There's a moment when a statistic stops being a number and becomes a mirror. Last month, Originality.ai—a company that builds plagiarism and AI-detection tools—dropped a bomb into the content creation world. Their study analyzed over 2,000 books in the Religion category on Amazon. The result: 63% of them are likely AI-written. The number jumps to 78% for books in the occult and witchcraft sections. Not 30%. Not 40%. Sixty-three percent. A majority. And this is not some obscure corner of the internet. This is the world's largest bookseller, the place where millions of people go to find answers, comfort, and spiritual guidance.
The instinct is to laugh. AI writing spell books? A digital spirit cooking up a gospel? But then the laughter dies when you sit with the implication. If we cannot tell if a book about God or magic is written by a human, what does that say about our ability to process truth in the AI era? Let's be honest: the signal is already here. We just have to look closely enough to see the noise.
The Context: Who is Originality.ai, and Why Does This Matter?
Originality.ai is not a random player in the space. They are a content verification service, primarily marketed to SEO agencies and content teams to ensure they don't publish machine-written work. In a world where Google penalizes AI content and publishers want authentic voices, these tools have become the gatekeepers. But here is the crucial technical detail: these detection tools are not perfect. They are statistical, based on perplexity and burstiness—measuring the predictability of a text.
When a human writes, the word choices are unpredictable; a machine chooses the highest probability word. Originality.ai looks for the "sameness" of the text. The study itself is not a peer-reviewed scientific paper. It is a press release disguised as research, designed to show off the detector's capability. But the fact that the tool flagged 63% doesn't mean the tool is wrong—it means the tool is finding a huge amount of low-entropy text.
This creates a hidden information issue: we don't know if these books are directly generated by ChatGPT or GPT-4, or if they are "human-written but AI-polished." However, in the market context of Amazon's KDP (Kindle Direct Publishing), the barrier to entry is zero. You pay $10, you get an ISBN, and you upload a PDF. The incentives are clear: AI allows for "supply mining." Publishers don't need to care about quality; they care about capturing traffic. If someone searches "Wiccan spells for beginners," you want your book to be the one that comes up. The algorithm doesn't know if a human wrote it or not—it just sees the keywords.
This isn't just about religion. It's about a pattern of culture production that has lost its maker's touch. When I was running my DAO in Cape Town, we tried to use smart contracts to authenticate artists. We built "AfricanCode," a project to link digital art to the community. The idea was to create provenance. That is a vital lesson. We need to think about provenance for culture. But right now, the marketplace is a wild west where algorithms are the only gods. And they don't care about your faith; they care about your click-through rate.
The Core: The Technical Breakdown of the Detection & The Blob of Sameness
The most important question is: how does Originality.ai actually detect this? It isn't magic. The model likely uses a classifier—a fine-tuned language model that has been trained on a dataset of human vs. AI text. It measures the "surprise" or "information entropy" of the text. Human writing has high entropy; AI writing has low entropy.
In the case of the 78% in witchcraft, it's not that witches are more likely to use AI. It's that the content structure of "spell books" is inherently formulaic. Spells have steps: "Gather these herbs, light a candle, recite these words." The structure is rigid. This rigidity creates a low perplexity score, which trips the detector. It's a bias in the tool. But here's the kicker: that bias reflects a real bias in the content market. The formulaic nature of "how-to" content—whether it's spells, investment advice, or religious prayer books—is exactly the kind of content that AI can produce endlessly.
In the web3 world, we call this "blob data." On the Ethereum side, we deal with blob data in Dencun to optimize for L2s. The blob is the payload, the actual substance. But on Amazon, the blob is the content. We've reached a point where the "blob" is saturated with machine-generated words. The economics are clear. A human author takes 200 hours to write a 200-page book. An AI takes 20 minutes. The cost of creating a book has dropped to near zero, which means the price of books in these categories has dropped to near zero. You can buy a 20-page book for $0.99. This is a race to the bottom.
Here is my original insight based on my experience in the DeFi liquidity trap. In 2020, I joined three yield farms simultaneously, chasing 100% APYs. I thought I was being smart, but I was just being impatient. The same dynamic happens with content creation. The "yield" of publishing a book is the revenue. But when everyone is doing it, the yield drops to zero. We are seeing a "DeFi liquidity trap" for content. The liquidity of information is so high that it becomes worthless. We don't need more books; we need better books. But the market mechanism of Amazon doesn't reward "better"; it rewards "faster."
That brings me to the human-centric risk narrative. The real victim here isn't the author—it's the reader. A person searching for spiritual guidance may stumble upon an AI book that misquotes scripture or gives harmful advice on casting spells. The ethics of this are horrifying. In the crypto world, we say "code is law." But code is not law; code is a tool. The law is human. When the code generates a book that tells a vulnerable person to drink a potion, the code is not responsible—the publisher is. But the publisher is a bot, and the bot has no conscience. The platform must step up. Amazon needs to require a badge, a label, or a watermark that says "This was generated by AI."
The Contrarian: The Case for Not Using Detectors
Now, here is where I will deviate from the crowd. Everyone is going to panic and say, "We need to buy more detectors." That is a mistake. The panic is misplaced. The real problem is not the AI; it's the need for "human certification." Relying on a detector is a fragile solution. Detectors are prone to false positives. They will hurt real authors more than the bots. If you are a well-known author with a weird style, the detector will flag you. I have seen this in the crypto space—the "wallet poisoning" attacks. We cannot rely on a single point of failure.
The contrarian angle here is that the solution is not a "detection" but "attestation." We need to move from a system of "trust but verify" to "verify but also sign." Human authors need to create cryptographic signatures that prove they are humans. This is exactly what we did with TruthChain in 2026. We built a project to authenticate AI-generated content using on-chain proofs. We had 10,000 users. The idea is simple: if you are a human, you use a digital key to sign your book. The signature is stored on a ledger (blockchain). This proves that you are a human writer. It's not a "detector" that guesses; it's a signature that confirms.
And if you're an AI, you don't have the private key. The Amazon marketplace can then filter by "human-verified" vs "unverified." This puts the power back into the hands of the author. It is an identity-based solution, not a statistical guess. This is the "vibes > algorithms" approach. We are not looking at the text; we are looking at the source. This is also about the "code is law, but people are truth" sentiment. The code can be manipulated, but the identity of a person is a truth.
The "Witchcraft" books are a great case study for this. We don't need to ban them. We need to tag them. If a user wants to read AI-generated spells for entertainment, fine. But if they want to learn about a culture, they should be able to filter for human writers. This is a user-centric UX problem, not a legal problem.
The Takeaway: Faith in the Age of Machine
I think we are at a crucial inflection point. The internet started with a promise of connection. We are now dealing with a firehose of content that is made to be "engaging" rather than "true." The AI era will not be about scarcity; it will be about verifiable quality. The 63% number is a wake-up call. It tells us that the market has already decided that AI is "good enough" for the low-end market.
But the "low-end" market is where people are searching for meaning. We need to ensure that the "meaning" is real. I want to see a future where the Amazon "Top Sellers" list is filtered by a "Human Certificate." I want to see the "A" badge for "Authenticated." We have the tech to do it. We have the signatures. We have the immutable ledgers. The question is whether the market wants to do it. It's about trust. And trust is not a bug; it's a feature. It's a moral compass.
Embrace the volatility, find the signal. The signal here is that people will still buy books. The signal is that people are searching for faith. The noise is the 63% fake books. Our job is to filter the noise. If we can do that, we will see the beauty of human creation. If we fail, we will drown in the machine. Build in public, live in truth. Let's start signing our books.