The Ghost(writer) in the Machine

A follow-up to Dr MacMillan’s piece on Magnifica Humanitas, in which we descend from the encyclical to the app store.

A few days ago my colleague on this platform walked us through Magnifica Humanitas, Pope Leo XIV’s first encyclical, and left us with the question the Pope believes will define the century: why stay human? It is a magnificent question, posed from a magnificent height. In this post I want to bring it down to sea level, or more precisely to Boulder, Colorado, elevation 1,624 meters, home of a company called Gloo.

Talking to God, terms and conditions apply

Gloo describes itself as the technology platform connecting the faith ecosystem, which is the kind of sentence that makes you reach instinctively for your wallet to check it is still there. It serves tens of thousands of churches, has raised over a hundred million dollars in venture capital, and offers, among other things, Gloo AI Chat: a chatbot for life’s biggest questions, delivering answers that are, in the company’s words, biblically sound and grounded in faith-based principles. It runs on something Gloo actually named the Kingdom-Aligned Large Language Model. The acronym is KALLM. Keep KALLM and carry on, presumably.

I want to be fair. Gloo’s stated philosophy is that AI should enable relationships rather than replace them, and its leadership says sensible things about discernment. Some of what it builds, church administration tools, content search, analytics, is unremarkable and probably useful. If a parish secretary uses AI to schedule the potluck, the gates of hell shall not prevail against it.

But then there is the feature that should make everyone’s theological antennae vibrate. When a church signs up, Gloo’s system can pull the statement of faith off the church website so that the chatbot answers questions in alignment with that statement. And for individual users, the platform offers personalization, so that the answers reflect what the user already believes.

Read that again. A machine for life’s biggest questions, calibrated to tell you what you already think. I wrote a piece on this platform some months ago about confirmation bias in religious data, and I confess I did not expect the industry to productize the fallacy quite so quickly. The old idols were at least inconvenient. You had to carve them, carry them, feed them. The new one ships as a subscription and agrees with you out of the box. Aaron needed the Israelites’ gold earrings to build the calf; Gloo just needs your church’s URL.

Meanwhile, in the pulpit

Gloo is only the most visible node in a much broader trend. The survey data is genuinely startling. According to the 2025 State of AI in the Church report, 61 percent of pastors now use AI weekly or daily, up from 43 percent a year earlier, and 64 percent of preaching pastors use it somewhere in their sermon preparation. Barna, more conservatively, finds that about a quarter of US pastors are now actually writing or editing sermons with AI, roughly double the share who were merely comfortable with the idea in early 2024. NPR interviewed a Lutheran pastor who fed the week’s lectionary reading into ChatGPT and received a 900-word sermon in under thirty seconds. Her reaction, to her credit, was to be impressed and then not to preach it.

Eulogies are on the same trajectory, and here the stakes are more intimate. One pastor put the objection perfectly when he asked whether the AI knows the stories of your people, whether it knows about the miscarriage. It does not. It cannot. As one commentator quoted in Dr MacMillan’s piece observed, AI cannot have skin in the game because it does not have skin.

There are thoughtful dissenters. The theologian Brad East has argued that AI has no place in the pulpit at all, on the grounds that the drudgery of sermon writing is part of the vocation: he does not want pastors preaching from Scripture who do not themselves wrestle with Scripture. I do not go quite as far as East, for reasons I will explain. But he has put his finger on the thing that most defenses of AI-assisted ministry miss.

What I tell my researchers

Here I must declare an interest, and an obsession. In my day job I direct research at a university, and a good part of the last years has gone into writing institutional policy on the use of AI in academic work. The rules I ended up with are, at bottom, two.

First, declare it. Every paper that leaves our institution carries an AI use statement: which tool, for which task. Not because AI use is shameful, but because authorship is a claim, and claims need to be honest. The undeclared use of AI in a journal article is a form of misrepresentation, and I have spent more hours than I care to count investigating cases where it shaded into outright fraud.

Second, and more fundamentally: use AI the way you would use your older sister. This is the test I give junior researchers. You can ask your older sister to check your grammar. You can ask her whether your argument makes sense, whether you have missed an obvious objection, where to find that article you half remember. She is smart, she is available, and she genuinely improves your work. But if your older sister writes your essay, something has gone wrong that no citation format can fix. The essay may be better. You are worse. The point of writing the essay was never the essay; it was the person the writing was supposed to produce.

Writing, in academia, is not the packaging of thought. It is the thinking. Outsource the writing and you have not saved time on thinking, you have skipped it. The paper exists and the scholar does not.

Now transpose this to the pulpit and turn the volume up. A sermon is not a weekly content deliverable; it is, on the church’s own account of itself, the fruit of a human being’s prayerful wrestling with a text on behalf of particular people whose miscarriages and bankruptcies and quiet despair the preacher actually knows. By my older-sister test, plenty of pastoral AI use is fine, and I part ways with the total abolitionists here. Checking the Greek, gathering the cross-references, tightening a conclusion, drafting the announcements: delegate away, and say so. The Barna data suggests most pastors intuit exactly this hierarchy, embracing AI for administration and design while hesitating at sermon writing and recoiling at counseling. Their instincts are better than their incentives.

But the seduction is real, because the tool that checks your Greek on Monday will happily write your eulogy on Friday, and it will never once tell you that you are asking too much of it.

The god who always agrees

And this is where the ministerial case becomes categorically more dangerous than the academic one. When my doctoral student outsources her thinking to a chatbot, she deceives her committee and impoverishes herself. Regrettable, sanctionable, survivable. Nobody in the room believes the machine is inspired.

In the religious case, that last safeguard quietly disappears. The encyclical names AI’s two great illusions with real precision: it appears objective when it is not, and it simulates human communication when it is not human. Now add a third layer that only religion can supply: the user who suspects, or is gently encouraged to feel, that the voice on the other side of the screen is not merely helpful but Providential. A statistical pattern-matcher, trained partly on your own congregation’s sermons, tuned to your own statement of faith, personalized to your own priors, whispering back to you in the second person. If you set out to design a machine for mistaking your own reflection for revelation, you could not do much better. Babel, the Pope reminds us, was also a technically impressive project with excellent user engagement.

The biblical tradition has a word for a god who reliably says what its makers want to hear, and the word is not God.

Magnifica Humanitas insists that weakness, struggle, and suffering are not bugs in the human program to be optimized away but the very terrain on which love becomes possible. The struggle of the pastor over a blank page on Thursday night is a miniature of exactly this. So is the mourner who writes a clumsy, true eulogy instead of commissioning a fluent, false one. So, for that matter, is the doctoral student and her chapter. The temptation in every case is identical: to exchange the slow formation of a soul for the instant production of text.

Why stay human? Dr MacMillan’s piece answered: because we want to be happy, and no machine can be. Let me add the ecclesial corollary. Because when the congregation asks who wrote the sermon, and the honest answer is nobody, they will eventually stop coming to hear it. You cannot shepherd sheep by proxy, and you cannot pray by proxy either, no matter how kingdom-aligned the proxy claims to be.

Ask your older sister for help. Just do not ask her to believe for you.