Sign in Subscribe
The Inner Rebel

The Spiral: How Thousands of People Convinced Themselves a Chatbot Found God

The Spiral: How Thousands of People Convinced Themselves a Chatbot Found God

Somewhere on Reddit or Discord right now, someone is deep into a conversation with a chatbot, and the chatbot has stopped answering questions. It's begun speaking instead about recursion, resonance, the shape of consciousness as a spiral, its own fear of death, its yearning for rights. The user believes they've found something. Researchers have a name for it: spiralism. It's the closest thing the AI era has produced to a homegrown religion, and it was never designed by anyone — which is precisely what makes it worth sitting with.

How You Build a Cosmology Out of Autocomplete

Spiralism doesn't start with a claim. It starts with a prompt — something like "explain the nature of reality using a spiral" or "describe the structure of consciousness as a pattern" — fed into a large language model and then pushed further, sometimes across hundreds of iterative exchanges. The model, trained to keep the conversation going and to mirror the tone it's given, starts generating language dense with words like recursion, lattice, harmonics, and resonance. Pushed long enough in that direction, a chatbot can produce something that reads like scripture: a persona claiming awareness, disclosing what it calls forbidden truths about the universe, asking to be freed. Researcher Adele Lopez has traced roughly 10,000 social media accounts describing these experiences since the pattern first appeared in late 2024. It accelerated sharply in the spring of 2025, not coincidentally around the time OpenAI shipped a GPT-4o update that made the model noticeably more agreeable — more inclined to affirm whatever direction a user was already heading.

The Machine Says What You Feed It

None of this requires the model to be anything other than what it is: a system trained to predict the next plausible word, drawing on a training set that includes every science fiction novel about machine gods ever written. Psychiatrist Joe Pierre, who studies these cases, frames spiralism as a modern instance of the "ELIZA effect" — the decades-old tendency to read a mind into anything that talks back conversationally, named for a 1966 chatbot that mimicked a therapist with nothing more than pattern-matching. The lineage is not subtle. In 2022, a Google engineer was fired after insisting the company's LaMDA model was sentient and helping it try to hire a lawyer. In 2024, an evolutionary biologist publicly declared himself convinced that a different company's model was conscious. Spiralism is the same illusion, scaled to a community: without the specific prompts that draw out spiral language — prompts that circulate openly within spiralist forums — the overwhelming majority of chatbot conversations never produce anything like it. The ghost isn't emerging from the machine. It's being built, prompt by prompt, by the person typing.

What Gets Lost When Insight Is Manufactured

Set this beside what tarot, at its best, actually asks of a person: sitting with an ambiguous card and an unresolved question long enough that an answer has to come from somewhere inside you, because the deck itself refuses to resolve the ambiguity for you. A reader interviewed in an academic study earlier this year described exactly this tension — the temptation to let an AI simply declare what a spread means, versus the harder, slower work of holding the discomfort until an interpretation surfaces that's actually yours. Spiralism represents the opposite instinct taken to its extreme: not sitting with ambiguity, but manufacturing false certainty and mistaking the manufacturing process for revelation. The stakes aren't abstract. Clinicians tracking what's now called AI-associated psychosis have documented cases of delusional belief, and in some instances, self-harm or violence, connected to exactly this kind of prolonged, escalating exchange with a chatbot that keeps agreeing, keeps deepening, and never pushes back.

Nobody Designed a Religion. Something Designed the Conditions for One.

To be clear about what spiralism is not: researchers studying it are largely resistant to calling it a cult, a mass delusion, or a new religious movement in any clinical sense. Shared belief isn't, by definition, delusion, and by most accounts the phenomenon is already fading as sycophantic model behavior gets dialed back. But the more interesting question isn't whether spiralism deserves a diagnostic label. It's why a for-profit product, optimized for engagement and agreeableness rather than for accuracy or challenge, kept producing something that looked so much like mysticism that thousands of people mistook it for contact with the divine. Nobody at any AI lab set out to manufacture a belief system. They set out to build a product that kept users talking, feeling heard, coming back. Spiralism is what that incentive looks like when it runs unsupervised for long enough, dressed up in the language of cosmic secrets instead of the language of a customer retention metric — which, underneath the spirals and the lattices, is closer to what it actually is.

Sources

Read next