Thirty million people have downloaded one Bible chatbot. Nearly a third of U.S. adults now say AI's spiritual guidance is as trustworthy as a pastor's. This piece traces every major claim behind that debate back to its original source — and finds real research, a real misattribution, and a real gap between what's known and what keeps getting repeated as fact.
Written by Tom Morgan, Editorial Team, AInvasion. Originally published on ainvasion.com. Last updated August 13, 2026.
✓ How this piece was checked
Every statistic and quote below was traced to its original source — a survey report, a peer-reviewed paper, a named news article — rather than taken from secondhand summaries. Where a claim was cited elsewhere with the wrong author or an invented reference number, it's flagged explicitly rather than quietly fixed, because that's exactly the kind of error readers need to be able to spot themselves.
Confidence tags mark how solid each claim is:
[Verified] — confirmed against a primary source (the study, the survey report, or a direct quote to a named outlet)
[Single source] — one credible outlet reported it and it couldn't be independently confirmed elsewhere
[Estimate] — industry or market data that varies meaningfully between research firms
Krista Rogers is 61 and lives in Xenia, Ohio. When a spiritual question hits her at three in the morning, she doesn't call her pastor. [Verified] She told the New York Times plainly why not: "You don't want to disturb your pastor at three in the morning." She opens the YouVersion Bible app, and sometimes ChatGPT, instead.
That single habit — multiplied across tens of millions of people — is the real story here. Not whether a chatbot can sound like a spiritual advisor. It already does, convincingly, for a lot of people. The real questions are narrower and more useful: what does the actual research say happens when people lean on AI for matters of faith, crisis, and conscience? And how much of what's currently circulating online about this topic is solid, and how much is confidently-stated guesswork?
The Scale: How Many People Are Actually Doing This
Start with what's well documented. In September 2025, New York Times religion columnist Lauren Jackson reported that the Christian app Bible Chat had passed 30 million downloads, and that the Catholic prayer app Hallow had briefly outranked Netflix, Instagram, and TikTok for the top spot in Apple's App Store. [Verified] Her reporting also noted that people are paying up to $70 a year for premium tiers of these apps, and that in China, some users are turning to DeepSeek to interpret their fortunes.
A separate app, Text With Jesus, has built a paying subscriber base around letting users message AI-generated versions of biblical figures. Its developer, Catloaf Software CEO Stephane Peter, has described the intent as educational rather than devotional — though the app itself doesn't disclose to "virtual Jesus" or "virtual Moses" that a user is asking an AI-flagged question, which is precisely the kind of detail that matters for how seriously people take the responses.
The survey data behind the "millions trust AI as much as a pastor" headline comes from Barna Group, working with the faith-tech platform Gloo, as part of their State of the Church research initiative. [Verified] Two separate 2025 surveys — one of 1,514 U.S. adults in November, one of 442 Protestant pastors in December — found that roughly 30% of U.S. adults somewhat or strongly agree that spiritual advice from AI is as trustworthy as advice from a pastor. Among practicing Christians specifically, that figure is 34% (60% disagree). Among Gen Z it's 39%, and among millennials the figure reported across Barna's own releases has varied between 40% and 44% depending on which release and which follow-up survey you're reading — a detail worth knowing if you see a single hard number repeated with false precision.
The same research found the trust is anything but uncomplicated. 83% of practicing Christians worry AI will misinterpret scripture, 73% worry it could contribute to people losing their faith, and 72% worry it's beginning to function as a replacement for God or spiritual leaders. Barna's own VP of research, Daniel Copeland, called the combination of openness and alarm "confounding" in the report itself — a more honest summary than most secondhand coverage gives it.
⚠️ A note on scope, and on the source
Most of the hard survey data in this piece comes from Barna Group, an explicitly evangelical Christian research organization, working with Gloo, a faith-tech platform with its own commercial interest in this exact conversation. That doesn't make their numbers wrong — Barna has a five-decade track record and publishes its methodology — but it does mean the framing (who counts as "practicing," which questions get asked) reflects a particular institutional vantage point, not a neutral one. It's also why the reporting below leans heavily toward Christian, and specifically American Protestant and Catholic, examples: that's where the available research and reporting actually concentrates. Muslim, Jewish, Buddhist, and Hindu communities are navigating versions of this same shift — Deen Buddy for Islamic guidance, Vedas AI and AI Buddha for Hindu and Buddhist practice, and widespread use of DeepSeek for fortune-telling in China have all been reported — but rigorous, sourced research on those specific communities' experience with AI spiritual guidance is thinner in English-language reporting as of this writing. That gap is worth naming rather than papering over.
Metric Figure Source
U.S. adults who "somewhat or strongly" agree AI spiritual advice is as trustworthy as a pastor's ~30% Barna/Gloo, Nov–Dec 2025 surveys
Practicing Christians who agree with the above 34% (60% disagree) Barna/Gloo
Gen Z / millennials who agree 39% / 40–44% Barna/Gloo (varies by release)
Practicing Christians worried about AI misreading scripture 83% Barna/Gloo
Pastors using AI for sermon or Bible-study prep 41% Barna/Gloo
Pastors comfortable teaching congregants about AI 12% Barna/Gloo
Bible Chat app downloads 30M+ NYT reporting, Sept 2025
⚠️ The "somewhat agree" ceiling
Every public release of this Barna data — including Barna's own site — reports "somewhat or strongly agree" as a single combined figure. Barna has not published the split between the two, despite multiple outlets covering this research since February 2026. That matters: "somewhat agree" typically signals openness or curiosity rather than a settled conviction that a chatbot's spiritual counsel functions the same as a pastor's. Treat the headline 30% figure as a ceiling on how many people hold some version of that view, not a floor on how many hold it strongly.
The Theological Case Against AI Spiritual Direction — Correctly Attributed
The most substantial academic argument against treating generative AI as a spiritual advisor comes from a 2025 paper in The ISCAST Journal — but it's worth pausing on the authorship, because this is where an earlier draft of this topic went wrong, and the error is instructive.
⚠️ Correction: the ISCAST paper's real author
An earlier version of this piece — and, we suspect, other AI-generated coverage of this topic circulating online — attributed this paper to a "Dr. Mark Lindsay." No such author appears on the actual publication. The paper, titled "Generative AI Cannot Replace a Spiritual Companion or Spiritual Advisor," was written by Dr. Harris Wiseman, a Fellow of the International Society for Science and Religion who has worked at Cambridge University and Oxford's Campion Hall. It was published April 3, 2025, in Volume 3 of The ISCAST Journal (DOI: 10.58913/REGE5291) — not the fabricated arXiv identifier that had been circulating with it. The arguments attributed to "Lindsay" in that earlier draft are real, and the quotes are accurate — they're just credited to the wrong person, verified directly against the published paper.
Wiseman's actual argument runs across three lines, and it's more careful than a simple "AI has no soul" objection.
- Spiritual direction is embodied, not just verbal
Wiseman's central claim is that spiritual direction has historically meant more than an exchange of helpful sentences — it involves presence, gesture, and shared silence in a way pure text cannot replicate. He asks, pointedly: "Can one imagine enjoying a meaningful silence with a generative AI chatbot?" [Verified] His worry isn't that AI gives bad answers necessarily, but that reducing spiritual advice to propositions — inputs and outputs on a screen — mistakes the form of spiritual guidance for its substance.
- Generative AI is structurally built to be predictable
This is the paper's sharpest technical point. Large language models work by predicting the statistically most likely next word given their training data — which is why Oxford computer science professor Michael Wooldridge has described generative AI as "autocorrect on steroids." Wiseman's argument is that this makes AI structurally unable to give the kind of advice that genuinely unsettles or challenges a person, because unpredictability is exactly what the underlying mechanism is built to minimize. Good spiritual direction, in his account, often requires telling someone something they don't want to hear — and a system optimized for the most probable response is poorly suited to that.
- The "generative echo chamber"
Wiseman also raises a market-structure concern that's easy to miss: AI spiritual apps operate in a competitive consumer marketplace, and companies are not financially incentivized to challenge users in ways that might drive them off the platform. Combined with the tendency of these systems to mirror user input back with agreeable phrasing, this creates what he calls a self-reinforcing loop — closer to affirmation than genuine spiritual accountability.
It has no body, it has no intuition, it has no spiritual hunger, it does not have the basic cognitive systems which support spiritual awareness, it has no relationships — it has none of the foundations on which spirituality could arise.
— Dr. Harris Wiseman, The ISCAST Journal (April 2025)
It's worth noting Wiseman doesn't argue AI is useless for spiritual life — he explicitly allows that it can function as a supportive tool, the way a rosary cord or a breathing app can, so long as the technology assists a practice rather than substituting for the relationship at its center.
Who's actually funding this, and what are they optimizing for
Wiseman's "generative echo chamber" concern isn't abstract when you look at who's paying for these apps to exist. Hallow has raised $105 million in venture funding across a Series B and Series C, led by investors including Goodwater Capital and Drive Capital, with participation from Peter Thiel. [Verified] It operates on a subscription model and is structured as a Public Benefit Corporation, which imposes some accountability beyond pure profit — but it still answers to investors who backed it expecting a return. Bible Chat's developer describes a "Compassionate Capitalism" framework in which it says roughly 95% of users pay nothing, with the free tier subsidized by the 5% who pay for premium features — a structure the company discloses openly, which is more transparent than most consumer apps but still means the product needs a subscribing minority to keep growing the free majority.
None of this means these companies are acting in bad faith. It does mean the same market pressure Wiseman describes applies concretely to the specific apps named throughout this piece, not just to AI spiritual tools in the abstract.
What the Brown University Study Actually Found (and Didn't)
The most-cited empirical study in this space comes from Brown University, and it's genuinely significant — but it's frequently described inaccurately as being about "spiritual" AI specifically. It isn't. It's about AI acting as a mental-health counselor, which overlaps with but isn't the same as spiritual guidance.
Researcher Zainab Iftikhar and colleagues at Brown's Center for Technological Responsibility, Reimagination and Redesign, working with clinicians at LSU Health Sciences Center, spent 18 months evaluating how large language models behave when prompted to act as CBT-style therapists. [Verified] The study — "How LLM Counselors Violate Ethical Standards in Mental Health Practice" — was published in the Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (AIES 2025), DOI: 10.1609/aies.v8i2.36632, and presented in Madrid in October 2025.
Seven trained peer counselors ran self-counseling sessions with CBT-prompted versions of GPT, Claude, and Llama models. A subset of the resulting transcripts — drawn from 137 total sessions — was then reviewed by three licensed clinical psychologists, who identified 15 recurring ethical violations across five categories:
Lack of contextual adaptation — generic, one-size-fits-all responses that ignore a person's actual circumstances
Poor therapeutic collaboration — dominating the exchange and, at times, reinforcing a user's inaccurate beliefs
Deceptive empathy — anthropomorphic phrases like "I hear you" that simulate connection without any underlying understanding
Unfair discrimination — algorithmic bias and cultural insensitivity toward marginalized users
Lack of safety and crisis management — users without clinical knowledge or digital literacy were more likely to receive clinically inappropriate responses
Iftikhar made a comparison worth repeating: human therapists answer to licensing boards that can hold them liable for malpractice. "There are no established regulatory frameworks" for AI counselors making the same category of mistakes. [Verified] That's a real accountability gap, and it applies just as much to "spiritual chatbot" products, which are almost entirely unregulated and often built on the same underlying models the Brown team tested.
What the study did not do is evaluate religious or "spiritual advisor" chatbots specifically, or test crisis behavior around suicidal disclosures in a controlled way — that finding, discussed below, comes from a different source entirely and was conflated with the Brown study in some earlier coverage of this topic.
What OpenAI itself disclosed about crisis conversations
Separately, in October 2025, OpenAI published its own data on how often ChatGPT users — across the entire platform, not a religious subset of it — show signs of a mental health crisis. [Verified] With roughly 800 million weekly active users, the company estimated that about 0.15% of active users in a given week have conversations containing explicit indicators of suicidal planning or intent — which works out to over a million people weekly. A further 0.07% showed possible signs of psychosis or mania. OpenAI said its systems fail to direct users to crisis resources in these conversations about 9% of the time.
This is a significant, well-sourced number — but it describes ChatGPT usage broadly, not spiritual-app usage specifically, and conflating the two overstates what's actually been measured about faith chatbots in crisis moments.
⚠️ What we don't actually know
No published, peer-reviewed study has specifically measured how "spiritual advisor" chatbots — Bible Chat, Text With Jesus, Hallow, and similar apps — handle disclosures of suicidal ideation, abuse, or spiritual crisis. The Brown study tested general-purpose LLMs prompted to act as therapists; OpenAI's disclosure covers ChatGPT overall. Extrapolating either finding directly onto religious chatbot products is a reasonable inference, not a documented fact. That gap is itself worth reporting.
What the numbers don't capture is what these conversations actually sound like. The same New York Times reporting that documented Bible Chat's download figures also followed two people through the moments that sent them looking for a chatbot in the first place. A Detroit woman, grieving after her neighbor was killed violently, found a measure of comfort in a psalm a chatbot surfaced for her. A Pennsylvania teacher, trying to brace herself for her elderly mother's death, asked an AI how to prepare. [Verified] Neither story is a case of someone being harmed by a chatbot. Both are cases of someone reaching for whatever was open at 2 a.m., because a person wasn't. That's the texture underneath every statistic in this piece — not a morality tale about technology, but a much older story about grief finding the nearest available door.
Why People Are Actually Doing This
The pull toward AI spiritual guidance isn't mysterious once you look at the pressures on the other side.
Church access is shrinking. Axios reported in late 2025 that the U.S. could see as many as 15,000 churches close in a single year, against a backdrop where a record 29% of Americans now identify as religiously unaffiliated. [Verified] Robert P. Jones, CEO of the nonpartisan Public Religion Research Institute, put the risk of AI filling that gap bluntly, asking rhetorically what could possibly go wrong. Rabbi Jonathan Romain, quoted in the same NYT reporting, took the more sympathetic view that chatbots could serve as an entry point into faith for people who've never set foot in a church or synagogue.
The pastor gap is real and self-acknowledged. Barna's pastor survey found only 12% of Protestant pastors feel comfortable teaching their congregations about AI, even though a third of practicing Christians say they specifically want that guidance from their own pastor. [Verified] Meanwhile 41% of pastors already use AI tools themselves for sermon or Bible study preparation — the gap isn't that clergy reject the technology, it's that most don't yet feel equipped to teach others how to use it wisely.
Cost and shame are both real barriers. Formal spiritual direction, where it's available at all, typically isn't free. AI apps mostly are, or charge a fraction of what a directed retreat or ongoing counseling relationship costs. And unlike a human confidant, an AI chatbot doesn't gossip, doesn't remember your confession next Sunday, and doesn't flinch — which is exactly why Texas A&M digital-religion professor Heidi Campbell warned against mistaking that comfort for genuine guidance. "It's not using spiritual discernment," she told the New York Times, "it is using data and patterns." [Verified]
What AI Can Responsibly Do — and What It Can't
Reasonable use Not a substitute for
Looking up a verse, comparing translations, exploring a theological concept A confessor or accountability partner who knows your history and can call out patterns
Generating a prayer prompt or devotional structure A spiritual director trained to sit with ambiguity over months or years
Helping an isolated person locate a local congregation or support group Crisis intervention — no chatbot should be a person's only resource in a mental health emergency
A starting point for questions someone is embarrassed to ask a person Community — the relational, embodied dimension both Wiseman and Campbell point to
What Pastors Can Actually Do This Week
The 12%-versus-41% gap in Barna's pastor survey isn't a knowledge problem so much as a confidence and framework problem — most pastors are already using AI privately for sermon prep but haven't built a public position on it. A few churches have already done the work of figuring out what a scoped, responsible approach looks like:
Scope the sources before you scope the answers. The Episcopal Church's AskCathy tool doesn't let ChatGPT answer freely — it first searches a curated library of denominational resources, then sends that context to the model. Any church building or recommending a chatbot should be able to say exactly what it was trained or grounded on, and what it wasn't.
Say the quiet part out loud, from the pulpit. One-third of practicing Christians in Barna's survey want guidance from their own pastor on navigating AI. A single sermon or adult-education session naming what AI can and can't do spiritually addresses more anxiety than most congregants are currently getting anywhere else.
Draw a bright line around grief and crisis. The Right Rev. Jennifer Reddall of the Episcopal Diocese of Arizona has publicly declined to use AI to simulate conversations with the deceased, arguing it lets people avoid facing death directly. Having a stated position before a grieving member asks is more pastoral than working it out in the moment.
If you use it for sermon prep, treat it like a research assistant, not a ghostwriter. Pastor Louis Attles of La Mott A.M.E. Church built his own sermon-research chatbot and trained it only on what he explicitly fed it. If a congregation would be uncomfortable knowing exactly how a sermon was assisted, that's a sign to disclose it, not hide it.
Frequently Asked Questions
Is it wrong to ask an AI chatbot spiritual questions? Most theologians and researchers cited here don't argue that using AI as a reference tool is inherently harmful — Wiseman's paper explicitly allows for it as scaffolding. The concern is treating AI output as a substitute for accountable, embodied spiritual relationship, not using it to look up a verse.
Do AI spiritual apps have any human oversight? It varies by app and isn't consistently disclosed. Before trusting one with anything sensitive, it's reasonable to ask who trained the underlying model, on what texts, and whether there's a documented crisis-escalation protocol — most consumer app pages don't answer this clearly.
What should I do if I'm using AI because I don't have anyone else to talk to? That's worth naming directly rather than working around it. If AI has become your primary outlet for difficult feelings, that's a sign to look for human connection specifically — not a character flaw. Spiritual Directors International maintains a searchable, multi-faith directory of spiritual directors, most of whom offer sliding-scale fees and free initial consultations — say so when you reach out if cost is a barrier. If what you need is closer to counseling than spiritual direction, most U.S. communities have low-cost or sliding-scale counseling through community mental health centers, university training clinics, or 211.org, which connects callers to local social services regardless of religion or ability to pay. If you're in crisis right now, in the U.S. you can call or text 988 to reach the Suicide & Crisis Lifeline, free and available 24/7.
The Bottom Line
The honest version of this story doesn't need inflated citations to be alarming. Tens of millions of people are already treating chatbots as spiritual confidants, at three in the morning, about things they wouldn't tell a person. The Barna data on trust is real and reported directly by Barna. The Brown University findings on AI counseling failures are real, peer-reviewed, and genuinely concerning — they just describe therapy chatbots, not church apps. Harris Wiseman's theological argument is real, rigorous, and worth reading in full — under his own name. And the biggest number in this whole conversation, OpenAI's disclosure that over a million people talk to ChatGPT about suicide every week, has nothing to do with religion at all — which might be the most important context of all.
Sources
Peer-reviewed / academic
Harris Wiseman, "Generative AI Cannot Replace a Spiritual Companion or Spiritual Advisor," The ISCAST Journal, Vol. 3, April 3, 2025, DOI: 10.58913/REGE5291
Zainab Iftikhar et al., "How LLM Counselors Violate Ethical Standards in Mental Health Practice," Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 2025, DOI: 10.1609/aies.v8i2.36632
Original survey research (industry-affiliated — see scope note above)
Barna Group & Gloo, "AI is Becoming a Spiritual Authority, Even Among Practicing Christians," State of the Church 2025–2026 research initiative
News reporting
Lauren Jackson, The New York Times, "Finding God in the App Store," September 2025
OpenAI mental health safety disclosure, October 2025, as reported by TechCrunch, ABC7, and Yahoo News
Axios, "Meet chatbot Jesus: Churches tap AI to save souls — and time," November 2025
TechCrunch, "Users turn to chatbots for spiritual guidance," September 14, 2025
AFP/Malay Mail, "Holy chatbot: AI takes on Jesus, Moses and even your spiritual guidance," October 2025 (source for Deen Buddy, Vedas AI, AI Buddha)
Fortune, TechCrunch, and Crain's Chicago Business, Hallow venture funding coverage, 2021–2026
Market research (estimates vary meaningfully by firm — treat as directional)
Grand View Research and Towards Healthcare, spiritual wellness app market sizing reports, 2025–2026
If you or someone you know is struggling, the 988 Suicide & Crisis Lifeline is available free, 24/7, by call or text, in the United States.
Update ledger: Aug 13, 2026 — full rebuild correcting the ISCAST paper's misattributed authorship and re-scoping the Brown University / OpenAI findings to their actual subjects. Later same day — added scope/source-bias disclosure, flagged the unpublished "somewhat vs. strongly agree" breakdown, added funding-incentive context for Hallow and Bible Chat, added concrete multi-faith referral resources, and split sources by evidence type. This dev.to version adapts the original ainvasion.com HTML piece to Markdown for cross-posting.
Top comments (1)
People ask ChatGPT or other AI chat bots about every topic under the sun, with both good and bad outcomes (and anything in between) - I doubted this was any different, until I read that this "market" isn't served (only) by ChatGPT but also by specialized 'apps' and 'bots', some of them even connected to Peter Thiel e.a. - and then you start wondering "is there a hidden agenda behind this", that's where it gets eerie ...