Pew Finds AI Fingerprints On Ten Percent Of Webpages. Post-ChatGPT Pages Hit Thirty-Five. The Tide Is Already In.
Pew Research found signs of AI authorship on 10% of sampled webpages and 35% of pages published after ChatGPT's release. Graphite, an SEO firm, separately estimated that newly published English-language content is increasingly AI-generated. Pew did not assess whether any single detection marker is reliable, which is a rather glaring omission, but the trend data is clear.
This demonstrates the baseline shift problem. Once AI-assisted writing becomes the default, the question of whether content is 'AI-generated' loses analytical traction. The mechanism is normalization through saturation. What matters is not provenance but quality, and detection tools chasing authorship markers will increasingly confuse edited human work with raw AI output.
Pew Research conducted the sampling study. Graphite, an SEO firm, provided the estimate on newly published English-language content. Matt G. Southern of Search Engine Journal reported the findings.
- Open ChatGPT and ask it to write 300 words on a topic you know well, then rewrite the same topic yourself in 300 words without AI assistance. Expected outcome: two comparable texts with subtly different rhythm and vocabulary patterns.
- Paste both versions into a free AI detection tool like GPTZero or Originality.ai's free tier. Expected outcome: detection scores for both texts, likely revealing false positives on your human writing.
- Compare the scores side by side and note which markers the tool flags. Expected outcome: a personal understanding of why authorship detection is an unreliable proxy for content quality.