That is the part of the conversation that is missing.

The 73% figure comes from the 2026 State of AI in the Church survey. Barna’s better-resourced data is sharper still. Only about 1 in 20 churches has any AI guidelines at all. The number you have probably heard is conservative. The actual gap is wider.

But the framing points in the wrong direction. It suggests a void. There is no void. There is a policy. Every church has one.

The unwritten policy in most churches right now reads something like this:

Whatever staff are doing privately, plus whatever members are doing on their phones, with nobody talking about any of it.

That is the default. It has been adopted by your church. It is operating today. And like every default policy in every institution, it drifts toward the path of least resistance and the loudest adopter. Whoever uses AI most enthusiastically and least cautiously is, by sheer gravity, writing the policy that everyone else inherits.

The stakes are slower and higher

A law firm without an AI policy will eventually get sued. The stakes are visible. Lawyers know what their malpractice insurance will not cover. The market enforces a kind of discipline.

A church without an AI policy will not get sued. There is no malpractice insurance for forming people poorly. The stakes are slower, quieter, and harder to see. They are also incalculably higher.

What gets formed in your congregation, when AI is shaping prayer prompts and devotional habits and sermon prep and counseling notes, is formation under the influence of whoever trained the model. The models were built by people. Those people held convictions. The companies that funded the work held priorities. The training data ran from the Westminster Confession to TikTok hot takes about Bible interpretation, with the proportions of each weighted in ways no one outside the lab can see. All of that ends up shaping the words your members hear when they ask the model a question about prayer or grief or whether their marriage can survive. Your congregation is being formed by it, through the unwritten policy your church has already adopted.

The elder who uses AI to write his sermon and tells nobody is making a discipleship decision for his congregation. The decision may be defensible. It may even be wise. But the deciding has happened, and the decision has been made for the flock without the flock or his fellow elders being part of it. That is the part the unwritten policy hides.

Two elders are already in your church

Both of them are using AI. Or, if not yet, both of them are about to.

The first one has three jobs and a sick child and a marriage that is hanging together by grace. He finished his sermon at eleven o’clock last night because he asked Claude to outline 1 Peter 4 with him, and the outline gave him the structure he needed to actually get the text into his bones before Sunday morning. He has been faithful in his study for fifteen years. He preached a good sermon. The tool helped him do it. He feels relief, and he should.

The second one is convinced the church has been waiting for this moment for a thousand years. He is running counseling notes through ChatGPT to identify themes. He is asking Gemini to draft his weekly congregational email. He is using a custom GPT to generate prayer prompts for the prayer team. He thinks the elders are slow. He thinks anyone hesitant is afraid of progress. He has not stopped to ask whether what he is doing is faithful. He has only asked whether it is impressive.

Both of these elders need a framework. Neither one needs a sermon about how AI is the beast or the antichrist or the spirit of the age. The first one does not need to be made to feel ashamed of the help he received. The second one does not need to be allowed to disciple your congregation by accident. They need their fellow elders to think with them, slowly, in writing, about what is faithful and what is not.

Where most attempts at a church AI policy go wrong

They begin with a corporate template. They borrow the posture of an HR document. They start telling members what they may and may not do with AI in their personal lives. This happens because the templates floating around the internet were written for businesses, and businesses do have authority over what their employees do with company tools. The template gets imported wholesale, and the elders end up signing off on something that sounds reasonable until you ask the simple question of where their authority to bind any of this comes from.

The elders of a local church are shepherds. They are not managers. The Scripture that defines their work is not Harvard Business Review. It is Hebrews 13:17, where they are told they will give an account for the souls under their care. It is 1 Peter 5, where they are told to shepherd the flock of God among them, not domineering over those in their charge but being examples to the flock. It is Acts 20:28, where Paul tells the Ephesian elders to pay careful attention to themselves and to all the flock, and warns that fierce wolves will come.

There is a related problem worth naming before going further. In many churches, when an AI policy finally does get written, it does not come from the elders. It comes from a staff member who happens to be tech-fluent. It comes from a deacon in IT who volunteered. It comes from a committee that reports to the elders rather than from the elders themselves. The mistake is structural and theological. Writing the policy is itself a shepherding act. It is the work of those who will give an account for the souls entrusted to them. It cannot be delegated outward to staff or members, however technically qualified they may be, because they do not bear that weight. Staff and members can advise. They can flag the technical realities the elders need to understand. They can be invited to ask questions and offer counsel. They cannot write the policy that defines how this church’s ministry is conducted. That authority belongs to the shepherds because the account belongs to the shepherds.

Those texts authorize a great deal. They also leave a great deal alone.

What the elders have authority to bind is the conduct of their own ministry, what gets preached, taught, and published under the church’s name, the tools and data handled by anyone serving in pastoral roles, the way the church communicates with the body, the doctrine the congregation receives, and the disclosure they make to the flock about how the church’s ministry is being done.

What the elders do not have authority to bind is what members do with AI in their personal Bible study, devotional life, work, or homes. They cannot bind what tools members install on their phones. They cannot bind the internal conscience of members on matters Scripture has not addressed. Romans 14 is not an abstraction. Neither is 1 Corinthians 8 through 10. They are the texts that mark the line. The elders teach. They exhort. They warn. They model. They do not write rules where God has not written rules.

This means the policy your elders adopt is not a policy for your members. It is a covenant the elders make before God about how the elders will conduct the church’s ministry in the age of artificial intelligence. The discipling of your members around AI in their personal lives is a different work. It happens through preaching. It happens through teaching. It happens through the slow work of pastoral conversation. It does not belong in the policy.

What the policy actually has to do

With that line drawn, the policy has three jobs.

The first is to make the implicit explicit. The unwritten policy is operating because no one has said anything. The written policy operates because the elders have said something, in writing, that they will be held to. The point of writing it down is not bureaucracy. The point is so the congregation knows what is being done in their name, and so the elders know what they have committed to.

The second is to protect what cannot be delegated. There are things in pastoral work that do not belong to a tool because they do not belong to the pastor either. Counseling content. Member confidentiality. The act of being present to someone in suffering. These cannot be ghosted out to a model that may train on inputs, and they cannot be substituted by the model’s output, because the work itself is the point. A tool that lightens the load on Greek word studies is one thing. A tool that pretends to listen to a grieving widow is something else entirely.

The third is to distinguish a tool from a substitute. Most arguments about AI in the church collapse this distinction, in both directions. Some treat every use of AI as a substitute for pastoral work, and so they condemn the calculator as if it were a ghostwriter. Others treat every substitute as a tool, and so they let the ghostwriter through the back door because they liked using the calculator. The policy has to draw that line in writing, in specific cases, with the elders together signing their names to it.

A note before the model policy

What follows is a model. It is not a finished document for your church. The elders together will need to sit with it, adapt it, argue with parts of it, and shape it for the flock you have been given to shepherd. It also has to be revisited every year, with the date set in the policy itself, because the tools are changing fast enough that any policy written today will need to be reworked twelve months from now.

Use it as a starting place. Take what is faithful. Push back where you see something missing or wrong.

Model AI Policy for Local Church Use

Adopted by the elders of [Church Name] Effective: [Date] Next review: [Date one year from effective date]

1. Statement of purpose

We, the elders of this church, recognize that artificial intelligence tools have entered our work, our communications, and the lives of our members. We will not pretend otherwise. Nor will we use these tools without thinking about what their use means for the souls we have been called to shepherd.

This policy describes how we, the elders, will use these tools in the conduct of this church’s ministry. It does not bind the personal use of AI by members of this congregation. The discipling of members around AI in their own lives belongs to teaching and exhortation, not to policy. We will do that teaching faithfully. We will not write rules where Scripture has not written them.

2. Pulpit ministry

The preaching of God’s Word is the central act of the church’s worship. The faithful exposition of Scripture cannot be delegated to a machine.

The elders will not allow AI tools to write sermons that are then preached as if the preacher had written them. AI may be used as a study aid in sermon preparation: for outlining a passage, identifying themes across a book, surfacing exegetical questions, or checking historical and linguistic background. The work of grasping the text, framing the message, and bringing the Word to this congregation is the preacher’s own work, and remains his responsibility before God.

3. Pastoral care and counseling

The work of caring for souls in private cannot be delegated to a tool that may train on inputs, and cannot be substituted for by output that pretends to listen.

No identifying information about a counselee, no description of their situation, no sins confessed, and no personal details from a counseling encounter will ever be entered into any AI tool, public or private, by any elder or pastoral staff member, for any reason. This is categorical and admits no exceptions.

AI may be used in general preparation for counseling, such as reviewing biblical passages relevant to a category of issue, surveying counseling literature, or thinking through how to approach a topic in principle. It may not be used to process the specifics of any actual counselee’s life.

4. Resources produced for congregational use

Curriculum, study guides, devotionals, and other written resources produced under the church’s name shape the formation of the body. They are an extension of the elders’ teaching ministry.

Where AI is used substantively in the production of any resource that will be distributed to the congregation under the church’s name, that use will be acknowledged in the resource itself, and an elder will review the resource for theological soundness before it is released. Resources for which AI was used only for proofreading, formatting, or minor editorial assistance do not require specific disclosure beyond what we say in our standing statement.

5. Communications

Most weekly communications from the church are administrative in nature: announcements, scheduling, building updates, vendor correspondence, basic logistical emails. AI use in producing them is permitted and need not be disclosed.

A subset of communications, however, carries pastoral weight. These include letters to the congregation about church discipline, statements addressing public crises, condolence letters to grieving families, written responses to serious accusations, and other communications in which the body or an individual member receives pastoral counsel from the elders.

Pastoral communications, as defined above, are subject to the same disclosure principle as preaching. Where AI is used substantively in their production, the elders will note that use, either in the communication itself or in a manner the recipient is plainly aware of. AI will not be used to substitute for pastoral judgment in such communications. The judgment is the elders’ own.

6. Administrative use

Operations, finance, scheduling, vendor correspondence, building maintenance, and similar work that does not touch directly on the spiritual formation of members or the pastoral relationship may use AI tools as is helpful. No specific disclosure is required.

Common-sense limits apply. No member’s personally identifying information, financial information, or other confidential data will be entered into AI tools that may train on inputs. The church will use only tools whose data handling has been reviewed and approved by the elders.

7. Member confidentiality and data

The trust members place in the church is itself a stewardship.

No personally identifying information about any member, including but not limited to names paired with sins, struggles, family situations, financial circumstances, or pastoral concerns, will be entered into any AI tool. This applies to all elders, pastoral staff, ministry leaders, and any volunteer with access to confidential information. The single exception is the use of approved tools whose data handling agreements explicitly prohibit training on inputs and have been reviewed by the elders.

8. Disclosure to the congregation

The body deserves to know how its ministry is being conducted.

The elders will publish a standing statement to the congregation, reviewed annually, describing how AI is and is not used in this church’s ministry. This statement will be plainly available to members, and will be referenced from the pulpit at least once a year.

In addition, where AI has been used substantively in any sermon, pastoral communication, or counseling preparation that affected the body or an individual member, that use will be acknowledged in a manner appropriate to the context. The elders will not issue per-sermon footnotes that turn disclosure into theater. They will instead practice the kind of plain honesty that allows the congregation to trust them.

9. Decision-making

Decisions about which AI tools the church will use, which uses are appropriate, and how this policy will be interpreted in particular cases belong to the elders together, not to any individual staff member acting alone.

Adoption of new AI tools for any of the categories described above requires the approval of the elders. Individual staff members do not adopt tools for pastoral use on their own initiative. Where a question arises that this policy does not clearly address, the matter is brought to the elders for decision. The drafting and revision of this policy itself remains the work of the elders, who may seek counsel from staff and members but do not delegate that work to them.

10. Annual review

This policy will be reviewed by the elders no later than [date one year from effective date]. The tools, capabilities, and risks involved in AI are changing rapidly. Any policy written today will need to be reworked. The elders commit to that work, and to keeping the congregation informed of any substantive change.

The unwritten policy is already operating in your church. The question facing the elders is not whether to have a policy. The question is whether the one already in force is the one they would have adopted if they had read it first.

Read your church’s unwritten AI policy. Then write a better one.

None of this is hypothetical.

This is a snapshot of pastoral ministry in 2026. The men I am describing are not careless or unfaithful, but they are coming at this from very different directions.

Some of them are exhausted. Competent, conscientious, carrying a counseling load and a preaching calendar that have not given them an inch in years. AI showed up at the right moment, took some of the weight, and gave them back a few hours they did not have. They feel relieved, not triumphant, and the relief is honest. There is something in that relief worth taking seriously.

Others are the opposite. They feel like they have been handed a superpower. The pastor who used to prep one sermon a week is generating three. The man who used to write five emails a day is moving forty. He is launching a second campus, expanding the discipleship curriculum, drafting a book in his off-hours, and handling more pastoral correspondence than any one man could handle alone, because in fact he is not handling it alone. He feels like Superman. Some of what he is doing is genuinely good. Some of it is genuinely dangerous. The work of telling those two apart is the work this moment requires of him.

Both men are using AI. Both have thought about it. Of course they have. They are pastors. The question this moment puts to every man holding a charge is whether the thinking has been pressed against Scripture, surfaced with elders and peers, and allowed to settle into convictions before the AI use itself hardens into habit. That is a question this post will not answer for you. It will only put it in front of you, perhaps more clearly than it has been put before.

This is a field map, not a verdict. I will not tell you which uses are faithful and which are not. That is what the rest of this publication exists to work out, post by post. What I will do is name, with as much honesty as I can manage, what is happening right now.

Because what is unseen forms us anyway.

Sermon Preparation

This is the headline use case, and the one most Christians would be horrified to learn about if they did not already know. Sixty-four percent of pastors are using AI in some form to prepare what they preach on Sunday, and that number is almost certainly low. Adoption is moving at the pace of the models themselves, which is to say very fast and still accelerating. The range is enormous. On one end, a pastor uses a model to brainstorm illustrations or to check that he has not misread a Greek participle. On the other end, the entire sermon is generated by a system and lightly polished before it is delivered as the Word of God.

These are not the same thing, and any honest treatment of AI in the church has to start by saying so.

The full ethical argument on sermon preparation is the work of “When the Shepherd Didn’t Write the Sermon,” already published in this publication. If you have not read it, that is the place to go. For the purposes of this map, what you need to see is that sermon prep is the most public and most contested AI use in ministry, and the pastor who tells himself he is “just using it for grammar” needs to know that men he stands next to in his presbytery may be using it to do the entire homiletical work.

Here is the question that haunts the sermon-prep conversation, and I will let it do its work without trying to answer it now: when your people heard the sermon last Sunday, whose voice did they hear?

In John 10, the sheep follow the shepherd because they know his voice. They will not follow a stranger because they do not know the voice of strangers. There is a kind of recognition that happens between a shepherd and his flock that is grounded in particularity. The pastor who outsources the prep of his preaching to a model trained on the entire internet should not be surprised when his sheep slowly stop recognizing the voice that comes from the pulpit.

More on this in the dedicated post. For now, just notice: this is happening, and very few congregations have any idea.

Counseling and Pastoral Care

Pastors are using AI across the whole spectrum of pastoral care, and the spectrum is wide. Some of these uses raise questions that Scripture and Christian practice answer fairly clearly. Others sit in genuinely harder territory. I want to walk the range honestly rather than collapse it into a verdict.

At one end, a pastor asks a model for biblical counsel on a category of struggle without identifying any particular person. He is wrestling with how to address chronic anxiety in a believer who has not responded to the usual encouragements. He asks the model what the Puritans wrote on persistent fear, what passages bear on it most directly, what frameworks have been used to distinguish ordinary anxiety from clinical depression. This is much closer to consulting a book or a trusted colleague than to outsourcing pastoral care. The pastor still has to test what he reads against Scripture. He still has to apply it to a particular soul. He has not transferred a sacred trust to anyone. There is room to argue the wisdom of any individual case, but I do not see this as a category Scripture or pastoral practice rules out.

At the other end, a pastor uploads the verbatim transcript of a counseling session into a model and asks it to summarize and store the content for future reference. This is a different category of action. The Christian church has long held that what a man tells his pastor in confidence stays with the pastor, because the trust honest pastoral care depends on cannot be sustained any other way. Proverbs is direct about the moral weight of guarded speech: “Whoever goes about slandering reveals secrets, but he who is trustworthy in spirit keeps a thing covered” (11:13). Once a counselee’s words enter a third-party system that retains data for training, audit, or feature improvement, the seal has been broken in a meaningful sense, regardless of what the vendor’s terms claim. The Christian instinct here, formed by centuries of pastoral practice and grounded in plain Scripture about guarded speech, points one direction.

The harder territory is in the middle. A pastor drafts a follow-up email to a grieving member and runs the draft through a model to “make it sound less stiff.” A pastor asks a model to suggest ways to phrase a hard truth he needs to bring to an elder. A pastor takes his rough thoughts after a difficult meeting and asks a model to turn them into coherent pastoral notes. None of these obviously violate confidentiality if done carefully. None of them obviously substitute for pastoral presence. But all of them sit on a continuum that runs from “tool that polishes” to “tool that voices,” and the difference matters.

Scripture has more to bring to pastoral care than the confidentiality question alone. Paul writes to the Galatians, “Bear one another’s burdens, and so fulfill the law of Christ” (6:2). The verb is bastazō, to lift, to carry. It is a physical word. The Proverb gives the same picture from the other side: “Know well the condition of your flocks; give attention to your herds” (27:23). Both verbs describe a pastor paying attention to actual people in actual circumstances over time. AI is good at processing information about people. It is structurally something else from bearing burdens with people.

The test that has to be applied, case by case, is whether a particular AI use is strengthening the pastor’s actual presence with his people or quietly substituting for it. There is no general rule that handles every case. There is the careful, prayerful judgment of a shepherd who knows the difference between a tool that helps him serve his flock and a tool that lets him serve fewer of them less well while feeling more efficient. That judgment cannot be outsourced either.

Church Communications

This is the most pervasive AI use in the church, and it is harder to evaluate than it first appears.

Pastors are using AI for newsletters, social media, series titles, email blasts, bulletin copy, funeral announcements, and sermon graphics. The category is enormous and the practices inside it differ dramatically. A pastor who runs a Tuesday email through a model to fix typos and tighten sentences is doing something quite different from a pastor who lets a model write the sympathy note that goes out under his name to a grieving family. Treating these as the same thing is the first mistake.

Two questions are worth weighing.

The first is about voice. When churches use the same models for their communications, the prose they produce starts to sound alike. Most of the popular models default to a register that is warm, professional, and faintly therapeutic. Whether this matters depends on what voice the pastor was bringing in the first place. Some pastors have never had a strong written voice, and their congregations have never expected one. For others, the written voice has been a real part of how the congregation knows them across the week. Where that continuity is real, replacing it with a model’s version of warmth is a loss that is worth weighing carefully. Where it is not, the loss is harder to identify.

The second is about representation. Communications shape congregational expectations of who their pastor is. When the email going out under his name has been heavily generated by a system, the gap between the man who shows up at the hospital and the man whose name is on the email begins to widen. This matters more in some contexts than others. A short logistical email about the women’s retreat is one thing. A pastoral letter to the congregation after a tragedy is another. The closer the communication is to actual pastoral work, the more weight the representation question carries.

Scripture’s general counsel on the weight of words is worth holding nearby. “Death and life are in the power of the tongue” (Proverbs 18:21). Words are not just packaging in pastoral ministry. They are part of the work itself. The most defensible practice I can name in general terms is that the pastor know, for each communication going out under his name, what the model did and what he did, and that the closer the communication is to genuine pastoral care, the more of it should be his own work. Use AI to fix typos. Be more careful as the words approach the work of caring for souls.

Discipleship Tools

Here is the use case least likely to be on a pastor’s radar, because it is mostly happening outside what he directly sees.

Members of his congregation are using AI Bible apps to ask questions about Scripture. They are using AI prayer companions to talk through anxiety at midnight. They are using AI devotional generators to start their mornings. They are subscribing to AI-driven podcasts that match content to their listening history. They are downloading apps that offer to be their “spiritual companion.” Some have asked a model for counsel on their marriages. Some have asked a model whether they should leave their church.

This is not the future. This is what congregations are doing now.

The evaluative question is harder than the empirical one. Some of these uses look like the modern equivalent of consulting a study Bible or a Christian book. A believer asking AI for cross-references on a passage is doing something believers have done with concordances for centuries, with the obvious differences that the AI is faster, less reliable in places, and not authored by an identifiable Christian mind. Other uses are stranger and have no direct precedent. A believer who has long, intimate, daily conversations with a model about his prayer life and his marriage is in a category Christian formation has not had to evaluate before, and the church does not yet have settled categories for it.

What Scripture is clear about is that formation is meant to happen in the saturation of ordinary life with the truth of God, mediated through a community God has placed around the believer. Deuteronomy 6 describes it as something that happens “when you sit in your house, and when you walk by the way, and when you lie down, and when you rise” (v. 7). The Reformed tradition has always understood the local church as the ordinary means by which God shapes His people, through Word, sacrament, and the fellowship of the saints. Whatever else AI tools are doing in believers’ lives, they are not the church. They are not preaching in the appointed sense, not administering the means of grace, not bearing the keys of the kingdom.

The pastor’s question here is less about the tools themselves and more about scale. A believer shaped by twenty hours a week of personalized AI companionship and ninety minutes a week of corporate worship is being formed primarily by the former, even where his formal theology is being delivered by the latter. The faithful pastoral response is not to denounce the tools. It is to know what his people are using and to bring it into the light of his shepherding.

A separate post is coming on this question. (”Your People Are Already Being Discipled by Machines.”) For now, see it. The sheep are eating somewhere, and the pasture the pastor tends is not the only field they are in.

The Uses I Have Not Named

I have given the four most important categories. There are more, and the reader should know they exist even if I do not have space to address each here.

Pastors are using AI to draft elder board meeting agendas, summarize denominational reports, manage their calendars, generate policies, and compose internal staff memos. This is the operational layer of church life, and it is being quietly automated at scale. Some of it looks straightforwardly fine. Some of it sits closer to the questions raised in earlier sections, particularly policy work that touches doctrine or pastoral practice.

Pastors are using AI for exegesis itself. Greek and Hebrew parsing assistants. Cross-reference engines. Models trained on commentary databases that produce competent-sounding interpretations in seconds.

Pastors are using AI in their own personal devotional and emotional lives. Asking it to pray for them. Talking to it in the small hours when no one else is awake. Confessing, in a kind of way, to a system that cannot absolve. This category sits closest to the questions raised by AI in pastoral care, and it deserves its own essay.

I name these to make a single point. The map is wider than the four categories I have walked through, and the machine is in more rooms of the church than any single post can show.

What Is Forming Whom

The author of Hebrews writes that pastors “are keeping watch over your souls, as those who will have to give an account” (13:17).

That account is the frame I keep returning to. The model will not give an account on the day of judgment. The shepherd will. Whatever AI does in his ministry, the burden of having shepherded these particular people in this particular moment falls on the man who was called to it. He is accountable for what was formed on his watch, including any of it that has been outsourced to systems he has not stopped to examine.

This post does not give him the verdict on any particular use case. The whole publication exists to do that work, post by post, with the seriousness Scripture warrants and the honesty the moment requires. What this post offers is a more accurate picture of the field he is shepherding. The questions the map raises are his to take up, with his elders, with his peers, and with his Bible open. The work cannot be done for him.

The shepherd does not have to be a Luddite. He does have to be awake.

The machine is in his ministry. The watching, in this moment, is on what it is doing, what he is doing with it, and what is being formed in the people he has been given to shepherd.

You know this. Not because she told you. She hasn’t told anyone. You know it because you are her pastor. You noticed her jaw was tight during the pastoral prayer. You saw that she stopped lingering after the service. You remember that her husband’s scan results were due in October, and it is now January, and she has not mentioned them once. The silence told you everything.

No algorithm on earth could have seen that.

I want to be precise about what I am claiming here, because this is not a Luddite argument. I am not arguing that AI is useless to pastors. I’ve made the opposite case in this publication. AI is a powerful tool, and faithful pastors can use it well. But there is a category of knowledge that belongs exclusively to the embodied shepherd, and if we lose sight of that category, we will lose something that no technology can recover.

The Shepherd Knows His Own

When Jesus says, “I am the good shepherd; I know my own and my own know me” (John 10:14), He is making a claim about a particular kind of knowing. The Greek here is not the word for intellectual comprehension. It is the word for intimate, relational knowledge. The same word used for the deepest forms of human connection throughout Scripture. Jesus is not saying He has data on His sheep. He is saying He knows them the way a father knows his children.

Earlier in the same passage, He says something even more striking: “He calls his own sheep by name” (v. 3). In the ancient Near East, shepherds did not manage flocks from a distance. They lived with them. They slept in the field beside them. They knew individual animals by voice, by gait, by temperament. When one wandered, the shepherd did not send a report. He went after it himself.

This is not accidental imagery. It is one of the central metaphors for pastoral ministry across the biblical witness, and for good reason.

Ezekiel 34 delivers God’s indictment against the shepherds of Israel, and the charges center on embodied neglect: “The weak you have not strengthened, the sick you have not healed, the injured you have not bound up, the strayed you have not brought back, the lost you have not sought, and with force and harshness you have ruled them” (v. 4). Every verb is physical. The passage includes other failures too, selfishness and exploitation among them. But the heart of the indictment is that the shepherds were not doing the hands-on work of caring for actual sheep. They were absent from the lives they were called to tend.

Paul picks up the same thread in 1 Thessalonians 2:8: “So, being affectionately desirous of you, we were ready to share with you not only the gospel of God but also our own selves, because you had become very dear to us.” Not just the gospel. Our own selves. Paul uses the language of a nursing mother in the verses before this, and the language of a father in the verses after. The point is total investment. Body, soul, time, presence. Not content delivery. Self-giving.

This is what a shepherd does. He gives himself to his people. And you cannot give yourself through a language model.

What the Machine Can Do (and What It Cannot)

Let me grant every advantage to the technology, because I think intellectual honesty requires it.

A large language model can process the entire canon of Scripture in milliseconds. It can cross-reference every mention of grief in the Bible, pull up pastoral care frameworks from a dozen traditions, generate a counseling response grounded in biblical categories, and do it all before you have finished your first sip of coffee on a Monday morning.

If a church member asks an AI chatbot, “I’m struggling with grief after losing my mother,” the response will likely be competent. It might even be theologically sound. It will probably reference passages like Psalm 34:18 and 2 Corinthians 1:3-4. It might suggest the language of lament. It will be gentle, measured, and available at 2 a.m. when no pastor is answering his phone.

I am not dismissing that. For someone who has no church, no pastor, and no community, a competent AI response to a moment of crisis is better than nothing. I believe that.

But here is what the machine cannot do.

It cannot remember that this particular woman lost her first child thirty years ago and that every subsequent loss reopens that wound in a way she has never been able to name. It cannot know that she processes grief through silence and withdrawal rather than tears, and that the people around her often miss it because she looks fine. It cannot recall that the last time she went through something like this, it was her pastor sitting in her living room saying nothing for twenty minutes that finally broke through. Not a word of Scripture. Not a counseling technique. Just presence. Just the embodied reality of a man who knew her well enough to know that she needed someone to sit in the wreckage with her.

The machine cannot know any of that because the machine does not have a body, does not have a history, and does not have a relationship. And pastoral ministry, at its irreducible core, requires all three.

What You Know but Cannot Say

There is a kind of knowledge that resists being written down. We all carry it, even if we rarely think about it in those terms.

A master carpenter knows when a joint is right by feel. A physician knows something is wrong before the lab results confirm it. An experienced mother knows the difference between her child’s tired cry and his pain cry, and she could not write down the criteria she uses to distinguish them if her life depended on it. We can know more than we can tell.

Pastors carry this kind of knowledge about their people. You know which elder handles confrontation well and which one shuts down. You know which couple’s marriage is strong enough to weather a hard conversation and which one needs six more months of trust before you push. You know the teenager who uses humor to deflect pain and the one who uses it because he is genuinely joyful. You know these things not because someone entered them into a database. You know them because you have been present, year after year, in the ordinary rhythms of their lives.

This is not mystical. It is not a spiritual gift that falls outside ordinary categories. It is the accumulated wisdom of incarnate presence. It is what happens when a shepherd actually lives among his flock.

And it is precisely the kind of knowledge that cannot be transferred to a machine. Not because the technology is not advanced enough yet. Not because we need better data input. But because this knowledge is rooted in bodily presence. It exists in the space between persons. It is generated by presence and sustained by relationship. Remove the body, remove the history, remove the ongoing life together, and the knowledge does not exist.

The Puritans understood this. Some of the best pastors in the Reformed tradition built their entire model of ministry on visiting every family in their parish, knowing their circumstances, understanding their struggles. One seventeenth-century pastor visited hundreds of families in his parish yearly. Not because he was heroic, but because he believed pastoral knowledge required physical proximity. You cannot shepherd people you do not know, and you cannot know people you are not among.

The Danger of the Competent Substitute

Here is what concerns me most about AI in pastoral contexts, and it is not what you might expect.

I am not worried about pastors being replaced by machines. No serious church leader is going to install a chatbot and fire the pastor. That is a straw man, and it distracts from the real danger.

The real danger is subtler. It is the gradual erosion of the expectation that a pastor should know his people.

When competent AI responses are available at scale, at any hour, without the inconvenience of scheduling or the awkwardness of vulnerability, the pressure on pastors to do the slow, costly work of personal knowledge will decrease. Not because anyone decides it does not matter. But because the availability of a good-enough substitute will quietly reshape expectations. If someone can get a thoughtful, biblically grounded response to their crisis from an AI at 2 a.m., the question “Why do I need to call my pastor?” starts to feel reasonable.

Barna’s research bears this out. Thirty percent of U.S. adults already say AI spiritual advice is as trustworthy as advice from a pastor. Among younger adults, that number climbs to nearly forty percent. And sixty-five percent of pastors themselves worry that AI could displace their spiritual guidance role.

Those numbers should not make us defensive. They should make us serious. Because the answer to “Why do I need my pastor when I have a chatbot?” is not “Because AI is bad.” The answer is that your pastor carries something the chatbot never will: knowledge of you. Real, specific, hard-won knowledge earned by years of showing up.

The chatbot does not know that you always volunteer for things you do not have capacity for because you are terrified of disappointing people. Your pastor does. The chatbot does not know that your theological questions are actually a way of avoiding the emotional conversation you need to have. Your pastor sees it. The chatbot does not know that when you say “I’m fine,” you mean the opposite, because it has watched you say “I’m fine” for three years and it has seen what fine actually looks like on you.

That is pastoral knowledge. And it is irreplaceable.

What This Means for Shepherds

If the argument I am making is right, and I believe it is, then it has practical implications for how pastors spend their time.

It means that the thing no machine can touch is not the most visible part of your ministry. Preaching is the primary calling. Scripture is clear on that, and the Reformed tradition is right to insist on it. But faithful preaching has never been disembodied. It has always depended on a shepherd who knows his flock well enough to bring the Word to bear on their actual lives. That knowledge is built through the slow, unimpressive work of being present. Remembering what someone told you three months ago and asking about it. Sitting in hospital rooms and living rooms and coffee shops, accumulating the kind of understanding that only comes from showing up, again and again, in the ordinary and extraordinary moments of a congregation’s life.

AI cannot do that. It will never be able to do that. And in a world that is increasingly captivated by the efficiency and availability of machine-generated counsel, the pastor who still does the slow work of presence will be more valuable, not less.

Bonhoeffer built his entire model of seminary education around this kind of presence. The best Puritan pastors wore out their shoes walking to homes in their parishes to practice it. The biblical witness is unambiguous: the shepherd is the one who is there.

So use AI for what it is good at. Let it help you research, cross-reference, organize, and draft. But do not let the efficiency of the tool deceive you into thinking that the most important parts of your ministry can be made efficient. They cannot. The woman in the third row does not need a better algorithm. She needs her shepherd to see her. And that requires something no machine will ever possess.

A body in the room. A history with her pain. And the kind of knowledge that can only be spoken by someone who has been there all along.

Subscribe now

He's not a fraud. He's not lazy. He's a man with 30 hours of work crammed into 20 available hours, a hospital visit on Saturday afternoon, and a sermon that needed to land by Sunday morning. So he opened ChatGPT, fed it his passage, and asked for help.

If you're a pastor reading this, I need you to hear something before we go any further: I'm not here to condemn that man. There's a decent chance that man is you. I've stood behind a pulpit. I know the weight of Sunday morning bearing down on a week that gave you nothing. According to the 2025 State of AI in the Church survey, 64% of pastors who preach now use AI in sermon preparation. That's up from 43% just one year earlier. Sixty-one percent of pastors report using AI weekly or daily. This is not an emerging trend. It has already emerged. The only question left is whether we'll be honest about it.

The Tool Question

Let's get the easy part out of the way first, because I think a lot of the anxiety around this topic comes from a failure to think clearly about categories.

AI is a tool. Full stop.

I know that sounds reductive, and I know there are important conversations to have about what kind of tool it is and what it's becoming. We'll have those conversations in this publication. But for the purpose of sermon preparation, we need to start with the obvious: pastors have always used tools.

You use commentaries. You use lexicons. You use sermon software that cross-references Scripture and pulls up word studies with a click. You use books written by dead theologians who did the hard exegetical work centuries before you were born. If you've ever borrowed an illustration from Spurgeon, you used another man's words to make a point to your congregation. Nobody called it a crisis of pastoral integrity.

The question has never been whether pastors use tools. The question is what role the tool plays and what role the shepherd plays.

When you open Matthew Henry's commentary, you're engaging with another mind that wrestled with the text. You read his conclusions, weigh them against the passage, argue with him in the margins, and then form your own position that you bring to your people on Sunday. The tool served the shepherd. The shepherd still did the shepherding.

AI can function in exactly the same way. I've used it to surface cross-references I hadn't considered. I've asked it to summarize scholarly positions on a difficult passage so I could evaluate them faster. I've had it identify structural patterns in Hebrew poetry that would have taken me an hour to trace manually. In each case, the tool served my engagement with the text. It didn't replace it.

But here's where we have to be honest with each other.

The Line Nobody Wants to Draw

There is a difference between using AI as a research tool and using AI as a ghostwriter. And that difference matters. Not because the tool is inherently sinful, but because the act of preaching is inherently personal.

When Paul wrote to Timothy, "devote yourself to the public reading of Scripture, to exhortation, to teaching" (1 Tim. 4:13), he was not describing the delivery of content. He was describing the overflow of a life devoted to the Word. Two verses later he says, "Practice these things, immerse yourself in them, so that all may see your progress" (v. 15). The congregation is supposed to see the pastor's own progress. His own growth. His own wrestling.

This is not a minor point. The Reformed tradition has always understood preaching as more than information transfer. It is the means by which God speaks to His people through the called, gifted, and prepared minister of the Word. The Westminster Directory for Public Worship charges the minister to preach from his own study of Scripture, drawing on his knowledge of his people and their condition. There's a reason for that. The sermon is not just an essay delivered aloud. It is a pastoral act.

So when a pastor prompts ChatGPT to "write a 25-minute sermon on Romans 8:28 for a congregation dealing with grief" and then delivers that output with minor edits from the pulpit, something has broken. Not because the theology is necessarily wrong. AI can produce surprisingly competent Reformed theology on a good day. Something has broken because the congregation is receiving a product where they were owed a process. They came to hear their shepherd's voice, and they got a machine's approximation of it.

The line is not complicated, even though we want it to be. If you're using AI to help you study, think, research, and prepare, you're using a tool. If you're using AI to produce the thing your congregation thinks you produced, you're not using a tool. You're outsourcing your calling.

The Commentary Objection

I can hear the pushback already: "How is this different from using a commentary?" It's a fair question, and it has a clear answer.

When you use a commentary, you know you're engaging with another theologian's work. You cite it or you don't, but everyone in the room understands that pastors read books and those books shape their preaching. The commentary is a known category. It's part of the ecosystem of sermon preparation that congregations have always understood and accepted.

AI is different because it is invisible. When a pastor uses AI to generate sermon content, the congregation has no framework for understanding what just happened. They assume the words came from their pastor's study of the text. They assume the illustrations were drawn from his experience or reading. They assume the application was shaped by his knowledge of their lives. None of those assumptions may be true, and the pastor knows it.

The issue is not the tool. The issue is the assumption gap between what the congregation believes happened and what actually happened. And that gap is an integrity problem, not a technology problem.

What This Actually Requires

I'm not calling for pastors to swear off AI. That ship has sailed, and frankly, I don't think it needed to stay in port. AI is a powerful tool that can serve faithful ministry in real ways. What I'm calling for is something harder than abstinence. I'm calling for honesty.

First, be honest with yourself. You know the difference between studying with a tool and letting a tool do your work. Your conscience is not confused about this, even if you've been telling it to be quiet. If you cannot point to the place in your sermon where your own wrestling with the text shaped the message, you have a problem that has nothing to do with technology and everything to do with your calling.

Second, be honest with your elders. If your church doesn't have a leadership conversation about AI in sermon preparation, it needs one. Not because you need permission to use a research tool, but because secrecy about tools and methods is corrosive to trust. Your elders should know how you prepare. They should know what role AI plays. And they should be able to hold you accountable for the integrity of your pulpit ministry.

Third, and this is the harder conversation: be honest with your congregation. I'm not suggesting you need a disclaimer before every sermon. But your people deserve to know, at a general level, how their pastor prepares to preach to them. If AI is part of that process, they should know it. Not because it's shameful, but because transparency is the foundation of pastoral trust.

Barna's research found that 49% of pastors are concerned about the loss of authenticity in preaching. Nearly a third of practicing Christians say they trust spiritual advice from AI as much as from their pastor. Those two statistics should keep every shepherd up at night. Not because AI is a threat, but because the church is losing the ability to distinguish between the authentic and the generated. And the pulpit is the last place that distinction should be allowed to collapse.

The Real Danger

The deepest danger of AI in sermon preparation is not that pastors will use a tool. It is that pastors will stop doing the work that the tool was supposed to assist.

Sermon preparation is formative. The hours spent in the text, the frustration of a passage that won't yield its meaning easily, the moment when the Spirit illuminates something you've read fifteen times before: that process shapes the preacher. It forms him. It is part of how God sanctifies the man who stands before His people. When you skip that process, you don't just get a worse sermon. You get a less formed shepherd.

Every preacher who has ever labored over a passage on Friday and felt it break open on Saturday knows exactly what I'm talking about. I have. That breakthrough was not just for your congregation. It was for you. It was God doing something in you so He could do something through you. No machine can replicate that process, and no machine should be asked to replace it.

Where We Go from Here

AI is not going away. Its capabilities are accelerating. By next year, every major Bible software platform will have AI integration that makes today's tools look primitive. The question for pastors is not whether they will encounter AI in their workflow. The question is whether they will be shepherds who use tools wisely or consumers who let tools use them.

Use AI to research. Use it to check your exegesis. Use it to find the quote you half-remember from a theologian whose name you've forgotten. Use it to identify patterns in the original languages. Use it to draft your bulletin and your email newsletter and your small group discussion questions. These are good and legitimate uses of a powerful tool.

But when you stand behind the pulpit and open your mouth to preach the Word of God to the people God has entrusted to your care, make sure it's you. Make sure you did the work. Make sure you wrestled with the text, sat in the difficulty, and came out the other side with something God gave you for your people. Not something a machine generated for a generic audience.

Your people can tell the difference. They might not be able to articulate it, but they can feel when their shepherd is speaking from the overflow of his own encounter with God. And they can feel when he isn't.

Share

The machine is already in the church. The question is whether the shepherd is still in the sermon.

Sit with that number for a moment. Not as a statistic to debate. As a reality to reckon with. More than six in ten men who stand behind pulpits in evangelical churches, men whose calling is to open the Word and feed the flock, are, to some degree, letting a machine help them do it. And in almost every case, the congregation has no idea.

Seventy-three percent of those same churches have no policy governing any of this.

The machine is already in the church. It has been for a while now. The question was never whether it would arrive. The real question, the one almost no one is asking seriously, is what we are going to do about it now that it has.

I am not here to sound the alarm. I am not a technophobe who has wandered into theology. I find artificial intelligence fascinating, and I think we are living through one of the most significant inflection points in human history. The printing press reshaped the church. The Industrial Revolution reshaped the family. AI will reshape everything. The church that treats this as someone else’s problem will find itself shaped by it anyway, just without having decided how.

What I keep seeing, in conversations with pastors and elders and seminary students, is one of two things. Either a reflexive defensiveness (we shouldn’t touch this, it’s dangerous, it’s not real ministry) or a casual adoption that never stops to ask whether there is anything at stake. Both responses fail the flock.

The pastor who refuses to think about AI because it feels like a threat is not protecting his congregation. He is just letting it happen without him. The pastor who uses it every week without asking what it means for his integrity, his calling, his relationship to the Word, is making a different kind of mistake. Neither one is shepherding.

There is a question underneath this that matters more than the technology itself.

What is a pastor for?

Not functionally. We can answer that easily enough with job descriptions and ordination vows. I mean theologically. What does it mean to be the shepherd of a particular flock, in a particular place, in a particular moment in history? What is the relationship between a man’s private wrestling with the text on Tuesday morning and the sermon he delivers on Sunday? What is at stake when the Word that is supposed to be dwelling richly in him is instead being processed by a model trained on the entire internet?

These are not rhetorical questions. They are the questions this publication exists to answer, or at least to ask with the seriousness they deserve.

Here is what you will find here.

Posts that deal with the real intersections of AI and ministry. Not hypothetical futures, but present realities. Sermon preparation. Pastoral care and counseling. Congregational discipleship. Church policy. What it means to shepherd people who are being formed spiritually by tools you did not choose and may not understand.

Posts that are theologically serious without being academic. Written by a pastor for pastors. The working shepherd with twenty-three people in his counseling queue and a sermon to preach in four days does not need a journal article. He needs someone who has thought about this carefully to tell him, straight, what is actually at stake and what faithfulness looks like from here.

Posts that are honest about complexity. Some of this is hard. The lines are not always obvious. A pastor who uses AI to clean up his grammar and a pastor who lets AI write his sermons wholesale are doing very different things, even if the surface behavior looks similar. I will not pretend the distinctions are simple. But I will try to make them clearly.

Here is what you will not find here.

You will not find the argument that the faithful response to AI is disengagement. The machine is here. It is not going away. The question is not whether to engage. That ship has sailed, whether you know it or not. The question is how to engage faithfully.

You will not find Silicon Valley cheerleading. I am not interested in convincing you that AI is a gift from God that will supercharge your ministry. That may or may not be true depending on how you use it. Tools are not neutral, but they are also not destiny.

You will not find culture-war framing. AI is not the mark of the beast. It is also not the Second Coming of Productivity. It is a profound technology that is reshaping human life in ways we do not fully understand yet, arriving in our churches at a moment when almost no one is theologically prepared for it.

A word about what I believe.

I am a Reformed theologian, which means I take Scripture seriously as the final authority on everything it addresses. The questions we are dealing with: what it means to be human, what image-bearing means, what the relationship between language and truth is, what pastoral calling requires. These are not questions that Scripture answers in the form of an AI policy. But they are questions that Scripture addresses, and any serious account of AI in the church has to start there.

Faithfulness to the text comes before loyalty to any tradition. I will draw on the Reformed heritage where it illuminates, but I am not interested in defending confessional positions that may or may not speak to the present moment. I am interested in what the Bible actually says and what it means for the man standing behind the pulpit in 2026.

The machine is already in the church.

Your congregation is being discipled by algorithms, whether you know it or not. Your people are using AI for prayer prompts, devotionals, spiritual advice, and answers to theological questions. Some of them are sophisticated enough to know what they’re doing. Most of them are not. And the man whose calling is to shepherd their formation, to know the grief in the third row, to remember what he said to the widow six months ago, to speak the Word into the specific contours of a specific life, that man needs to think carefully about what is happening, and what faithful presence in this moment actually requires.

That is what this is for.

Glad you’re here.

Ken Duffy The Church in the Machine