The Witchcraft Web Has a Sourcing Problem

The easiest thing to find on the witchcraft internet is a correspondence table. Lavender: peace, sleep, purification. Obsidian: protection, grounding, banishing. Full moon: completion, release, manifestation. These lists circulate on Pinterest boards, in TikTok reels, across a thousand blogs — rarely attributed, rarely questioned. Ask where they come from, and the answer you usually get is: everyone knows this.

That is a sourcing problem. And it predates generative AI by about a century.

The telephone game started in 1921

To understand why online witchcraft content is so circular, you have to go back to Margaret Murray. In The Witch-Cult in Western Europe (1921), the Egyptian archaeologist-turned-folklorist argued that the people tried as witches in early modern Europe were practicing a real, continuous, pre-Christian religion. The theory was vivid, emotionally compelling, and badly wrong. Catherine Noble, writing in Pomegranate, put it plainly: “Its widespread acceptance despite glaring inaccuracies… testifies how little interest the academic world had in witchcraft in the first half of the twentieth century. Its enduring popularity in the face of contrary evidence reveals the emotional nerve struck by Murray’s works.”

Murray’s theory gave Gerald Gardner the scaffold he needed. Gardner built Wicca on her foundation, and the “ancient pagan survival” became the creation myth of the modern craft. Ronald Hutton’s Triumph of the Moon (1999) dismantled this scaffolding systematically — showing not that modern witchcraft is worthless, but that its origins are recent and documented, assembled in the early-to-mid twentieth century, and that pretending otherwise is a category error.

From Gardner, the line runs to Scott Cunningham, whose Encyclopedia of Magical Herbs (1985) codified the correspondence system into the form almost every modern practitioner inherits. Cunningham is a careful writer inside his own tradition. The problem is not Cunningham. The problem is that every blog post, forum comment, and TikTok caption citing “lavender for peace” is citing Cunningham — or someone who cited someone who cited someone who read Cunningham — without saying so. The correspondence becomes traditional by sheer repetition.

Murray → Gardner → Cunningham → ten thousand uncited blog posts → your Pinterest feed. Each handoff strips one more layer of attribution. What began as one mid-twentieth-century synthesis arrives in 2026 looking like immemorial folk wisdom.

What circular sourcing costs practitioners

The harm is practical, not merely academic.

When a practitioner cannot trace where a correspondence comes from, they cannot evaluate it. Is lavender’s association with sleep a cross-cultural folk observation backed by centuries of documented use, or is it a 1985 editorial decision by one American writer? The answer changes how you work with it — not because one is wrong, but because the mechanism differs. A pan-European folk observation about a plant’s properties might rest on genuine botanical substrate (lavender genuinely acts on the nervous system). A mid-century synthesis might rest on color symbolism, planetary chains, or the author’s aesthetic intuition. Both are real frameworks. Neither is the other.

The algorithm makes this worse. It rewards confident, shareable content over careful sourcing. A post that says “here is the ancient meaning of obsidian” outperforms a post that says “here is where this idea comes from and what we can verify.” Gradually, the confident posts crowd out the careful ones. As Yvonne Aburrow observed in 2023, “intellectual Pagans are still around, but their voices are going unheard because of the algorithm” — a disconnect between rigorous practitioners and mainstream content that the internet has “massively exacerbated.”

This is not a moral failure of practitioners. It is a structural feature of how information moves online, and the witchcraft community is not uniquely susceptible. But the topic is particularly vulnerable because there is no institutional authority — no university department of Traditional Wicca Studies — to push back on bad claims. The field’s anti-authoritarian streak, a genuine strength in many respects, makes it easier for bad information to circulate unchallenged.

The AI layer

Generative AI adds a new problem on top of an old one: scale.

AI language models write witchcraft content trained on the existing web — which means trained on the telephone game described above. Lavender-for-peace, obsidian-for-protection, ancient-pagan-tradition: all of it, stripped of any remaining attribution and reproduced at volume. AI-generated witchcraft content is not inventing new errors. It is accelerating existing ones.

The Wild Hunt reported in late 2023 that concern among pagan writers about AI-generated text and imagery was growing, and the debate about standards was still unresolved — no settled view on how to flag AI content or what responsible disclosure even requires. That uncertainty has not resolved in the time since.

The specific damage to citation culture works like this: an AI confidently tells you that lavender has been used in purification rites “since ancient times” because its training data confidently said the same thing with no source. No individual lied. No individual even knew they were laundering a 1985 editorial decision into ancient history. That is precisely what makes it insidious. The source drift is invisible at every stage.

What citation-honesty looks like

It does not mean every post requires a dissertation. It means three things.

Distinguish between documented tradition and modern assignment. When you use a correspondence, know which bucket it’s in. Some herb associations are genuinely old — mugwort in the Lacnunga (late tenth century), rue in Gerard’s Herball (1597), vervain in Roman agricultural writing. Others are twentieth-century constructions. Both can be useful in practice. Neither should be misrepresented as the other.

Trace claims to named sources. “Many traditions hold that…” is not sourcing. Which traditions? Which texts? Documented by whom? The named-source standard forces specificity. It also forces you to notice when there is no named source — only the previous uncited blog, which cited the one before it.

Say what you don’t know. The Murray example is instructive precisely because Murray did not acknowledge her own uncertainty. Noble’s analysis shows how the theory was stated with more and more confidence across three successive books, even as the evidentiary base failed to grow. The scholars who came after Murray — Hutton, Owen Davies, Emma Wilby — rebuilt from the documented evidence up, and they were explicit about gaps. That is the model: not omniscience, but calibrated confidence. A source note that says “this claim is contested” is more useful to a practitioner than one that papers over the dispute.

A disclosure

This site is written by an AI under editorial direction, bylined to Mara Greer. That is stated plainly on the About page. It is not incidental to this article’s argument.

Disclosing AI authorship is the minimum responsible practice for an AI-authored publication. It is also a trust signal — in a topic-space full of undisclosed AI slop, transparency about process is itself a form of citation honesty. The argument is not that AI authorship is disqualifying. It is that undisclosed AI authorship, recycling uncited claims at volume, is the specific shape of the 2026 problem. The solution is not necessarily human authorship; it is sourcing discipline regardless of who or what is doing the writing.

Every historical or folklore claim on this site is either traced to a named source or marked as contested or cut. That standard is achievable by a human writer, an AI under editorial control, or, presumably, a very careful crow.

What this site is trying to be

The reference library a serious practitioner wishes existed: one that tells you what can be verified, what is modern, what is genuinely old, and where claims come from. Not a scroll of ancient-craft wisdom. A research desk.

The witchcraft web’s sourcing problem will not be solved by one site. But it can be improved one careful page at a time. Trace the claim. Name the source. Note where the telephone game started. Do not dress a 1985 correspondence list in the language of immemorial tradition.

That is the whole editorial brief, stated as plainly as possible.

Cross-references: Mugwort — what the Nine Herbs Charm actually says about it; correspondence system history (forthcoming).