Yesterday, MinistryWatch, a non-profit that tasks itself with holding Christian ministries accountable for financial transparency and efficiency, published what can accurately be described as a hit piece against Ligonier Ministries.
The piece, written by freelance writer Kim Roberts, delves back into an issue Protestia has been reporting on since mid-March:
UPDATE: What’s Happening Around the Ligonier Network?
Stephen Nichols Statement Makes the SAC/Ligonier Conflict Both Clearer and Blurrier
Bodycam Footage, New Letter Deepen Dispute in Saint Andrew’s Chapel Conflict
Julie Roys Abuses YouTube Copyright System to Remove Nichols Bodycam Footage From Protestia Article
As I listened to Roberts’ article (driving in the car, having my phone read it to me), I couldn’t help but notice that the rhetorical techniques being used seemed very familiar. While the article (published under MinistryWatch’s #churchtoo category) comes across as dispassionate reporting at first glance, and reports responses from Ligonier, the same Julie Roys-esque implication techniques I had exposed at length in 2022 were being used here.
It’s no secret that we’re no fans of MinistryWatch, particularly after its president, Warren Cole Smith, took a detour from his purportedly neutral examination of Christian ministry transparency to attack Megan Basham’s book Shepherds for Sale (and our assistive research) on behalf of his friends back in 2024. Yet this latest article targeting Ligonier read like a Julie Roys piece against John MacArthur, complete with denials of causation while implying it, a lack of evaluation regarding truth claims, a critique of Ligonier’s lack of transparency while itself using selective sourcing, conflation of categories (legal, theological, institutional), and (most Roys-like), building narrative weight through proximity rather than proof.
Then it occurred to me – large language model AI might be able to quickly shed some light on these kinds of techniques, and (unironically) bring some transparency to the foggy journalistic lens these issues seem to always be viewed through.
Setting aside for a moment the relative value or ethical concerns of AI (I did take issue with it being used as a replacement for substantive argumentation a few months ago), it can be quite useful as a tool to quickly parse a given piece for summarization and analysis. Back in the dark ages of 2022, of course, these tools were unavailable, and we simply did the analysis ourselves and tried our level best to get those who should have known better about John MacArthur to understand that they were being hoodwinked.
Simply put, even talented and brilliant folks in conservative evangelicalism were being dazed and confused by the rhetorical emotionalism being used. Roys’ claimed conclusion was stated. But the mechanics—the how behind that conclusion—were harder to pin down in real time.
I argued then that the narrative was constructed, timelines were twisted, and conclusions were implied rather than demonstrated. To this day, neither Roys nor any of her supporters has been willing or able to lay a glove on our takedown of her attempted takedown of MacArthur. Yet today we have access to tools that can confirm (or object to) our analysis and reasoning, and do it much quicker. Of course, AI makes plenty of mistakes, but for those who are truly interested in seeing behind the curtain of how modern internet hit pieces work, the exercise is worthwhile.
With modern language models, readers can now take these same articles, run a simple prompt—
“Where are the logical gaps? Where is rhetoric doing the work of proof?”
—and see patterns surface almost instantly.
What took pages to explain in 2022 can now be identified in seconds. So let’s do exactly that.
Below are three case studies:
- Roys Report (2022) — MacArthur / Grace Community Church
- Roys Report (2025) — Michael Tait / Newsboys
- MinistryWatch (2026) — Ligonier Ministries
Different targets. Different facts. Same techniques.
Note: In all three cases, the analysis evaluates the rhetorical and logical structure of the articles, not the truth of the underlying events.
AI Analysis #1 — Roys Report, (MacArthur Case, 2022)
Article: EXCLUSIVE: John MacArthur Shamed, Excommunicated Mother for Refusing to Take Back Child Abuser
Prompt: Identify the logical and rhetorical inconsistencies with this piece.
Summary of the analysis: The article builds a powerful narrative not by clearly proving wrongdoing, but by arranging facts, timelines, and emotions in a way that makes wrongdoing feel obvious. Basically, exactly what we identified back in 2022.
Summary of issues:
- Headline Certainty vs. Evidentiary Complexity: The headline presents a complete moral narrative, while the article itself describes a developing situation with partial knowledge.
- Timeline Compression: Events separated by time are presented as if they occurred with full knowledge at once, implying intent that is not demonstrated.
- Hindsight Bias: Later criminal convictions are used to judge earlier pastoral decisions, without clearly establishing what was known at the time.
- Category Conflation: Legal obligations, moral outrage, and church discipline are blended into a single implied failure, despite being distinct domains.
- Emotional Anchoring: Graphic abuse details are placed early, shaping reader interpretation before analytical evaluation occurs.
- Insinuation Without Assertion: The article arranges facts to imply wrongdoing while often avoiding explicit claims that would require direct proof.
- Guilt by Proximity: Multiple issues are clustered together to create a sense of systemic failure without independently establishing each claim.
Summary Insight: The narrative persuades not by proving wrongdoing, but by structuring information so wrongdoing feels obvious.
AI Analysis #2 — Roys Report (Michael Tait / Newsboys Case)
Prompt: Identify the logical and rhetorical inconsistencies with this piece.
This one is interesting because not only is the article being analyzed a direct exhibit in the Newsboys/Wes Campbell lawsuit against multiple parties, including Julie Roys herself, but Morris and her reporting on the case recently received an award from the Religion Journalism Association for story of the year.
Summary of issues:
- Headline Certainty vs. Allegation: The headline presents specific claims as definitive, while the article relies on an accuser’s account and belief-based language.
- Character Reinforcement as Evidence: Prior admitted misconduct is used to strengthen new, unproven allegations, despite being different categories of behavior.
- Emotional Priming: Severe allegations are foregrounded, anchoring reader interpretation before evidence is fully examined.
- Elastic Language (“Cover-Up”): Loaded terms are used without clear definition, allowing readers to assume the strongest possible meaning.
- Blurring Belief and Fact: Statements of belief are embedded within factual narrative flow, reducing clarity about what is proven vs. alleged.
- Pattern Construction: Multiple allegations are presented together to reinforce the perceived credibility of each individual claim.
- Timeline Ambiguity: Past events are interpreted through present knowledge without clearly distinguishing what was known at the time.
- Institutional Expansion: An individual allegation is used to imply broader organizational failure without clearly establishing institutional knowledge.
Summary Insight: Even where real wrongdoing exists, rhetorical structure can extend conclusions beyond what is directly demonstrated.
AI Analysis #3 — MinistryWatch (Ligonier Case, 2026)
Article: Fallout at Ligonier Ministries and Reformation Bible College
Prompt: Identify the logical and rhetorical inconsistencies with this piece.
- Denied Causation, Implied Connection: The article claims events are unrelated while structuring the narrative to suggest they are connected.
- Selective Skepticism: Conflicting claims are presented, but not equally evaluated, allowing tone to guide reader conclusions.
- Transparency Critique Built on Partial Transparency: The article criticizes lack of transparency while relying on selectively presented donor reactions and incomplete context.
- Category Confusion: IRS classification, theology, and organizational identity are blended into a single implied problem.
- Timeline Structuring: Events are arranged to suggest causation without clearly establishing it.
- Guilt by Association: Multiple unrelated concerns are presented together to create a cumulative sense of institutional failure.
- Donor Reaction as Evidence: Individual dissatisfaction is treated as indicative of systemic issues.
Summary Insight: The article constructs a narrative of institutional concern through implication rather than direct demonstration.
The Pattern Across All Three
Step back and look at what’s happening. Both Roys and Smith fancy themselves as watchdogs over Christian ministries. Yet both employ deceptive techniques that are designed to confuse average Christian readers and lead them to distrustful conclusions about (overwhelmingly conservative) institutions and people in evangelicalism. And the question must be asked: If their targets are indeed chosen based on behavior and not theology/beliefs, why are these manipulative techniques necessary?
The playbook is consistent:
- Lead with a morally compelling narrative
- Compress or rearrange timeline
- Blend categories (legal, moral, institutional)
- Anchor the reader emotionally
- Use elastic, undefined language
- Cluster concerns to amplify impact
- Allow implication to replace proof
These are not random errors.
They are repeatable rhetorical techniques.
“But Protestia Is Biased Too”
Of course we are. And we’re clear about it.
Protestia operates from a stated theological framework, explicit doctrinal commitments, and openly argued positions. You know, like Christian ought to do. We are not feigning neutrality or claiming to be “above it all” while we rhetorically advance an unstated agenda.
In this analysis, the issue lies in the method these outlets employ, not their perspective per se. There is a difference between an outlet that declares its framework, makes arguments explicitly, and invites evaluation of those arguments, and one that uses narrative-driven pieces to imply conclusions, arrange facts to support them, and use rhetoric to bridge gaps.
Even though both Roys and Warren have objectionable theology from a biblical perspective, for the purposes of analyzing their journalistic methodology, AI doesn’t care about theology.
Instead, it surfaces:
- where claims exceed evidence
- where implication replaces demonstration
- where narrative does the work of proof
Why This Matters Now
In 2022, unpacking these deceptive techniques required lengthy rebuttals, detailed timeline reconstruction, and extensive argumentation. No matter how carefully the issue was discussed, readers and social media commenters routinely misstated basic facts or asserted claims clearly disproven by the analysis. Information-deluged readers lacked the endurance to understand the core content of the issues being written about, much less understand the rhetorical trickery being used.
Today, readers may be able to get a head start by running an article through an AI, asking basic analytical questions, and seeing these patterns surface immediately. This won’t replace theological discernment nor journalistic fact-finding, but it’s perfectly useful in quickly uncovering the emotionalized fog too often weaponized by “watchdogs” who are secretly advocates.
The Bottom Line
While I didn’t dive as deeply into Jessica Morris’s Tait article or yesterday’s Ligonier piece, this analysis lines up nearly perfectly with the 2022 pre-AI analyses we published on Roys’ MacArthur crusading.
Outlets like The Roys Report, MinistryWatch, and others that feign impartial watchdog status too often do not persuade primarily because they prove wrongdoing. Instead, they persuade by constructing compelling narratives, leveraging emotion, compressing timelines, and implying conclusions. And the ends of their unstated (or at least deemphasized) theological commitments justify these means.
One final note, in the interest of methodological transparency:
If this analysis is valid, it should hold up when applied to our own writing as well. So we ran it. What follows is a condensed, AI-assisted analysis of the above article’s rhetorical structure.
The goal is not to claim neutrality—we don’t—but to show that the difference is not the absence of rhetoric, but the willingness to expose and examine it.
7 responses to “Roys, Smith, and What AI Can Now Reveal About “Watchdog” Journalism”
[…] Roys, Smith, and What AI Can Now Reveal About “Watchdog” JournalismInside the $50 Billion Lawsuit Shaking Christian MusicApologist William Lane Craig Says Belief In A ‘Young Earth’ Is Akin to Belief In A ‘Flat Earth’Worship Artist Cory Asbury Says ‘Sin is Rampant’ In Christian Music Industry, But Again Won’t Name Names […]
This is a helpful summary. Institutional Expansion is what stands out to me the most from these “watchdog exposes.” When GRACE did their report on BJU, for instance, it was the same sort of thing. Rebecca Davis of “Here’s the Joy” website uses similar tactics in her writing about John Street, the SBC, Biblical Counseling, Jay Adams, etc. Emotional Priming and Timeline Ambiguity are pretty apparent, too.
They themselves are deceived by the same deception from the sin of Adam, and are only appeasing and appealing to those who, like them, are already deceived, and in the wickedness of their heart are looking for excuses and want to be deceived. Those who are truly born again, indwelled by the Holy Spirit, with the implanted word, and grounded in the word, are not so easily deceived, see through the deceit, and are fairly immune to it. We can certainly be deceived, but not quite as easily, and only when the Lord allows it for His purposes.
What you’ve done here, David, is to further show those who are deceived that they are indeed deceived and how they are deceived. It is brilliant, and clearly done by leading of the Holy Spirit, there is no doubt in my mind about that. May many read and see that they are deceived by the same deception since the sin of Adam, and be drawn to the true Gospel of Jesus Christ. For the Lord is not willing that any should perish, but that all should come to repentance. (2 Pet. 3:9)
I’m not commenting upon whether any of the reports were right or wrong. But, as a matter of methodology, note that any AIs you’re feeding them into are likely to have already ingested not only the original report, but reams of Internet commentary upon them. As such, the AI analysis of the report is likely to be not simply based upon analysing the report itself, but also affected by the existing commentary.
So, for example, there’s been recent discussion over research into AIs that evaluate CVs. The research indicated that any particular AI has a bias not simply towards AI-generated CVs, but specifically towards CVs generated by *the same* AI – https://arxiv.org/pdf/2509.00462 .
i.e. If an AI repeats criticisms that you’ve earlier made of a report, when fed that report, it has to be borne in mind that the AI has likely ingested your criticism already as part of its training data, and already associates those criticisms with the report. So, the methodology of “ask an AI what it thinks about this report”, when it turns out to agree with previously-published criticisms – it can’t be assumed that this is a coincidence.
In other words, it’s “reasoning” is baseless. Very true. Still a useful tool for many things I believe, but to be taken with a grain of salt, no doubt about that.
Reading this:
“Setting aside for a moment the relative value or ethical concerns of AI (I did take issue with it being used as a replacement for substantive argumentation a few months ago), it can be quite useful as a tool to quickly parse a given piece for summarization and analysis.”
Reminded me of reading this:
“The 2019 Southern Baptist Convention (SBC) approved the use of Critical Race Theory (CRT) and intersectionality as “analytical tools” to help understand and address social ills like racism, provided they remain subordinate to Scripture. This decision, known as Resolution 9″
Just sayin
I don’t see the equivalence. Critical Theory (not just CRT) is severely at odds with scripture, and is pretty much an entire false religion. They claimed it was just a tool, but it’s not just a tool. But the AI tool that David used here, is, in fact, just a tool. Useful for many things, but it’s output should always be taken with a grain of salt, because the old GIGO standard still very much applies.