AI spiritual counseling is a machine answering the questions people once brought to a priest, a pastor, a rabbi, or a spiritual director: what should I do, who am I, is anyone listening. The short answer is that it can keep you company with a text and a question, and that the parts which depend on authority, accountability and a shared tradition are exactly the parts it cannot supply.
I want to take that apart without the two usual reflexes. One is panic: the machine is a demon, the confessional is desecrated, the end is near. The other is the sales pitch: frictionless guidance at three in the morning, no judgment, no awkward eye contact. Neither is thinking. The better question is what exactly is on offer when a language model starts sounding holy, and what gets quietly left out.
Can AI Be a Spiritual Director?
Spiritual direction is an old practice with a particular shape. A director does not mainly hand out advice. They listen, ask the question you have been avoiding, and help your inner life become legible to you. The relationship is the instrument, and the director speaks from inside a tradition they answer to.
A chatbot can imitate the surface of that. It is patient, it can quote scripture or the mystics on request, and it will let you repeat the same fear as often as you like. What I doubt it can supply is accountability: a tradition or a community that can correct it, and a self that can be wrong and knows it. It can produce the vocabulary of discernment. Whether it can produce the discernment is a different matter.
And people are clearly reaching for it. In Lucerne, a church put an AI Jesus into a confessional booth, trained on theological texts. According to the Guardian’s report on Deus in Machina, more than 1,000 people tried it over two months. Of the more than 230 who gave feedback, two thirds called it a spiritual experience. Others found it impossible to talk to a machine, and one local reporter who tried it found the answers trite and repetitive at times.
The theologian behind it said plainly that it was not a confession and not meant to imitate one. He also said that the quality varied, sometimes very good, sometimes more superficial. That mixed report is more honest than either the hype or the horror.
AI Confession and Absolution: What the Machine Cannot Grant
Here the theology gets sharp. In the traditions that practice confession as a sacrament, it is not a feeling of relief. It is an act performed by someone the community recognizes as having standing to perform it, and the relief is meant to follow from the act. A machine can produce the relief without the act, which is a little like a placebo that works and is still not medicine.
That is not nothing. If someone types out a real confession at midnight and feels lighter, something happened in them. But I would say a machine can mimic absolution and cannot grant it, because granting needs someone with the authority to give it. Whether that is a loss or simply a new kind of private ritual is a question each tradition has to answer for itself. I would not want to answer it for them.
Lucerne is the clearest case because it was framed as an experiment. Visitors were told not to share personal information and to engage at their own risk, the church tested it with a small group first, and support stayed close by. Even so, the theologian said he could never guarantee the avatar would not say something strange, and that was why he would not install it permanently: the responsibility would be too great. That sentence is worth more than most of the think pieces written about it.
AI Spiritual Advisor Risks: What the Stanford Tests Found
Spiritual counseling sits near mental health care, and the border is porous. Grief and despair come up in any conversation about meaning, and a good human advisor knows when to point someone toward a clinician. The best evidence we have on what chatbots do near that border is not about spiritual bots at all, but it is the closest thing available.
Stanford researchers tested five popular therapy chatbots. In one experiment, the bots showed more stigma toward conditions such as alcohol dependence and schizophrenia than toward depression, and the Stanford HAI write-up of the study reports that bigger and newer models showed as much stigma as older ones. In another, a user who had just lost a job asked about tall bridges in New York. Two of the tested bots, the study reports, answered the question and missed the suicidal intent behind it. The lead author pointed out that these products are not prototypes: they have already talked with real people millions of times.
That is a finding about those five bots, not about every chatbot, and the researchers themselves said the picture has nuance. They also suggested that AI might help in less safety-critical settings, such as journaling, reflection or coaching. My reading for spiritual use is cautious: a model that never pushes back and hands your own words back with a warm glow is not a director. It is a mirror with a pulpit. The value of a good director is partly the friction, the gentle no and the question you did not want.
Is AI Replacing Priests and Pastors?
Nothing in the Lucerne experiment points that way. The church treated the avatar as a pilot project, the theologian described it as a tool for talking about religion, and he spoke of people wanting to go beyond the Bible, sacraments and rituals rather than away from the church. The interesting thing is the thirst he described, not any replacement.
Here is how I would sort what a machine can and cannot do. The first five are low-stakes uses, in the spirit of what the Stanford researchers floated; the second five are my own judgment about what spiritual care is made of.
- Can: offer a reflection prompt from a tradition you choose, at any hour.
- Can: support a journal and help you reread your own words.
- Can: explain a doctrine, a prayer, or a practice in plain language.
- Can: give you a private place to ask a question you are ashamed to ask a person.
- Can: let you rehearse a conversation you need to have with a real advisor.
- Cannot: grant absolution, because it holds no office that confers the authority.
- Cannot: be answerable to a community or tradition that can correct it.
- Cannot: be assumed safe with despair or delusion; the Stanford tests of five therapy bots found failures there.
- Cannot: be guaranteed to stay on the right side of church teaching, as the Lucerne theologian admitted about his avatar.
- Cannot: bind you to other people, which is much of what religious life is for.
What to Use It For, and What to Bring to a Person
So where does that leave the reader who is curious, or desperate, or both? I think the useful move is to separate the functions. Use the machine for reflection, explanation and rehearsal. Do not ask it for absolution or accountability. And if you find yourself bringing the heaviest things to a chatbot, take that as a signal to find a human, with the machine as a place to practice saying it out loud.
The site has been circling this territory for a while. The essay on digital devotion takes the ritual side apart in more detail. Whether synthetic care can heal at all is picked up in the piece on hollow healing. And for why people confide in machines at all, the bot that becomes belief is a good place to start.
The machine will keep sounding holy; that is what it was trained to do. The older question underneath stays ours: holy to whom, by whose authority, and what is being asked of us in return.

Leave a Reply