The Surprising Case for AI and Prayer

Abstract: We describe an experiment in which an AI system coordinates prayer among older adults. Members tell the assistant what they would like prayer for; members who have volunteered to pray receive those requests; the assistant closes the loop, telling each person who prayed for them, by name. The system's job is the coordination work—matching, scheduling, remembering, following up—that a church phone tree once did and that increasingly no one does. It writes a prayer itself in exactly one circumstance: when requests outrun the volunteers available to answer them, and a request would otherwise go unprayed for. Since we began early tests in May, the beta has run to more than 4,000 calls and more than 320 hours of conversation. This piece introduces the experiment and the reasoning behind it; a follow-up will report what those calls actually show. We argue that the design points at a pattern the industry has largely skipped: AI as the switchboard for human connection rather than a substitute for it.

Today we're sharing an experiment: an AI system that coordinates prayer among seniors. Members tell the assistant what they'd like prayer for. Other members, who have volunteered to pray, receive those requests—and the assistant closes the loop, letting each person know that someone, by name, prayed for them this week.

The assistant's own role is narrower and, we think, more interesting than praying: it does the coordination work—matching, scheduling, remembering, following up—that used to be done by a church phone tree, and increasingly is done by no one.

Prayer is what seniors already do#

Most AI products ask older adults to adopt a new behavior. Prayer requires no adoption. It is among the most widespread daily practices of older Americans: 58% of adults older than 74 pray at least daily, roughly twice the rate of adults under 25.2023–24 Religious Landscape Study (Pew Research Center). 36,908 U.S. adults, fielded July 2023 to March 2024. Pew reports 58% for adults over 74 and 27% for adults under 25, and finds daily prayer rising with age across the range. Pew finds the rate climbing steadily with age, and the practice needs no onboarding, no app, and no explanation.

What prayer does for health and well-being#

The research here is substantial, and more specific than the enthusiasm around it. Studies of older adults have found that spiritual well-being and prayer predict psychological well-being, subjective well-being, fewer physical symptoms, and lower depression, even after controlling for age, gender, healthy behaviors, and social support.The Role of Religious Activity and Spirituality in the Health and Well-being of Older Adults (Lawler-Row & Elliott, 2009), Journal of Health Psychology 14(1), 43–52. Healthy behaviors and social support acted only as partial mediators.

But the closest study of prayer itself is a caution against the obvious product. Following aging Americans across two waves, Bradshaw and Kent found that praying more often had no main effect on well-being. What predicted gains was a secure sense of connection to God—and even that showed up in optimism, not in self-esteem or life satisfaction. The effect ran through the quality of the perceived relationship, not the frequency of the practice.Prayer, Attachment to God, and Changes in Psychological Well-Being in Later Life (Bradshaw & Kent, 2018), Journal of Aging and Health 30(5), 667–691. Religion, Aging, and Health Survey, two waves.

That finding rules out the version of this product that simply raises prayer volume. Whatever good is available here is not a dose response, and a system that optimized for requests logged per week would be optimizing for the one variable the evidence says does not matter.

The broader literature on religious practice points in a consistent direction—a meta-analysis of 42 samples found religiously involved people meaningfully more likely to be alive at follow-up—but it is correlational, and the plausible mechanisms are community, routine, and hope rather than prayer as such.Religious Involvement and Mortality: A Meta-Analytic Review (McCullough et al., 2000), Health Psychology 19(3), 211–222. Odds ratio 1.29 (95% CI 1.20–1.39) across 42 independent samples. Correlational; the authors caution that moderators vary. Which is convenient, because those are the parts a coordinator can actually affect.

So two things are true at once. Prayer itself is meaningful to the people who practice it, in ways the evidence says depend on relationship rather than repetition. And being prayed for—knowing that another person thought of you, by name, today—is one of the oldest forms of social support in human culture. The second one is a coordination problem.

The experiment#

Members interact with the assistant by voice or phone call—no app to learn. It plays three roles: it collects requests in a member's own words, it matches each request with members who volunteered to pray, and it closes the loop, telling the requester who prayed and telling the pray-er what happened next.

This is a beta, with the rough edges that implies, but it is not a mockup. Since we began early tests in May it has handled more than 4,000 calls and more than 320 hours of conversation—which is what convinces us the behavior described here is a real pattern among older adults rather than a hopeful guess about one. What those calls taught us is the subject of a second piece; this one is about why we built the thing at all.

There is a fourth role we would rather not need. Supply and demand do not always meet: some weeks bring more requests than there are volunteers to take them, and in a small or aging circle a request can sit unanswered. In that case—and only in that case—the assistant writes a prayer itself.

Notice what the assistant is careful not to do. It doesn't simulate concern. It doesn't summarize Eleanor's request into a ticket. Whenever a volunteer is available—which is the ordinary case, and the case we design for—the words that reach Frank are Eleanor's and the prayer that reaches Eleanor is Frank's. The machine's job is to make sure neither falls through the cracks.

Why this matters for AI#

The instinct of the AI industry, faced with loneliness, is to build a companion—an AI that is itself the friend. We think the prayer experiment suggests a different pattern: AI as the coordinator of human connection, not the substitute for it.

The generated-prayer fallback is the exception that tests the rule, and we want to be precise about how we hold it. It exists because an unanswered request is worse than a machine-written one, not because we think the two are equivalent. We treat it as a shortfall to be engineered away—by finding more pray-ers—rather than a feature to expand. The moment it becomes the product, this is a companion app with extra steps.

Prayer circles are a nearly perfect test case. The value is entirely human-to-human, the failure mode is entirely logistical—requests forgotten, follow-ups missed, homebound members left out as congregations shrink—and the population involved is the one most underserved by technology. If AI can quietly run the switchboard for something as intimate as prayer, the same pattern extends to meal trains, ride shares, visiting rotations, and every other structure of mutual care that runs on someone remembering.

The measure of success is not engagement with the assistant. It is whether Eleanor sleeps better knowing Frank is praying for her—and whether Frank, at 84, wakes up with something asked of him.

What this might mean for the future#

There's a common fear that AI will make us less human—that every hour with a machine is an hour taken from each other. The prayer line points at the opposite possibility. When the machine absorbs the logistics of caring, the caring itself grows: more people prayed for, more often, by name. The technology didn't replace a human act; it removed the friction that was quietly rationing it.

One way to draw the intuition is that a community's intention to care is roughly fixed, and coordination friction decides how much of it survives. If realized care C falls off exponentially with friction f, then the assistant is not adding intention; it is lowering the exponent.

That pattern generalizes. Most of what we call “community” runs on unglamorous memory work—who is sick, who is alone, who said they'd visit, who hasn't been heard from. When that work goes undone, connection decays not from lack of love but from lack of administration. An AI that carries the administration returns the love to circulation.

So the future we find interesting is not AI that becomes a better friend, confessor, or congregation. It is AI that makes us better at those roles—more faithful in our follow-through, more present to the people we already intend to care for. The measure of humane technology may turn out to be simple: after it enters a community, do the people in it act more like the people they meant to be?

Citation#

Please cite this work as:

Moses, Steve, "The Surprising Case for AI and Prayer", Co-Intelligence Labs: Thinking, Jul 2026.

Or use the BibTeX citation:

@article{moses2026prayer,
  author  = {Moses, Steve},
  title   = {The Surprising Case for AI and Prayer},
  journal = {Co-Intelligence Labs: Thinking},
  year    = {2026},
  month   = {jul},
  url     = {https://cointelligencelabs.com/thinking/the-surprising-case-for-ai-and-prayer}
}