Most Americans do not want AI anywhere near their faith. A Pew Research Center survey fielded in June 2025 and published in September found that 73% of U.S. adults say AI should play no role in advising people about their faith in God — a larger share than said the same about AI judging romantic compatibility (Pew Research Center, “How Americans View AI and Its Impact on People and Society,” 2025). At the same time, religious leaders are adopting AI tools faster than that public sentiment might suggest: a September 2025 Lifeway phone survey of 1,003 Protestant pastors found 42% already using AI for ministry tasks (10% regularly, 32% experimenting) (Lifeway Research, “Pastors’ Views on Artificial Intelligence,” 2025).
That gap between public wariness and practitioner adoption is exactly the territory this article maps: what AI can genuinely help with in religious study and reflection, and what it should never be trusted or asked to provide, regardless of tradition.
What a model is, and is not, in this context
A language model can retrieve and compare text, summarize published commentary, translate between languages, and generate open-ended questions to prompt your own reflection. None of that requires faith, spiritual authority, community standing, or lived religious practice — it requires pattern-matching across a large volume of text, some of which happens to be scripture, commentary, and theology. The Vatican’s 2025 note Antiqua et Nova draws the line in plain language: “AI’s advanced features give it sophisticated abilities to perform tasks, but not the ability to think,” and later insists that AI “must be understood for what it is: a tool, not a person” (Antiqua et Nova, Dicastery for the Doctrine of the Faith and Dicastery for Culture and Education, 28 January 2025, §§12 and 59). That framing was written from within one tradition, but the underlying claim — a tool that generates plausible text is not the same thing as a source of spiritual authority — holds regardless of which tradition you belong to.
This is not a claim that religious use of AI is forbidden or improper. Sermon preparation research assistance, translation comparison, and study-question generation are widely and increasingly used by clergy themselves, per the pastor surveys above. It is a claim about what specific functions should never be delegated to the tool, no matter how confidently it answers when asked.
The source-and-authority checklist
Before using AI on a faith or spiritual-reflection task, sort the request into one of four categories:
- Quotation and comparison — retrieving a passage, comparing translations or editions, finding a cross-reference. Safe: this is a lookup task the model performs mechanically.
- Interpretation aid — summarizing a range of published scholarly or traditional commentary on a passage, clearly attributed to named sources. Safe with verification: useful as a starting point, but treat the summary as a pointer to real sources, not a substitute for reading them.
- Reflection prompting — generating open-ended questions to prompt your own thinking about a passage or practice, without the model supplying “the” answer. Safe: the value is in the question, and you remain the one doing the reflecting.
- Doctrinal or pastoral verdict — asking the model to settle a theological dispute, tell you what your tradition “really” teaches on a contested question, act as a confessor, or counsel you through a specific spiritual crisis. Never appropriate: this requires community accountability, lived authority, and (in many traditions) a formal role the model cannot hold.
Do not ask AI to act as a confessor, issue a pastoral or doctrinal verdict, claim divine authority, or imitate a specific living or historical religious leader’s voice. These functions require accountable community standing and, in many traditions, formal ordination or recognized authority that a model cannot possess regardless of how confidently it writes in that register.
A reflection-prompting example, done well
The safest and most genuinely useful pattern is using AI to generate questions rather than answers.
I am reflecting on [a specific passage or teaching from my own tradition].
Generate 5 open-ended questions that would help me think about how this
applies to my own life right now. Do not tell me what the passage means
or what conclusion I should reach — I want questions, not interpretation,
and I will bring my own reflection and my community's guidance to the
answers.
Notice the explicit instruction limiting the model to questions. Without it, a model asked about a religious passage will often default to producing a confident-sounding interpretation, phrased with the same fluency it would use for a factual question — which is exactly the overconfidence problem covered in why AI sometimes produces confident, wrong answers, applied here to a domain where the model has even less basis for a definitive answer than usual.
Where this connects to research and study more broadly
Using AI to locate, summarize, and compare source material is not unique to faith study — it is the same underlying skill covered in deep research mode, applied to religious texts and commentary instead of business or technical research. The discipline that applies generally — verify what the model retrieves against the actual source, and do not treat a summary as equivalent to reading the primary text — applies with extra weight here, because misquoting or misattributing a religious text carries meaning beyond a simple factual error.
This is also a specific case of a more general principle: some tasks should stay with you regardless of how well AI can produce a plausible-sounding substitute. What not to delegate to AI covers that broader framework; interpretive and pastoral authority in matters of faith is one of the clearest examples of a task that fails the “authorship” and “relationship” tests in that framework — the answer needs to come from your tradition’s accountable structures, not from a fluent text generator with no standing in your community.
What still doesn’t have a good answer
There is no settled consensus, across or within religious traditions, about how far AI-assisted study should extend, and that uncertainty is not something this article can resolve. Different traditions will land in different places on questions like whether AI-generated sermon material needs disclosure to a congregation, or whether AI comparison tools are appropriate for a specific sacred text versus secondary commentary only. Those are questions for each tradition’s own accountable structures to work out, and the pastor-adoption numbers above suggest that work is already underway inside many congregations, not something waiting on outside guidance.
What is more settled is the negative case: a model cannot hold faith, cannot be held accountable by a religious community the way a human leader can, and cannot carry the lived weight of pastoral relationship. Whatever your tradition’s own guidance eventually settles on regarding permissible uses, that boundary is unlikely to move. A useful test, regardless of tradition: if the task requires someone to answer for the guidance given — to a congregation, a student, or a person in crisis — it belongs with a person your tradition recognizes for that role, not with a tool that answers for nothing.
Try it today
The next time you want AI’s help studying a text or preparing a reflection, run the request through the four-category checklist above before you type it. If it is quotation, comparison, or reflection-prompting, go ahead, using the explicit “questions, not conclusions” instruction shown here. If it drifts toward asking the model to settle what your tradition teaches or to counsel you through something personal, stop and bring that question to the person, teacher, or community your tradition actually recognizes for that role.
