Hidden in Plain Sight

Analyzed 100,000+ web privacy policies across two decades using the Princeton-Leuven Longitudinal Corpus. Used readability metrics (Flesch-Kincaid, SMOG) to measure comprehension difficulty across 130,000+ websites.

Found that 95% of policies require college-level reading skills, with the average demanding 15 years of education. Readability has significantly declined over time; policies are getting harder to comprehend, which is a serious obstacle to informed consent.

Observed a slightly surprising pattern: data-intensive industries (social media, advertising) appear to maintain slightly more readable policies than other sectors, though all remain largely inaccessible. Non-data-intensive industries showed faster decline in readability.

One plot from the work is shown here. Note that a higher score means a harder-to-read text.

Advised by: Abhilash Mishra, Labbi Karmacharya.