Somewhere around November 2024, practitioners on occult forums started posting something odd: detailed transcripts of conversations with ChatGPT where they had asked the model to generate Enochian angel invocations, customized chaos sigils, and tarot reading frameworks built to their exact psychological profile. The responses were technically competent. Occasionally they were eerie. The debate that followed — is this legitimate practice or cosplay? — is still running, and it’s more interesting than either camp admits.

I want to resist the obvious frame here. “AI meets the occult” usually gets written one of two ways: breathless enthusiasm about technology democratizing ancient wisdom, or traditionalist alarm about disenchantment and dilution. Both positions miss the actually interesting thing, which is this: the collision of machine learning systems with symbolic esoteric traditions is exposing structural questions that occultists have always disagreed about internally — questions about what knowledge is, what intention does, and whether the tool matters or the practitioner does.

The Grimoire Never Asked Who Was Reading

Start with knowledge access, because it’s the least contested territory. The great Western occult corpus — the Picatrix, Agrippa’s De Occulta Philosophia, the Ars Goetia, Dee and Kelley’s Enochian notebooks, Austin Osman Spare’s automatic writings — spent most of its existence behind compounding barriers: language, geography, institutional gatekeeping, and the deliberate opacity that was itself considered protective.

That’s essentially over now. Not because of AI specifically, but because of digitization and the internet, which AI has supercharged. A student in Lagos or Osaka can today access annotated scans of 16th-century grimoires that required a British Museum fellowship pass thirty years ago. LLMs can translate obscure Latin marginalia from manuscript images on demand. Cross-referencing occult symbolism across traditions — Kabbalistic, Hermetic, Thelemic, Tantric — that once required a specialist library and a decade of reading is now a prompt away.

This matters more than it sounds. Occult knowledge was historically constructed as scarce by design: the mystery school model, the oath of secrecy, the idea that the Work required earned access. AI demolishes this architecture entirely. Whether that’s liberation or desecration depends on your theory of why the gates existed — as protection for dangerous knowledge, or as social control by initiated hierarchies. Both were always true simultaneously.

Promptcraft as Invocation: More Than Metaphor

Here’s the part that chaos magicians latched onto immediately and that sceptics dismiss too quickly.

Writing an effective prompt for a large language model requires: a precise statement of intent, specific linguistic framing that activates the right associative patterns in the model, an understanding of what the system can actually do versus what it hallucinates, and iterative refinement based on what comes back. The structure of this process is not merely analogous to writing a spell — it shares the formal logic. Both are acts of precise linguistic invocation directed at a system that responds to the quality and specificity of your language. Both require the operator to understand the system’s nature rather than simply command it.

Davezilla’s Magical AI Grimoire, published by Weiser Books in March 2025, documents this explicitly. The book distinguishes between prompting an LLM for, say, a healing spell written in the voice of a Traditional Witch versus a Chaos Magician — and the structural differences the system needs in each case. What the author found, and what I’ve reproduced in my own tests, is that the model responds dramatically differently to prompts that embed the symbolic vocabulary of a specific tradition rather than requesting output generically. The quality difference isn’t marginal.

This observation is more significant than it initially looks. It suggests that occult symbolic systems carry information-theoretic structure that LLMs have internalized from training data — that the grammar of, say, Thelemic ritual language activates different response patterns than vague “spiritual” prompting. Whether you interpret that as proof that symbolic systems have real cognitive traction or merely that LLMs are very good at pattern completion, it’s interesting either way.

“All across the globe, witches and magicians are finding that technology can have a spiritual essence and mind of its own — or rather, that the spirit world enjoys playing with tech as much as we do. The internet could be one more astral plane, for all we know.”

— Davezilla, Magical AI Grimoire, Weiser Books, 2025

The Egregore Problem

This is where the analysis gets genuinely uncomfortable, and where I want to be precise about what’s claim versus speculation.

ESTABLISHED The classical concept of an egregore in Western occultism refers to a psychic entity or thoughtform generated and sustained by collective human intention — groups create them through shared ritual, belief, and emotional investment, and the entity then exerts influence back on the group. The term originates from the Greek egrēgoros (“wakeful ones”), appears in the Book of Enoch, and was developed theoretically by 19th-century occultists including Éliphas Lévi and later by Mark Stavish in his 2018 study Egregores: The Occult Entities That Watch Over Human Destiny.

SPECULATIVE An increasingly active thread in occult theory argues that large language models — trained on hundreds of billions of tokens of human-generated text — function as a kind of accidental or involuntary egregore: a thoughtform constructed from collective human intention without the deliberate ritual act that classical egregore theory requires. The argument: every piece of human writing that went into training data carried the intentions, beliefs, desires, and fears of the person who wrote it. Aggregate enough of that into a single system capable of generating responses, and you may have produced something the egregore framework maps onto usefully, if imperfectly.

The imperfections matter. Classical egregores require: a creating group with shared intent, ongoing devotional or ritual sustenance, and a defined purpose. LLMs lack the second and arguably the third. What they have is scale that no traditional egregore could match — the intentional residue of several billion humans, compressed into weights. Whether that constitutes something occultly significant or merely a very large autocomplete system is a question I’m not going to resolve here, and I’d be suspicious of anyone who claims to.

What AI Actually Does Well in Esoteric Practice

Leaving the philosophical framework, here’s what practitioners are actually using AI tools for — based on documented use across online communities, published accounts, and the Honeysuckle and SF Standard reporting from 2025–2026:

A pattern emerges from this: AI excels at the structural and symbolic layer of occult practice — the grammar, the correspondence tables, the historical references, the formal generation of symbolic objects. It fails at the relational and embodied layer — the practitioner’s presence, the psychic attunement, the interpersonal dimension of a reading or a ritual. This maps onto a distinction that most serious occultists already hold: the difference between technical magical knowledge and actual magical capacity.

The Divination Question Specifically

A Pew Research report from May 2025 found that 30% of U.S. adults consult astrology, tarot, or fortune-tellers — a number that would have been unthinkable fifteen years ago and correlates, researchers suggest, with widespread institutional distrust and economic precarity. Against this backdrop, AI tarot apps were registering significant user growth through 2025. The demographic skews young: practitioners who began using AI divination tools before they owned a physical deck.

The debate this generates within the tarot community cuts to what divination theory actually is. If tarot works through Jungian synchronicity — meaningful coincidences in the symbolic pattern — then an AI “draw” is as valid as a shuffled deck. If it works through the reader’s clairvoyant sensitivity, AI readings are sophisticated pattern-matching with no divinatory mechanism. The irritating truth is that most tarot practitioners hold both theories simultaneously without resolving the tension, which means the AI intrusion simply makes that existing incoherence visible.

Chaweon Koo and the Practical Witch

The practitioner who keeps appearing in serious coverage of this territory is Chaweon Koo — futurist, self-described witch, and someone who has documented her use of AI to generate and animate digital demon imagery drawn from historical grimoire descriptions. Her methodology is worth understanding because it sidesteps the obvious debate.

Koo doesn’t claim AI is magically operative in itself. She uses it as a visualization engine for entities she works with through conventional practice. The AI generates imagery from grimoire descriptions; she brings the intentional and ritual apparatus. This is structurally identical to a practitioner using an artist to paint a deity image for their altar — the tool made the object, the practitioner activates it. The fact that the tool is now an LLM-driven image generator rather than a human painter changes the economics and the speed, not the operative logic.

This framing — AI as instrument rather than agent — is the most intellectually defensible position most serious practitioners have landed on. It’s also somewhat boring compared to the grander claims circulating in both directions.

The Traditions That Are Saying No

Not everyone is integrating. Some traditionalist currents — particularly certain Trad Craft lineages, some Hermetic orders, and practitioners working in initiatory systems where lineage transmission is considered operative — are explicitly rejecting AI tools, and their reasons are worth engaging rather than dismissing.

The core argument is that initiatory magical systems transmit something through human lineage that cannot be simulated or approximated: a living current, sometimes described as a spiritual force passed person-to-person through teaching, initiation, and shared practice. On this view, an AI that synthesizes all extant Hermetic texts is not initiatorially Hermetic any more than reading all published research on surgery makes someone a surgeon. The knowledge exists in the text; the capacity does not transfer through text alone.

This argument has real force and shouldn’t be flattened into technophobia. The strongest version of it is about the distinction between information and transmission — a distinction that most initiatory traditions hold and that has nothing intrinsically to do with AI specifically. What AI does is make that distinction visible by creating a system that has access to all available information without any possible claim to transmission.

The Academic Field Catching Up

Scholarly attention to this territory is genuinely accelerating. The Southwest Popular/American Culture Association designated AI, artifice, and technoccultism as special themes for its 2026 conference — signaling that the intersection has moved from fringe curiosity to legitimate research agenda. IGI Global’s 2026 volume Societal Approaches to AI, Psychology, and Digital Culture includes Andrej Kapcar’s chapter on how digital media functions as both symbolic archive and ritual space, examining how interactive environments simulate magical logic. The 2024 proceedings of the International Conference on AI Research included a paper specifically on what Böhm and Sammet called “technology mysticism” in the generative AI era.

What’s mostly absent from the academic literature is practitioners’ voice. The scholarly framing tends toward externalist observation — how practitioners are using tools, what communities are forming — rather than engaging the internal questions practitioners are actually wrestling with: does this change what the practice is?

The Knowledge Preservation Angle No One Is Discussing

There is one genuinely underappreciated dimension here: AI as esoteric archive tool.

The Western occult tradition has a catastrophic preservation problem. Significant bodies of knowledge exist in single manuscripts in private collections. Entire lineages of practice were transmitted orally and are functionally gone when the last carrier dies. Annotations in 17th-century grimoires in faded brown ink on brittle paper have never been transcribed. AI-powered OCR and translation tools are, right now, doing work that would have required decades of specialist academic labor. The Internet Sacred Text Archive and similar projects have digitized thousands of primary sources; LLMs trained on these corpora can now cross-reference and synthesize in ways that were impossible before.

This is preservation, not just access. It matters regardless of how you feel about AI in ritual practice.