Billy Graham Had a Ghostwriter but TGC Won't Let You Have a Proofreader.
TGC mistakes technological purity for human faithfulness. Its AI ban will hit developing writers hardest—and it still won’t solve the real problem
I’m proud that The Gospel Coalition has drawn a line in the sand on AI usage in the writing we publish. In short: We don’t want AI’s tentacles on any of the written works we publish. We want our audience to trust that what they get at TGC is human-made, human-edited, human-assessed, made in human community for other humans—as an act of love, fellowship, and worship.
Our Line in the Sand on AI: TGC’s Missional Opportunity, August 5, 2026 | Brett McCracken
Before I read the whole article, it’s title ‘Our Line in the Sand on AI’ got my attention, as intended. I assumed it was a knee-jerk reaction to AI and then having read it I still think it is, although not quite for the same reasons.
The Gospel Coalition (TGC) has spotted a real danger with AI: it can churn out language that seems thoughtful when there's absolutely no thought behind it. You end up with a finished article dropping out of nowhere, missing all the usual background work.
But here is where TGC gets it wrong.
An AI system cannot believe in what it writes. It has no congregation to answer to, and if it invents a citation or misrepresents somebody’s theology, it feels no conviction.
An AI model can re-produce the words “Christ is risen,” but it cannot mean them. Which is why I do not want AI-generated Christian testimonies, invented pastoral encounters, and generic machine-written articles lightly dusted by the person as the author.
If someone asks an AI to 'Write a Christian article about hope,' tweaks a few words, and fires it off to a publisher, they're implying they did the hard work. They’re claiming they reached those conclusions, understand the arguments, and can stand behind them. When in reality, all they've really done is supervise a large language model.
TGC is completely right to reject that, but their mistake is lumping almost every other use of AI into the exact same category.
Under their policy, TGC writers aren't allowed to use AI to clarify a thesis, organise their thoughts, stress-test an argument, do basic research, or edit anything beyond spelling. They then try to give this ban theological heft by arguing that because AI can't worship, writing shaped by AI can never truly be an act of worship.
Really?
That’s such a glaring category error that even the limited AI use TGC bans—like asking a tool to stress-test an argument—might have caught it before publication. After all, a pen cannot worship either. Neither can a keyboard, a dictionary, or a guitar. We don’t ask whether the instrument loves God, because the person using it is the worshipper.
TGC has conflated two distinct things: they’ve taken the sound premise that AI cannot be an author and stretched it into an unwarranted conclusion—that a human forfeits their own authorship the moment they use it.
Then they ban almost everything
Suppose I’ve read the sources, formed my thesis/argument, and written a draft. Something in the middle isn’t working, but I can’t spot the problem, so I turn to an AI:
Tell me where I’m making a logical leap. Don’t rewrite my text or do the thinking for me—just ask questions to help me improve.'
The AI offers three observations. I discard two as unhelpful, but the third exposes a genuine blind spot. I return to the primary sources, realise the AI critique is fair, and rewrite the section myself. Afterwards, I ask the AI to assess my revisions, and finally, to generate a brief note on what I can learn from the process for next time.
Have I stopped writing?
I don’t think so. The AI didn’t give me the conviction, the conclusion, or the final wording. It simply asked a question—and I still had to judge whether that question was any good and decide what to do about it. That is worlds apart from asking a machine to write the entire piece, even if both happen inside the exact same chat box.
'Using AI' tells us almost nothing until we ask what the writer actually handed over. Transcription is not testimony, search terms are not research, finding a recurring argument is not the same as inventing one, and translation is not theological judgment.
Sometimes an AI model assists a person who remains fully involved, while at other times it performs the intellectual work and leaves the person watching from the side. Yet TGC refuses to separate those cases because it has chosen a purity rule that is easy to state over an authorship test that requires - ironically - actual human judgment.
To assess a written article, you have to ask whether the writer reached the central claim, whether they read the sources they cite, and whether they understand what their own sentences mean. You have to ask whether they could defend the argument without running back to the AI model for a ready-made defence, whether they would change it if someone exposed an error, and whether they are prepared to answer for the whole thing.
The supreme irony of TGC’s ban is that it exposes a profound superficiality in their understanding of both technology and human formation. The real intellectual frontier lies in how a tool is engaged—the questions it prompts, the rigour it tests, and the cognitive friction it provides—not its mere presence on a workspace. By drawing their blanket line in the sand, TGC reveals a double failure: a poverty of imagination regarding what AI actually is, and a fundamental blindness to how human beings learn, grow, and are deeply formed by the tools they use.
And the absurdity is that Christian publishing has outsourced writing for many years, long before the advent of AI.
Christian publishing
There is an irony that the article never addresses: Christian publishing has lived with ghostwriters, co-writers and heavily interventionist editors for decades, including writers hired to turn a pastor’s sermons, interviews and rough ideas into a polished book carrying the celebrity pastor’s name.
This is not a fringe practice, it is extant.1 Sometimes the collaborator receives an acknowledgement or a “with” credit. Most times the work remains invisible. Either way, Christian publishing has already spent years dealing with books in which the name on the cover did not write every sentence and may not have built the structure or produced the first or any draft.
That does not automatically make ghostwriting wrong. A pastor may supply years of sermons, developed convictions and extensive interviews, while a skilled writer turns that material into a readable book. The named author may examine every page and remain responsible for the overall message.
Fine. But once we accept that possibility, TGC’s blanket argument against AI starts to unravel. A human ghostwriter can love, believe, and worship; AI cannot. Yet if the ghostwriter produces the prose and the celebrity author takes the byline, whose act of worship is the book supposed to be? The person who formed the sentences isn’t named, while the person on the cover didn’t write them.
Meanwhile, a writer who develops their own thesis, reads the sources, writes the draft, and merely asks an AI to spot a flaw in their reasoning falls foul of TGC’s policy—simply because the critic they consulted cannot worship.
That is indeed, a very strange way to measure authorship.
Which person has the stronger claim to being the writer: the Christian celebrity whose manuscript was largely produced by another human being, or the unknown writer who did the research and composition, but used a machine to stress-test their argument?
The Christian publishing industry has long found room for the first arrangement. TGC’s policy outright rejects the second.
Perhaps the answer is that a ghostwriter works from the named author’s ideas and remains under their direction. If so, TGC should concede an important point:
Authorship can survive assistance when the named person retains intellectual control, understands the work and accepts responsibility for it.
Once that’s admitted, the real issue isn't whether the helper was human or artificial—it’s what work was actually outsourced. The uncomfortable questions are already familiar: Who shaped the argument? Who wrote the prose? What has the reader been led to believe? Is someone's labour being hidden? Can the author on the cover stand behind the work? Who takes responsibility when it turns out to be wrong?
Christian publishing needed honest answers to those questions long before ChatGPT arrived. AI didn't create the authorship crisis. It just made the industry’s existing compromises much harder to ignore.
TGC’s ban favours writers who need AI the least
The TGC AI policy flatters accomplished writers because people who already know how to build an argument will cope perfectly well; they have spent years learning how to hear a weak sentence and spot the place where a paragraph slips away from its subject, and many of them already have editors, teachers and other writers in their circle.
They will be fine.
TGC tells writers who need assistance to find a well-read friend, another writer or an English teacher at church, but where, exactly, are they supposed to get these people?
A writer in a well-connected church may hear that advice as an invitation into community, while somebody else hears an instruction to find a resource they simply do not have.
Think about a writer working in a second language who knows what they want to say but cannot yet make it sound like publishable American English, or a dyslexic writer who struggles to catch structural repetition across a long draft, or someone with ADHD who understands the subject but finds it difficult to hold the sequence of the article together.
Then there are people raising children, working long hours or writing without a professional network around them, who may have something worth saying but do not have an English teacher waiting to edit it for free.
Some will submit rougher work, some will use AI quietly, and others will decide that writing for TGC is not for people like them, while TGC will never see what it has lost because you cannot count the articles that were never submitted.
If human editing forms part of TGC’s ministry, it could provide human editing by paying people, building apprenticeships, reading work that is not ready yet and giving developing and second-language writers actual access to the community the policy assumes.
Telling writers to find a human editor does not preserve community; it assumes they already have one.
Research suggests this is more than a theoretical concern.
An experiment involving 453 professionals found that access to ChatGPT cut the time spent on writing tasks by 40 percent and raised independently judged quality by 18 percent, while also narrowing the gap between stronger and weaker participants.
This was not a study of theological writing and it did not measure wisdom, prayer or moral formation, but it still knocks down the idea that AI involvement can only make work worse.
A study involving more than 5,000 workers found that novice and lower-skilled staff gained much more than experienced staff, which is precisely why a total ban worries me:
It removes a tool from the people who may gain most from careful use while barely inconveniencing those who have already mastered the work.
There is another side to all this, however, and it is not minor.
Researchers studying consultants found that AI improved performance when the task suited the model, but once the task moved beyond what the model could do reliably, AI users became more likely to give a wrong answer; the researchers called this the “jagged technological frontier”.
In other words, AI can make weak thinking look settled and give a beginner work that sounds far more competent than they actually are, while the beginner may lack the judgment needed to recognise the gaps in their work.
A large study involving students using AI for mathematics found that unrestricted access improved their performance during practice, but when the AI was removed those students performed worse than students who had never used it, meaning they had completed the work without building the skill.
The researchers also tested a guarded version that gave hints rather than full answers, and most of the later damage disappeared because the system stopped doing the work for the learner and pushed the learner back into it.
That distinction should sit at the heart of an intelligent writing policy: did AI replace the thinking, or did it provoke more of it?
TGC does not ask. It simply bans both.
Not all struggle is holy
Some of the struggle involved in writing is the act of writing. A paragraph may resist understanding because the idea underneath it is confused. If AI smooths out too quickly for the writer, they can end up with a nice clean paragraph and no clearer understanding themselves.
I remember my PhD supervisor taking apart my writing, section by section, during three-hour meetings and repeatedly asking, “What are you trying to say here?”
Often I knew that I meant something before I knew exactly what it was. The thought was there as a pressure, a shape, an irritation—felt, but not yet well formed. We would stay with it while he kept asking questions. What did I mean by this word? Where had that claim come from? Had I actually shown the connection between these two ideas , or was I hoping the reader would do the missing work for me?
Eventually, I would wrestle my thought into something I actually understood and another person could follow.
Then my supervisor would say, “Good. Now go away and say that.”
I often left those meetings with a splitting headache because I was trying to hold the whole act of writing in my conscious attention: claims, evidence, the order of the arguments, my imagined reader and the sentences currently refusing to behave that I needed to edit and rewrite. Nothing had become instinctive yet. Every movement had to be chosen and worked through and practiced. And it was exhausting.
My supervisor had rewritten nothing for me. He had made me return to my thoughts until I knew what I meant, then return to my sentences until they said what I meant.
That was moving from conscious incompetence to conscious competence in practice. I could do the work, but only slowly, deliberately and at considerable mental cost. Over time, I began hearing my supervisor’s questions before he asked them, and decisions that once consumed a three-hour meeting started happening while I wrote, as I wrote.
The headache was what learning felt like before the skill became learned instinct.
I still hear his voice when I write, laid down in my neural pathways, permanently etched into my brain. “Jason what are you trying to say? Well say that then”.
But most people will never have a PhD supervisor sitting with them for three hours, taking their writing apart and refusing to repair it for them. But what if AI could give them some version of that experience?
Not an AI that rewrites the paragraph and hands it back polished, but one that stops at the weak point and asks, “What are you trying to say here?” One that follows the answer with another question, notices when the writer has jumped over part of the reasoning and keeps pressing until the writer can explain the thought in their own words.
Used in that way, AI would not remove the struggle involved in learning to write. It would create the conditions for it. It could teach people to question their own reasoning, review their own work and recognise the difference between a sentence that sounds finished and a thought they genuinely understand.
What if the people who cannot afford an editor, do not have a writing mentor and will never enter a doctoral programme could still experience that kind of sustained attention to their thinking?
AI does not have to write for them. It could teach them how to do the writing themselves.
That kind of friction is worth keeping, but some friction is simply a lack of time, money, access or appropriate tools.
Reformatting foot notes is not spiritual formation, a second-language writer does not become more faithful by spending another three hours correcting syntax, and a dyslexic person does not need to prove that a human wrote the article by finding the process harder than everybody else does.
The TGC article says writers should care enough about the work for its production to cost them and other people something. That line made me pause as I read because cost can come from love, but it can also be generated by an organisation shifting its labour onto writers, their friends and whoever may or may not be available to help.
If AI saves somebody an hour and they spend it checking a source, talking to a real person, looking after their children or getting enough sleep, what has the article lost?
Maybe nothing.
Christianity knows about discipline, but it also knows about grace and Sabbath, and we should be very careful before treating the more burdensome route as the more noble and faithful one because churches have made that mistake before.
The Incarnation is not an AI policy
TGC appeals to Christ’s Incarnation, arguing that God did not reveal himself as abstract information because the Word became flesh, which means embodiment and human presence matter.
Yes, they do, and an AI companion is not a friend; a model cannot become a pastor, share the life of a church, sit beside someone in hospital or carry another person’s grief.
But asking software to point out a contradiction in a draft is not an attempt to replace the body of Christ.
In an irony of ironies, TGC could have benefited from the very practice they ban: using an AI sparring partner to run a theological critique of their own draft and catch their category errors made over their incorrect use of the incarnation.
Christian communication has always passed through different forms—speech, memory, letters, scribes, translation, print, recordings and screens—and the article defending embodied communication reached most of us through software running on distant servers.
Paul used a scribe, Tertius identifies himself in Romans as the person writing the letter, Luke worked with earlier accounts, and Jeremiah dictated through Baruch.
And no - human scribes were not ancient large language models, and that is not what I am claiming; the point is that Christian authorship has never required one isolated person to originate and physically produce every word without assistance.
Our ideas are not entirely “our own” either, because Christians receive almost everything through Scripture, doctrine, language, teaching and correction, then borrow, forget where something came from, rediscover it and slowly decide what they believe to be true.
The Christian responsibility is to be honest about what we have received and take responsibility for what we pass on, which is harder than avoiding a tool.
Where I would draw the line
I would set firmer boundaries than many AI advocates want. Fabricated testimony should be out, along with invented pastoral encounters, quotations and citations; writers should not ask a model for theological conclusions and then present the result as the fruit of their own reading, prayer and reflection. And it should go without saying but still needs to be, that confidential pastoral information should never be entered into an external AI system.
An article generated mainly by AI should not carry the byline of someone who merely supervised its production, just as a book written by a ghostwriter should not carry the sole name of the celebrity, pastor or ministry leader who commissioned it IMHO. If you did not write the book, do not present yourself as its writer. Credit the person who did.
Writers should read the sources they cite, understand their argument and be able to defend every serious claim that is made.
I could accept a rule that says AI must not write the article, but it should be allowed to copy-edit a human draft, provided the author reviews every change and the AI does not alter the argument or erase the writer’s voice. But transcription, accessibility tools, translation, research leads and questions directed at an existing draft are different activities that transfer different kinds of work and should be judged accordingly.
Furthermore, TGC’s reliance on ‘suspected or confirmed’ AI use is unworkable. A policy that treats clean prose, clear structure, and a lack of typos as a smoking gun of machine authorship collapses under basic scrutiny.
Author Leonard Sweet exposed just how broken these detection metrics are. Before using the detector Pangram on student papers, Sweet ran excerpts from his own books—published decades before AI existed—and watched Pangram rate them as up to 80% AI-generated. Even worse, when he asked Gemini to rearrange his own genuine prose into marketing copy, the detector branded it ‘100% AI
The detector could not tell who had originated the language, when it had been written or how much intellectual work the machine had performed; it could only identify a style that resembled AI, even though AI had learnt that style from writers such as Sweet in the first place!
AI detectors do not produce certainty, and some have wrongly flagged human writing, particularly work produced by people writing outside their first language, so TGC should talk to the writer, examine drafts if the stakes warrant it, allow them to answer and give them a route of appeal.
A policy that claims to honour human beings should not begin by treating them as guilty until proven innocent because their prose does not look human enough.
Stop Policing the Tool. Start Forming Writers.
If Christian publishers care about the formation of writers, they should do more than prohibit the technology and hope the old routes into writing remain available.
Commercial AI companies largely sell speed: produce the answer, finish the task, generate more material, and move on. Christian publishers do not have to accept this as the only way these systems can be used. Because they already employ editors, theologians, teachers, and experienced writers who know the difference between polishing a sentence and teaching someone to think, they are uniquely positioned to build models of AI-assisted apprenticeship that commercial platforms have no incentive to create.
TGC could bring those people together and test writing processes in which the writer first develops the idea and produces a rough draft, uses AI to question the reasoning rather than replace it, and then works with a human editor who can address the theological, personal and relational dimensions a machine cannot touch.
The point would be to discover whether writers become more capable over time, not merely whether they produce a better article while the tool is present. Can they identify gaps in their own reasoning? Do they become better at reviewing a draft without assistance? Does working with AI prepare them for a more serious conversation with a human editor, or does it make them dependent upon the machine?
Christian publishers could publish what they learn and set standards that the wider industry currently lacks, showing where AI use becomes deceptive, what authors must never hand over, when disclosure is necessary and how a publisher can protect human responsibility without reserving publication for people who arrived with every skill already in place.
They could also open these programmes to writers outside the usual networks, actively looking for people with something to say who have never had access to the training, time or relationships through which Christian writers have traditionally been formed.
At present, TGC’s policy effectively says that the safest writer is the one who arrives fully formed. A better response would be: come and learn how to do this well.
The internet does not need another flood of machine-produced Christian content. Never mind AI slop—the commercial Christian publishing machine is already drowning in an ocean of ghostwritten, brand-driven material. What we need are people who can think clearly, write honestly, and take responsibility for their words. Christian publishers could use AI to widen access to that kind of formation, showing the wider church what responsible use actually.
TGC can preserve the system that produced its existing writers, or it could use its experience to help form writers who would never otherwise enter it.
I know which of those I think would serve the church better.
Mel White openly describes himself as a former ghostwriter of books, speeches and autobiographical material for Billy Graham, Jerry Falwell, Pat Robertson and Jim Bakker. Lisa Stilwell, who worked as a senior editor for major Christian publishers, has described publishers commissioning her to ghostwrite devotionals and memoirs for Christian leaders and public figures.