A recent study found that 63% of religious books on Amazon were likely AI-written, identifying 1,272 of 2,034 titles examined as flagged by the researchers; this result is presented as an assessment of AI-written religious books on Amazon. The authors described their classifications as indications of likelihood rather than proof and noted known limitations in automated detection methods. The study used a predefined scoring threshold to determine likely AI authorship and framed the findings cautiously without asserting definitive proof.
The researchers examined 2,034 recently published religious books across 14 religious and belief categories to assess patterns in authorship classification. They applied a numerical scoring system and designated any text scoring 50 or higher as “Likely AI” in their classification scheme. The study framed these classifications as indications of likelihood rather than definitive proof of AI authorship. The methodology relied on automated detection metrics and pre-established thresholds to categorize texts without asserting conclusive authorship.
The researchers reported known limitations of automated detectors, including false positives and inconsistent results across different tools. An October 2024 test of four detectors on the U.S. Declaration of Independence produced conflicting outcomes, which the study cited as an example of detector inconsistency. A March report involving Colombia’s Supreme Court described detectors flagging a legal filing as AI-generated, with that ruling later evaluated as 93% AI-generated when re-tested. The study therefore emphasized caution in treating detector flags as conclusive evidence of AI authorship.
AI-written religious books on Amazon: Rates by category
The study reported category-level rates for AI likelihood within a sample that comprised 14 religious and belief categories in total. In that sample, the Wicca, Witchcraft & Paganism category had the highest reported rate at 78%, Hinduism registered 76%, and Taoism registered 74%. The researchers also reported lower category rates including Mormonism at 42%, atheism at 40%, and Satanism at 22%. The study provided these percentages for the categories named without enumerating the remaining categories in the sample.
The percentages cited reflect the share of books within each named category that the researchers flagged as likely AI-written under their classification criteria. The presentation of these category rates was limited to the categories explicitly reported in the study, and the study did not extend those specific percentage figures to categories not detailed in its findings. The language used in the report presented these figures as measurements within the sample rather than definitive statements about all religious books on the platform.
Michael Fraiman stated that the research team does not claim certainty about whether books are definitively AI-written and described the study’s classifications as indicators of likelihood rather than proof. He said the research model “determines the likelihood that something was AI-written to a degree of certainty,” and that, in the team’s view, they can distinguish between likely AI-written content and human-written content.
Fraiman also described a customer profile for certain Wiccan titles, saying one can picture an ideal customer as someone seeking solutions to heal themselves, improve mental health, or “detox” from commonplace chemicals and drugs, and that such books can address those concerns without being held to scientific standards.
Fraiman reported that Amazon did not respond to a request for comment and said Amazon is aware of how common AI-written content is on the platform. He described Amazon’s approach as using an honour system for authors to self-identify AI-written work and added that the company did not share its findings with the researchers.
Fraiman said, “We have no data on how many authors actually disclose that,” reflecting a lack of available disclosure information from the platform.
The study reported a high prevalence of books flagged as likely AI-written among the sampled religious titles on Amazon, presenting that finding as a central result of the analysis. The researchers emphasized that their classifications indicate likelihood rather than definitive proof, and they noted methodological limits in automated detection tools, including known false positives and inconsistent results in other tested contexts.
