A survey has found that about 63% of recently published religion and faith books on Amazon are highly likely to have been written by AI.

Decryption, a blockchain news outlet, reported on Aug. 23 that AI detection company Originality.ai analysed 2,034 religion and faith books listed on Amazon and classified 1,272 as having a high likelihood of AI authorship.

The survey covered 14 religion and faith categories. The highest share was in an occult category grouping Wicca, witchcraft and pagan traditions. It classified 78% of books in that category as highly likely to have been written by AI. Hinduism followed at 76% and Taoism at 74%. Sikhism, Buddhism, Islam, Catholicism, Protestantism, Orthodoxy and Judaism fell in between. By contrast, Mormonism was 42%, atheism 40% and Satanism 22%, relatively lower.

Researchers said the figures should not be compared directly because sample sizes differed widely by category. They also stressed the results were not definitive findings but assessments of likelihood. Report author Michael Freiman said it did not conclusively determine that texts were written by AI, and explained the model calculates the likelihood of AI authorship based on probabilities.

Originality.ai analysed book descriptions, author bios and sample text together, and classified texts scoring 50 or higher as having a high likelihood of AI authorship. Freiman said researchers separately reviewed books that received detection results. He said the team can now distinguish differences between AI-written content and human-written content.

The occult category also raised concerns over fact-checking. Researchers assessed that 53% of verifiable claims in books in that category were likely false. The books mainly dealt with topics such as healing crystals, herbal remedies and cleansing negative energy. Freiman said the primary readers of such books may be people seeking ways to heal themselves or improve mental health and to find solutions to "detox" from everyday chemicals or drugs, and said these Wicca books can address such demands without being verified under scientific standards.

Limits of AI detection technology itself have also been confirmed. The outlet pointed to cases in which AI detectors produced false positives or different results. In an October 2024 test of four detectors on the U.S. Declaration of Independence, results were mixed. In March this year, Colombia's Supreme Court rejected a legal filing based on a detection result that it was an AI-generated document, but when the same detector was used on a court ruling it assessed a 93% probability of AI generation.

Amazon's response remains an issue. Freiman said Amazon knows AI-written content is common on its platform and relies on authors to disclose AI authorship themselves. Originality.ai did not share the survey results with Amazon, and added it could not confirm how many authors actually disclose it.

The issue of paper books being used as AI training data is also drawing renewed attention. The outlet reported that it has become controversial for developers to buy millions of books, tear off the bindings and scan pages to create training datasets. An investigation in August also found that rare books were moved to an Amazon facility in Las Vegas for scanning work, and Amazon confirmed it purchases books through commercial distribution channels for that work.

The survey shows AI is rapidly entering publishing distribution networks. In areas such as religion and faith, where information asymmetry among readers is large, disputes may grow not only over whether content is AI-written but also over accuracy and disclosure standards.