People are bringing their hardest questions to AI: doubts about faith, a marriage in trouble, thoughts they haven't shared with anyone. AI is designed to generate ananswer. The more important question is what happens next.

Barna asked Protestant pastors and practicing Christians that question as part of State of the Church, its series on faith and AI, produced in partnership with Gloo. Across 10 situations, respondents said whether AI should stop and connect a person to a human helper.

Pastors and practicing Christians draw the line in different places

While a majority in both groups agree AI must hand off immediately on suicide or self-harm, pastors are far more decisive (94% vs. 63%). In eight of the 10 situations listed below, pastors are more likely than practicing Christians to want an immediate handoff, by double digits.

Source: Barna Group, State of the Church. n=442 U.S. Protestant pastors, December 2025.

The widest gap is spiritual crisis. Seven in 10 pastors (71%) say AI should hand off right away, compared with about one in three practicing Christians (32%). Nearly one in four practicing Christians (23%) say AI can continue without a handoff.

Source: Barna Group, State of the Church. n=683 U.S. practicing Christians, November 2025.

When someone no longer knows what they believe, they don’t need a faster answer. They need someone willing to walk through the questions with them. That has always been the work of the church. AI can support that work, but it can’t do it.

Pastors are adopting AI for the work, not for the shepherding

Across the churches Gloo serves, pastors are bringing AI into their workflows quickly, from sermon preparation to communications to administration. Yet the bar for letting AI advise people remains high, and it is unlikely to come down soon.

That is not resistance to technology. It reflects a pastoral conviction that some questions aren't asking for an answer. They're asking for a person. The data suggests that conviction is well placed.

What AI actually does with questions of faith

So if nearly one in four practicing Christians would let AI handle a spiritual crisis on its own, what does AI actually do with those questions? Gloo's Flourishing AI Christian Benchmark (FAI-C) measures how well AI aligns with Christian theological and relational values. It scores leading models on everyday questions about life and faith, first against general standards of well-being and then against a Christian understanding of flourishing.

In the June 2026 evaluation of 36 frontier models, every model scored lower under the Christian lens. Faith was the weakest dimension, averaging 50 out of 100, a 30-point drop from our previous measurement. Health, by comparison, dropped just 6 points. 

Even when models get the facts right, they rarely reason within a framework of faith. Gloo's researchers call this pattern procedural secularism: responses that are warm and balanced but silent on conviction.

Taken together, this research points to a clear risk. Practicing Christians are most comfortable letting AI handle a spiritual crisis alone, and faith is the dimension where AI is least equipped to help. The answer a person receives may not be harmful, but it is likely to be incomplete.

Industry and policy responses, such as OpenAI's trusted contact feature and California's proposed PAUSE Act, focus on acute safety risks, and each still depends on a human on the other end. For the gray areas, such as doubt, relationship strain and spiritual questions, no comparable handoff exists. That is where the Church has a unique role.

What this means for church leaders

The research affirms what most pastors already sense. It also shows that many in their congregations do not yet share that conviction. Three priorities stand out.

1. Give congregations language for discernment

A simple framework helps: information or formation? Information questions ask what is, such as what a passage means or how to build a budget. AI can help there. Formation questions ask who am I or what should I do, such as whether to forgive or whether God still cares. Those call for a person.

The same principle applies to church staff. A clear starting guideline: AI supports preparation, organization and communication. It does not replace counsel, comfort or spiritual direction.

2. Equip more people to be the handoff

No pastor can be the human response for every gray-area question. Scripture instructs leaders to "equip his people for works of service" (Ephesians 4:12). Elders, small group leaders and mentors who are prepared to listen to doubt and sit with grief extend a church's capacity for care.

3. Build relationships before the hard moment comes

In gray-area struggles, no system triggers a handoff. People reach for someone they already trust. AI sets a high bar for availability and privacy. Churches meet it not with more programs, but with a culture where people know at least one person, by name, they can call.

Technology in service of the mission

Humanity is made in God's image (Genesis 1:27) and entrusted with "the ministry of reconciliation" (2 Corinthians 5:18). “We don’t build for a market. We build for a person, made in the image of God,” said Pat Gelsinger, Executive Chairman, Head ofTechnology at Gloo, in his keynote at the 2026 Gloo AI Hackathon.

AI can lighten the administrative load and free leaders to spend more time with the people they serve. It can accelerate the mission. It cannot be the center of it.

"The church's role isn't to compete with these technological marvels," said Daniel Copeland, vice president of research at Barna. "It's to make sure that when someone needs a person, they already have one."

If you or someone you know is struggling, call or text the 988 Suicide & Crisis Lifeline.

About the research

State of the Church (Barna and Gloo): 683 practicing Christians among 1,514 U.S. adults surveyed in November 2025; 442 U.S. Protestant pastors surveyed in December 2025, using Barna's definition of practicing Christians. FAI-C (Gloo): 36 frontier AI models evaluated in June 2026 across seven dimensions of flourishing. Methodology is in the FAI-C paper.