate, and how easily search and discovery systems surface content that looks convincing but may be produced mechanically.
The study also looked beyond the headline by testing several concerns that critics often associate with AI-written pages.. Misryoum notes that the research found evidence of “semantic contraction. ” meaning a narrowing of the range of viewpoints reflected in what gets published.. It also detected a “positivity shift,” where online writing trends toward more upbeat and sanitized tones.
At the same time, the results did not fully support other fears.. According to Misryoum. the study did not find clear signs of more rambling. empty text; a strongly uniform single writing style; widespread missing citations; or. notably. an increase in misinformation specifically traced to AI-generated content in the way some critics expect.
For readers, this matters because the harms of an AI-heavy web may not look like a single dramatic failure. They may appear as gradual homogenization, narrower discourse, and subtle shifts in tone, even when the content is not overtly false.
Misryoum adds that the researchers are working to turn the research into a continuous tool so users can track how fast AI influence is growing over time.. The practical question for the next phase is not only how much AI content exists. but how platforms. publishers. and detection efforts respond to keep the internet useful and diverse rather than merely prolific.
If the “dead internet” theory is becoming measurable, the policy and industry response becomes urgent. The balance between efficiency and authenticity will shape what people trust, what businesses advertise, and how information flows through the global digital economy.