This study is an example of how you don't do science.

It fails to define critical term definitions, and uses such terms in contextual scopes that depending on the scope meant may contradict itself, with the context absent (i.e. welfare).

Fails to account for impactful actions that occurred in the same time period (i.e. internal catalog search changes, external search changes, and other changes related to requirements of all large businesses doing business in the US related to FOSTA-SESTA Act 2018).

Fails to vet data collection methodology or identify limitations of the dataset (Similarweb, bad data in bad data out).

Most people searching for porn use protection, fails to address collection methodology shortcomings when data collection is thwarted. It is also entirely unclear how the study controls for duplicate signals.

Fails by inserting value-based statements and asserting false narratives or flawed reasoning (a null hypothesis without alternatives, in a stochastic environment), also without proper basis, (i.e. the loss of 80% content and drastic changes in site discoverability/usability in aggregate).

There are a few phrasings, coupled with the poor methodology, that make me think this paper/study was in large part generated by AI, potentially as a pre-fabricated narrative (soft-propaganda).

The reasoning does not follow logically, and fails at obvious points where an AI would fail. On its face, this doesn't look like a sound study.