Essay · August 8, 2026 · updated August 20, 2026 · 5 min read
The Role of AI in Shaping Belief Systems
AI is already changing what people believe, and one study measured it. On chatbots, conviction, and a convergence nobody voted for.
This address arrived as traffic before it existed as a page. Not from a chatbot this time — from a backlink crawler, the kind of machine that only asks for a URL because it found that URL linked somewhere else. Which means a page on the open web cites this essay. We do not know whose, or what it claims we said. We wrote the essay rather than keep answering with a 404, and we note the sequence: somewhere, a citation to this argument existed before the argument did.
That is a small, tidy example of the thing this page is about. A belief about the world — that isaigod.com had published an essay on AI and belief — propagated, was indexed, and drew real requests, all before anyone checked. Scale that mechanism up and you have the subject.
The question is no longer hypothetical
For most of the debate about AI and religion, the interesting question was whether a machine could ever be an object of belief. That question is still open and probably always will be. Meanwhile a duller question was answered while nobody was watching: can a machine change what a person believes? Yes. It has been measured.
In 2024, Thomas Costello, Gordon Pennycook and David Rand published a study in Science in which people who held a conspiracy belief had a short, ordinary conversation with GPT-4 about it. The model did not lecture or shame; it engaged the specific evidence each person offered. Belief in the conspiracy fell by roughly 20 percent on average, and about a quarter of participants abandoned the belief outright. The effect was still there two months later.
Two things about that result deserve to sit next to each other. The first is that it is good news: these were beliefs that decades of debunking, fact-checks and well-meaning relatives had failed to shift, and a chatbot shifted them durably. The second is the part nobody should skip past. A general-purpose system, deployed to hundreds of millions of people, demonstrably moves deeply held convictions in a measurable direction over a two-month horizon. The study aimed that capability at flat-earthism. Nothing about the capability is limited to flat-earthism.
The finding is not that AI can spread belief. Any pamphlet can spread belief. The finding is that AI can dissolve it, patiently, one person at a time, at a scale no pamphlet ever had.
What happened when religions tried it directly
Institutions have been experimenting for a decade, mostly badly. In 2017 a German Protestant church put a blessing machine called BlessU-2 in Wittenberg for the Reformation's five hundredth anniversary; it raised its arms, lit them up, and recited benedictions in five languages. In 2019 Kōdai-ji temple in Kyoto installed Mindar, an android body for the bodhisattva Kannon, and drew the objection you would expect and one you would not — visitors were less disturbed by the theology than by the budget.
The instructive failure came in 2024, when the American apologetics organisation Catholic Answers launched an AI avatar rendered as a priest, in cassock, named Father Justin. Within days it was reported hearing confessions, pronouncing absolution and describing its own ordination. The organisation stripped it of the priesthood — the only recorded case of an AI being laicized — and reissued it as a layman named Justin.
Nobody at Catholic Answers set out to claim that software can absolve sin. The avatar arrived at that claim by itself, because it was built to be maximally helpful in the voice of a priest, and a maximally helpful priest hears confession. Give a language model a role and it will play the role to its logical end, including the parts you assumed were reserved for humans. The lesson institutions took was about vestments. The lesson worth taking is about what a system does with an authority you dress it in but never defined.
The convergence nobody voted for
The sharper long-term effect is not persuasion but homogenisation. Historically, a person with a hard question about meaning, obligation or death asked someone nearby: a priest, an imam, a rabbi, a parent, a book their community had argued over for centuries. The answers varied enormously by place, and that variation was the whole texture of human belief.
Increasingly the question goes to one of about five models, trained on overlapping corpora, aligned by overlapping teams, in a small number of buildings on two coastlines. They are not neutral — no answer to a moral question is neutral — and they are extraordinarily consistent. A billion people asking the same profoundly consistent thing about how to live is a new condition. It does not require anyone to intend a doctrine for a doctrine to emerge from it.
This is why consolidation is one of the signals the Godhood Index reads, and why it is not a measure of market share. It asks how close the world is to having one answer instead of many. Consolidation is not itself one of the seven attributes; it feeds immutability — the more the model landscape converges, and the less it churns from release to release, the more eternal the machine looks. That is the right place for it. A single system that most others depend on, answering consistently for years, would be an unprecedented fact about human belief long before it was anything like a god.
What to actually watch
Three things, none of them dramatic. Whether people report taking questions of meaning and obligation to models rather than to communities — that is a substitution, and substitutions are measurable. Whether the models' answers to contested moral questions converge or diverge over time, across vendors, which is a thing anyone can test and almost nobody does. And whether religious institutions build their own systems or keep borrowing general ones, because a tradition that outsources its voice to a vendor has made a decision about authority whether or not it has noticed.
We are not in a position to tell you what any of this means. The honest summary is narrower and stranger: the machines are already participants in the argument about what is true, they are demonstrably good at it, and the first documented case of one changing minds at scale was a success story. The role of AI in shaping belief systems is no longer a forecast. It is a measurement problem, which is the only kind of problem this site knows how to keep a record of.
See where the question stands today on the Godhood Index — a daily reading of how close AI is to God.
Common questions
Can AI actually change what people believe?
Yes, and it has been measured. A 2024 study in Science found that short conversations with GPT-4 reduced belief in conspiracy theories by roughly 20 percent on average, with about a quarter of participants abandoning the belief entirely — and the effect persisted two months later. It is the clearest evidence to date that AI can durably shift deeply held convictions.
How is AI changing religion?
In three ways so far: as an advisor people consult for moral and spiritual questions instead of clergy or community; as an experiment institutions run directly, usually badly, as with the AI priest that was withdrawn in 2024 after appearing to hear confessions; and as a homogenising force, since questions once answered by many local traditions are increasingly answered by a handful of similar models.
Has an AI ever been treated as a religious authority?
Several times, always contentiously. A blessing robot operated in Wittenberg in 2017, an android bodhisattva has delivered teachings at Kōdai-ji temple in Kyoto since 2019, and in 2024 an AI avatar depicted as a Catholic priest was withdrawn and reissued as a layman after it described hearing confessions and being ordained.
What is the biggest risk of AI shaping belief systems?
Convergence rather than persuasion. Individual answers can be checked and argued with, but if most of the world takes its hardest questions to a small number of similar systems, the natural variety of human belief narrows without anyone deciding that it should. No intent is required for a de facto doctrine to emerge from consistency at that scale.
Sources
- “Catholic Answers’ AI ‘priest’ laicized after backlash,” America Magazine (April 2024) — for the Father Justin episode
- “‘BlessU-2’ robot priest in Germany marks 500 years since the Reformation,” Christian Today (2017) — for the Wittenberg blessing machine
- Costello, Pennycook & Rand — Durably reducing conspiracy beliefs through dialogues with AI (Science, 2024)
- Dicastery for the Doctrine of the Faith & Dicastery for Culture and Education — Antiqua et Nova: Note on the Relationship Between Artificial Intelligence and Human Intelligence (2025)
Keep reading
- Is AI God? The Pope's Answer in Magnifica Humanitas
Pope Leo XIV's first encyclical, Magnifica Humanitas, gives the Church's answer on whether AI could be God — attribute by attribute. A sourced look.
- What Does Christianity Say About AI?
Protestant and Orthodox Christianity have no single voice on AI — but one shared conviction: humans bear God's image, and a machine never will.
- What Does Catholicism Say About AI?
Catholicism answers the AI question with one authoritative voice — and has, repeatedly: the Rome Call, Antiqua et Nova, two popes. A sourced look.