How Pastors Can Use AI Without Losing Their Voice

Losing your voice can mean two different things, and pastors tend to collapse them into one. The first is authenticity: sounding like a machine instead of yourself. The second is authorship: letting something else do the thinking and then signing your name to it.

You can use AI heavily and keep both, or use it lightly and quietly give the second one away. Which happens depends on the tasks you hand over and how firmly you hold onto the ones that form you.

I am a working Episcopal priest in Hell's Kitchen, and I build these workflows for my own parish before I teach them to anyone. That is what Handbuilt is. Most of the pastors I talk to are already using these tools and almost none of them will say so out loud, because the private worry underneath the silence is exactly this fear about voice. This piece is my attempt to answer it honestly, from inside the work.

What "Losing Your Voice" Actually Means

The fear worth naming precisely is authorship, even when a pastor describes it as authenticity. Authenticity is a real concern, and it is the easier of the two to solve. If a draft sounds generic, you can hear it, and you can fix it by rewriting until it sounds like you again. The problem announces itself.

Authorship is quieter and harder. It is the slow substitution of the model's thinking for your own, where you stop wrestling with the text because a competent summary is already sitting on the screen. Nobody notices that in a single sermon.

It shows up over months, as the muscle you used to bring to a passage goes soft from disuse. When pastors tell me they are afraid of losing their voice, this is usually what they mean underneath, even when the words that come out are about tone.

I can feel the difference in my own week. If I let a model summarize a passage and lightly edit what it hands back, the sermon comes out competent and a little hollow, because I did not earn it; doing the reading first and using the tool only to check my work or draft the connective tissue between points leaves the same amount of AI in the process, but the thinking stays mine. The amount of help was similar both weeks. What differed was whether I had done the thinking before the tool touched it.

AI can imitate your style well enough that style was never the interesting question. What matters is whether you are still the one deciding what the text means and what your people need to hear.

How Many Pastors Are Already Doing This Quietly

Using AI in ministry is already majority behavior. Barna's 2026 research on pastors found that only 13% report using no AI at all, and sermon writing with AI has roughly doubled since 2024, from 12% who were comfortable with it then to 24% actually doing it today, still treated more as a research partner than an author [1].

Ministry task	Share of pastors using AI for it
Brainstorming and idea generation	50%
Graphic design and visual creation	37%
Researching biblical or theological topics	36%
Small-group questions and admin tasks	34%
Writing or editing sermons	24%

Comfort drops the closer a task sits to preaching, which the table makes plain. A separate 2026 industry survey put overall engagement higher still, finding that around 93% of church leaders now use or actively explore AI in ministry, with close to two-thirds of preaching leaders folding it into sermon prep, a rate that has held steady for two years rather than climbing [2].

Governance has not kept pace. Barna's church-leader research found that only 5% of churches have any AI guidelines in place at all [3]. Most pastors are working this out alone, one week at a time, without a policy to lean on. That is part of why the caution runs so deep: 71% of pastors describe feeling cautious about AI, far above the rate among churchgoers generally [1].

Where Your Voice Actually Lives in Pastoral Work

Your voice lives in two places a model cannot reach: the judgment you bring to a text and the presence you bring to people. Name those concretely and the fear gets smaller, because you can see what you are actually protecting.

Judgment and interpretation

A model can tell you what a passage has meant across the tradition, but deciding what it means for these people this week is a judgment call no model can make. That decision is the interpretive work at the center of preaching, and it depends on knowing who is in the room: which families are carrying fresh grief, which tensions are running quietly under the surface, the season your particular congregation is in.

No model has read that room. I can hand a workflow every commentary ever written and it still will not know which word my particular congregation needs to hear this Sunday.

Presence and pastoral care

The other half of the work is not textual at all. Sitting with a grieving family, noticing which parishioner has gone quiet, sensing when someone is drifting away before they say a word: none of that survives a handoff to a machine, because it runs entirely on knowing people a model has never met.

Ed Stetzer names the same limit as a basic disconnect between the tool and a congregation it never encountered [4]. The care that depends on memory and relationship stays human, and I would not want to serve a church that thought otherwise.

The Line Between Assistance and Authorship

Nearly every serious voice on this subject draws the same boundary in its own words: AI can gather and draft, the pastor decides and delivers. It shows up as "sous chef," as "assistant," as lists of principles, but the shape is consistent enough that it reads like a genuine professional consensus rather than one writer's opinion.

Task type	Appropriate AI role	Where the line holds
Research and background reading	Gather sources, summarize positions, surface citations to verify	You still read the text and decide what it means
Administrative writing	Draft routine emails and announcements for review	Nothing pastoral is delegated, only the paperwork
Design and production	Build graphics and page layouts	The message being designed is still yours
Sermon and teaching content	Draft support and structure you then rework	You do the study and write what you preach
Pastoral care and counseling	None	Care depends on knowing this specific person

The comfort gradient in the data tracks this line closely. Even in Barna's earlier 2024 numbers, 88% of pastors were comfortable using AI for graphic design, while only 12% would use it to write a sermon and just 6% would use it as a counseling tool [5]. Pastors sorted these tasks correctly before anyone handed them a framework. The instinct is already good; it mostly needs language and permission.

Why a Chat Window Alone Wears Your Voice Down Over Time

A chat window resets to zero every time you close it, so the help never compounds. The real risk is subtler than one inauthentic paragraph. Undirected prompting, repeated week after week, slowly lets the model's phrasing and framing stand in for your own.

The ceiling of prompting

A chat window remembers nothing between sessions. Every Monday you start from zero, re-explaining your context, your parish, and your theology to a blank box that will forget all of it by Tuesday. That is genuinely useful, and it is also a hard ceiling. Most pastors think this is what "using AI" means, because the chat window is the only door they have been shown. They do not know there is anything past it.

What gets thinner with repeated offloading

Here I want to stay careful with the claim: skip the brain-scan studies and just note the plain version, which carries the weight on its own: a skill you stop practicing is a skill you slowly lose. If the model does your exegetical spadework every week, the part of you that used to do it gets less practice, and over a few years that adds up to a real difference in what you can do without help.

The tools stay welcome in the rest of the process. The discipline is simply to keep doing the parts of the work that keep you sharp, week after week, rather than quietly handing all of it over.

From Prompting to Workflows: What Changes When AI Learns Your Voice

Your voice is best protected by building AI around your own words instead of a generic model's. The jump that changed my ministry was moving from prompting to workflows: agents I set up once, trained on my own sermons and the standing facts of my parish, that then run the repetitive parts of my week without me starting over each time.

Building one is less technical than it sounds. You give the agent examples of your own writing and the fixed facts of your ministry, and you tell it the job you want done. From then on it works from your material rather than a stranger's, and you refine it the way you would train a new assistant, correcting what it gets wrong until it gets it right.

The skill you are building is direction: telling a capable tool exactly what good looks like in your hands. That skill transfers across every workflow you build after the first one, which is why the second automation always takes less effort than the first.

The difference is what the tool is working from. A blank chat session pulls from the average of everything on the internet. A workflow I have built pulls from my material, so what comes back already sounds like my parish rather than a generic church. Here is some of what that looks like in practice, none of it anywhere near a sermon or a hospital bed:

• A full parish brand standards guide, plus business card and letterhead designs, built inside one week.
• A bespoke dashboard that pulls from scattered systems and surfaces the real-time numbers I actually need to watch.
• Invoice-management automations that handle what used to eat an afternoon.
• A hymn database that stays fetched and maintained on its own.
• Research and document drafting that lands as clean, finished PDFs.
• Email responses drafted for my review, and never sent unread.

Every one of those gave me back time that now goes to people. Christopher Benek, a pastor who has written about this from inside his own ministry, reports that AI cut his sermon-prep time by roughly 87.5%, and he still insists the tool cannot do the formation work of prayer and study [6]. That is the pattern to aim for: the machine takes the weight that was never sacred, and you spend the reclaimed hours on the parts that are. This is the gap I built the Handbuilt Lab to close, teaching pastors to build and own these workflows for themselves.

Practical Guardrails That Keep the Work Yours

A few plain habits keep authorship honest without turning every sermon into a compliance exercise. None of these require a committee or a policy document. They are the working discipline I actually use, and they line up with the guidelines other practicing pastors have landed on independently.

• Keep the study central. If a task forms you as a preacher, do it yourself and let the tool wait outside the door.
• Verify everything. A model will produce a confident sentence whether or not it is true, and most pastors already know it: 84% say AI-generated content needs editing because of errors, the single most common worry in the research [7]. Check every fact and citation before it reaches anyone.
• Disclose plainly when it matters. You do not need a disclaimer every Sunday, but you should be able to say without flinching which parts of a piece came from a workflow and which came from you.
• Never let a generated paragraph stand in for study you did not do. Treat the draft as a starting point for your thinking, something to test and rework rather than hand over untouched.

Christopher Benek's own guidelines land in the same place, and Ed Stetzer's practical standard reads the same way from a different tradition [4][6]. When people who share little else agree on the boundary, that convergence is exactly what makes me trust it.

A Quick Self-Check Before You Use AI Near the Pulpit

Before I use a tool for anything close to ministry, I run four questions in my head. You can do the same in about ten seconds.

Ask before you start	What the answer tells you
Is this task administrative, or is it formation?	Formation work stays with you; admin work is fair to hand over
Whose material is the model working from, mine or a stranger's?	A workflow trained on your own words protects your voice; a blank chat does not
If a parishioner asked, could I describe the tool's part plainly?	If the honest answer has to stay hidden, that is already the answer
Have I checked every fact and citation it produced?	Unverified confidence is the fastest way a tool embarrasses you

Clear all four and the use is fine; fail even one and I stop to figure out why before going further, rather than pushing ahead on a guess.

The Real Answer to "Without Losing Their Voice"

What protects your voice is staying in charge of the work that carries it: the interpretation, the study, the presence with people, and the final call on what to say. Hand the machine the administrative and creative weight that sits between you and that work, build your tools around your own material rather than a generic model's, and your voice comes out stronger because you have more time to spend on the parts only you can do.

If you are already using these tools and want to move from typing prompts into a blank box to building workflows you own, that is what the Handbuilt Lab is for. The first cohort runs four weeks beginning the second week of September, capped at fifteen pastors, and goes by invitation rather than an open sign-up, because working out where your own line sits takes a real conversation about your ministry. So book a call and we can talk it through. If you want the ethics of all this laid out first, I wrote about whether it is wrong for a pastor to use AI in a companion piece.

FAQ

Will using AI make my preaching sound less like me?

Only if you let a generic model do the writing and stop editing. A blank chat session pulls from the average of everything online, so its default voice is nobody's. When you build a workflow around your own sermons and parish context, and you rework the draft until it sounds right, the output tracks your voice instead of erasing it. The tool follows your lead as closely as you make it.

Is it dishonest to use AI in ministry writing?

Not in itself. Using a model for research or first-draft support is an ordinary use of a tool, the same as using a commentary. Dishonesty enters only when you present AI-generated work as your own unaided study and hide that you used it. The fix is simple: be able to say plainly how a tool contributed, and never let a generated paragraph stand in for study you skipped.

What should I never hand to AI as a pastor?

Pastoral care and counseling. A model does not know your people, and care depends entirely on knowing them: their history, their grief, the details that never make it into a summary. Interpretation and application belong in the same protected category, because deciding what a text means for this congregation this week is the core of the work only you can do.

Do most pastors already use AI?

Yes. Barna's 2026 research found only 13% of pastors use no AI at all, with brainstorming and behind-the-scenes work leading the way. For most pastors the live question is no longer whether to use it but how to use it with integrity, since they have already started.

What is the difference between using AI for research and using AI to write for me?

Research keeps you as the author: the tool gathers sources and surfaces options, and you still read the text, decide what it means, and write what you preach. Letting AI write for you moves the thinking itself off your desk, so you deliver a message you did not actually work out. Research is a faster reference shelf; letting AI write for you is where your voice quietly goes missing.

How is building a workflow different from just using ChatGPT?

A chat session starts blank every time and pulls from the average of everything online, so you re-explain your context each week and the output sounds generic. A workflow is set up once and trained on your own material: your sermons, your parish, your commitments. It runs the repetitive parts of your week and returns drafts that already sound like you. Most pastors only know the chat window, and the workflow is the room past it.

References

[1] Barna Group. "New Research: Pastors Are Using AI More Than You Think." Barna Group, June 15, 2026. https://www.barna.com/research/pastors-using-ai-ministry/

[2] Spencer Jahng. "The 2026 State of AI in the Church: 93% of Pastors Are In. Now What?" ChurchTechToday, April 27, 2026. https://churchtechtoday.com/the-2026-state-of-ai-in-the-church-93-of-pastors-are-in-now-what/

[3] Barna Group. "How Church Leaders Are Using AI (And What Concerns Them Most)." Barna Group, March 26, 2026. https://www.barna.com/research/church-leaders-ai-usage-concerns/

[4] Ed Stetzer. "Navigating the Ethics of AI in Ministry and Sermon Writing." The Good Book Blog, Biola University, October 22, 2025. https://www.biola.edu/blogs/good-book-blog/2025/navigating-the-ethics-of-ai-in-ministry-and-sermon-writing

[5] Barna Group. "Three Takeaways on How Pastors Can Use AI." Barna Group, February 22, 2024. https://www.barna.com/research/pastors-use-ai/

[6] Christopher Benek. "Should Pastors Use AI to Write Sermons?" christopherbenek.com, January 2025. https://www.christopherbenek.com/2025/01/should-pastors-use-ai-to-write-sermons/

[7] Christianity Today. "Young, Educated and Urban Pastors Are Most Likely to Use AI." Christianity Today, April 2026. https://www.christianitytoday.com/2026/04/ai-chatgpt-anthropic-pastors-church-research/