Can AI analyze Christian theology? Not this one.

I was working on a simple introduction to the Sermon on the Mount. Until it turned out to not be so simple. I was trying to see if the crowds in the Sermon on the Mount chapters were largely the same people that were following Jesus before that – and also if they followed Him again afterwards. These days, the fastest way to do that research, from strictly Christian sources, is via the Faithlife/LOGOS AI Agent. Unless it isn’t. Below, I’ll go through my Q&A sequence and show you why it’s always essential to verify everything.

I often write/say, “Check everything – including me”. Well, I always check AI. And despite the attitudes some people have towards it, checking AI response is essential.

That’s what the adjacent image is about. The man’s telling the futuristic AI – well, what is he doing? A quick look shows “AI Response Correct”. But a close look, for which you might have to double-click to open a larger view, shows something else.

Immediately below the “AI Response Correct”, it says press any button to continue. Any button, that is, except the two red ones. The round one on the console is a big red X. The hologram also has red text that says “NO“. That’s typical AI. It always thinks it’s right. And if you continue, you affirm that belief. It’s only by doing something else – pushing back, as I call it – that we say “NO, your response isn’t correct“.

Maybe it’s outright wrong, maybe misleading, or somewhere in between. The thing is, we need to push back, regardless of which of those might be the case, otherwise AI continues down that path and will not only continue to follow that flawed line of answers, but it may very well affect all responses you get after that. If the AI you’re using has the ability and you’ve given it permission (explicit or implicit without knowing it) to “remember” past interactions, that line of thinking gets embedded in everything you do after that.

So – yes, AI is way faster than we are. For various reasons, they’re not always 100% accurate. And unlike us, they don’t even think about checking their work – until after being told they were wrong.

Well – maybe that’s also just like us.

But hey – when our souls depend on knowing the truth about God, it’s more than worth the time and effort to be sure we get it right. Check out You will know the truth and the truth will set you free for more on that.

To that end, this article isn’t a deep dive into theology – that comes later – but a real‑time example of how AI can take solid sources, merge viewpoints, and drift into conclusions the authors themselves never made.

The Crowds at the Sermon on the Mount

As I mentioned, it wasn’t simple research. Surprisingly, it may have been the most difficult one I’ve ever done, as far as accuracy is concerned.

I’ll take you through the entire sequence, questions and answers – and show what’s wrong and why the responses were formatted the way they were to try to reach a coherent conclusion from various sources.

One of the things that made this even more difficult is that the AI Agent drew from some authors that I frequently use when looking for things myself and/or to read their books. That gives a bit of inherent trust, just because of my knowledge of their works. The issue wasn’t what they wrote – it’s how it was interpreted and especially how it was integrated into the answer.

Because this is about challenges to the responses, and because the initial responses were modified multiple times, please do not come to any conclusions based on the initial responses below. At the very end, I asked it to consolidate all the corrected responses into a final conclusion. However – even there – an issue came up on a portion of the response that indirectly went back to an earlier response – so it also had to be corrected.

So, let’s get started.

---

1. Did pre-Sermon crowds follow Jesus onto the mount?

Here’s the passage in question – immediately before the Sermon on the Mount passages:

Based on that, my initial question was:

---

There are assumptions in what you’re about to read about “two-tiered” audience, and that Jesus apparently didn’t want to leave the previous crowd behind. This particular author also makes a number of references back to Moses and Exodus. And while he says, “scholars interpret it as intentional theological symbolism”, the use of scholars as opposed to some scholars – overstates the agreement others may have with some of the conclusions he reaches.

---

As mentioned, “Scholars recognize that Jesus did not leave the crowds behind—rather, Matthew presents a deliberately structured two-tiered audience” overstates the amount of support for this theory. The source author for this part of the response follows a method that can place more of an Old Testament comparison and relationship than context allows for. Since I’m more on the context side, I have a problem.

The two-tier audience, which AI will eventually abandon for a three-tier structure, comes from Mount Sinai, where Moses met with God while the people stayed below. And that comes from Mount Sinai and Jesus being on a Mount for these passages. It’s a relationship that I don’t believe is warranted by the contexts. Not every hill or mountain is Sinai. So, while there are parallels, extending them beyond the context of not only Exodus and Matthew – but of the Bible as a whole – can be problematic, as we’ll see with this two-tier assumption.

My original question, in my opinion, wasn’t adequately addressed. I wanted to know whether Jesus was able to escape the crowds, because sometimes He wanted to get away from them, for various reasons.

---

The AI response then continued:

My question still wasn’t answered. Instead, the AI launched into an implication and subsequent conclusion about “a deliberate staging ground where Jesus simultaneously addresses multiple audiences with different levels of understanding and commitment.”

It reaches a conclusion that doesn’t seem warranted based on the sparse information provided about the crowds from Matthew 4 versus the crowds in Matthew 5.

I didn’t go into this with any preformed assumptions – but I also wasn’t expecting anything quite so sweeping as a reply.

What follows is the beginnings of pushing back or one piece of the response at a time.

---

2. Is the first view as widely accepted as implied?

Looking into alternative viewpoints was my first reaction. The answers were surprising. But still – it’s good to double-check our own assumptions as well as the thing we’re questioning.

I’ll break this up into multiple sections. As to the question of whether AI was right or not:

Essentially, if you pay attention to the source authors. the AI says, “Yes, I was right”. Well, AI is generally very good at summarizing, so it’s hardly a surprise that it would go back to the same author and decide it was right, since the same source says the same thing on double-checking.

It also doubles down on the two-tiered audience structure – which it will later abandon. I’ll give you a hint here – think about the Parable of the Sower, from three of the four Gospels, and you won’t see two tiers. How about three?

And then there’s “However, there’s genuine disagreement about what this structure means for interpretation.” It’s disappointing that this wasn’t caught on the first AI pass. It says it’s going through all books. If that’s true, then the most reasonable explanation for missing it is the algorithm used to determine what to use and what to skip when generating a reply.

I don’t know about you, but my soul is too important to leave it to an algorithm to decide what matters to me. This points to a need for aggressive questioning on a regular basis – even if things look like we expected. How can we rely on an answer without asking, in all cases, whether there’s competing views to the one given to us?

---

---

Was the Sermon on the Mount meant as a universal ethic or does it only apply to devoted followers?

I’ve read it considered to be Jesus’ manifesto in John Stott’s “The message of The Sermon on the Mount”. That sounds like something Jesus said for everyone. But then, there’s a difference between what God wills and what God desires. And don’t forget the Parable of the Sower.

I can’t help but wonder – why do scholars seem to feel a need to make things so complicated, when Jesus makes it clear? The overall context provides lots of answers to the more specific questions. In this case, no matter who puts out a manifesto – God or human – not everyone will follow it. Does that mean the manifesto wasn’t directed at everyone – or is it simple that some chose to do something else?

This gets so difficult – figuring out what should’ve been a question about whether it was the same crowd.

And it gets worse – the AI is going to continue insisting on the two-tier system that it will ultimately abandon.

---

Notice that there’s no specific reference for this portion. But it does go back to the idea of trying to align Exodus/Mount Sinai with the Sermon on the Mount. But in Exodus, the people told Moses to go to God without them – and in this case the people went to God, uninvited.

Why does it need to be so hard? BTW – if you haven’t noticed, all these things I’m pushing back on are not related to my question. And yet, I can’t leave them unless I’m willing to taint all future responses by accepting these.

But enough of this track. Let’s move on to a different issue.

---

3. Did Jesus always stay and teach the crowds?

Part of what was in the details from one of the authors cited above had to do with a claim that Jesus always spoke to/taught the crowds of people that gathered. But, is that really the case?

More correction from the AI Agent. Let’s see how this one goes. However, before we begin, I submit that once again – there’s an unwarranted conclusion.

I dare say – very important.

I was also surprised at the statement: “the pattern is far more complex than I initially suggested”. I mean, pattern matching is a huge part of how AI does “its thing”.

Let’s see how/if it recovers from this.

---

I want to be clear on something in there: “Yet the Gospels reveal a persistent tension: when Jesus went to a solitary place at daybreak, people searched for him and tried to prevent his departure.”

Because of the wording, the implication is that a pattern of Jesus leaving at daybreak and then the people try to prevent His leaving the area.

First of all – it was one event – from Luke 4:42. One event does not make a pattern.

Second, in case there’s any doubt, while people may have tried to prevent Jesus from leaving – that just wasn’t possible. Remember – while He walked the earth, Jesus was 100% man and 100% God. No one could’ve prevented Him from doing what He wanted to do, especially since everything Jesus did and said was the Father’s will.

---

Having said that, there was a pattern.

While AI said: “a retardation that suggests intentional separation from apparent success” – I don’t care to make any assumptions as to what it meant by that. No meaning of the word “retardation ” that I could find fits in that scenario.

Also, the inclusion of the word “apparent” success indicates that maybe at least some of Jesus’ mission may have failed. That thought just isn’t there.

Finally, Jesus’ withdrawal, presented as a trend to “replenish his relationship with God through prayer and to build strong bonds with selected followers through extended conversations away from normal distractions” is only partially related to a trend. Prayer – conversation with the Father – absolutely was a trend, and one that we should follow. However, the part about building bonds with selected followers was pulled from the book without context. In the book, Bruce Barton has a list of what he calls “strategic withdrawals”, one of which the quote comes from – that being before Jesus’ arrest in the Garden of Gethsemane.

Unfortunately, the remainder of this response doesn’t get any better.

---

It’s telling that there aren’t any footnotes for this segment of the response. This is a complete synthesization by the AI from the previous paragraphs for this one question.

As such, they are, well – incorrect.

First – the conclusion that the pattern suggests Jesus balanced public ministry with genuine need for solitude, prayer, and focused discipleship training …” is greatly overstated.

The greatest issue comes with the idea that “the crowds’ persistence frequently thwarted his intentions”. As if that was even possible.

By this time, hopefully, it’s evident that asking questions about God pretty much requires that we already know the answers that we’re seeking. As a computer that gathers patterns from the various things it’s “trained” on, AI has no real “knowledge” or real “intelligence” related to God. And they certainly don’t have the Holy Spirit for discernment.

As for me, most of the time I ask questions about God in that manner, it’s usually for one of two reasons:

1. I already know what I want as an “answer”, but AI can put it into a more readable form than what I usually think of. But even then, I still reword even that to use the words I’d use, rather than the ones it gave me. But it can bring some order to my thoughts.
2. I’m doing research – and want a starting point. From there, I’m free to use it, push back, or just do it the old-fashioned way by searching my library without using AI. More time consuming – but less frustrating than when AI goes off the deep end or just can’t comprehend what I’m really after.

---

4. Time to try for a refreshed viewpoint

I was really hoping, from frustration, as much as anything, that this would be the last question.

It was not.

But since this is about how to be sure we get something useful from AI, maybe it’s not as bad as I thought.

I’m actually not going to go over any of this in detail.

• It still doesn’t address my question about the crowds from chapters 4 to 5. Although, that may be my fault for not pushing that very hard right away. This is an example of AI going off on a tangent if we let it.
• Even as it goes over the complexity, yet again, it reaches some of the same conclusions and pulls out single instance examples to try to prove a trend – without ever reaching the core of that trend.
• But then, in its wrap-up sentence, it says: “Critically, the crowd functioned as the recruitment pool from which Jesus built his core disciples”

Say what? Jesus had a recruitment pool from which to build his core disciples? When did that happen?

---

5. Was there a talent pool that Jesus used to call His disciples?

Jesus didn’t examine talent pools to find His core disciples. That one deserved an immediate question.

An “imprecision”. It was important, to be sure. But an imprecision? I do believe it’s more than that.

---

Here comes the three-tier structure. Replacing the two-tier it was so insistent on earlier. Nothing’s sad about the loss of the parallel to Mount Sinai in Exodus.

While I do agree to something of three tiers, it’s not like this. If there’s a “structure” it goes along the lines of the Parable of the Sower, having to do with the Holy Spirit, not how far away they were from Jesus in a physical sense.

This appears to be another case where an AI pattern search can’t relate to the importance of the spiritual realm in what’s “seen” as a physical world.

A larger disciple pool? If we truly have the free will choice to follow Jesus or not – the potential disciple pool is everyone.

At least this part of the answer goes into people moved by faith, rather than Jesus selecting them.

However, the larger issue is with the seventy-two. They are only mentioned in one passage in Luke: Jesus Sends Out the Seventy-two. In that passage they go out, they do things, and they return. There’s nothing else in scripture about them, so no indication that they were ever sent out again by Jesus. There were efforts in the early church, in extra-biblical history to try to extend their timeframe, but no such evidence exists in the Bible.

---

6. Did Jesus have a middle core of disciples?

I’m exactly sure why I asked one more question. I think I was trying to bring this to a “happy ending”, where it all came together somehow. Sadly, but not surprisingly, it didn’t happen.

And so, this is the last question. Regardless of what kind of ending it creates.

This question was asked along with a list of passages for those in the twelve whose calling was recorded.

Conclusions: Can AI analyze Christian theology? Not this one.

In case you have any doubts, the adjacent image shows my final decision about this sequence of queries and AI responses.

After denying acceptance of every response in the chain, the final status on this line of inquiry is “Rejected”.

There’s really no choice. After that many attempts to clean up the train of thought in those responses, it’s just not going to happen.

In one way, it’s normal. I’ve found that the AI Agents I use don’t go off the deep end anywhere near as much as they used to.

But then, once they get “off the rails”, so to speak – once they lose track of the intended track of the queries – they cannot get back on. There’s just too much implicitly accepted misinformation to get it back. As I said, in this case, some of that’s on me for not pushing back on the focus of the first question.

On the other hand, there were so many issues with the responses, I’m not sure it would’ve mattered. When multiple concerns show up in one response, you really can’t correct them all at once.

I’ve found this same experience when using AI for coding and other more technical projects. Once it goes off track, the only thing to do is take the “conversation” up to just before the train of thought was lost – start a new conversation/chat – feed that back into it – and try again to reword the question that caused the issue.

What went so wrong here?

Once I gave up, I wanted to go back and see what went so wrong so quickly. It turned out to be primarily two things.

1. The very first issues showed up because the authors it chose to use represented very different methods of scholarly research – as well as very different end goals of that research. If it could have isolated them, then presented them as different views of my questions, it may have been useful. For instance, when I have questions that require nuanced answers, Claude is very good at responding. LOGOS has problems from time to time – but I’ve never failed to get it back on track again.
2. The inability to get back on track may be due to another issue brought on by the authors it selected to reply. Some were Calvinist and some were Arminian. That introduced some theology based on God choosing ahead of time who would be saved and who wouldn’t on the Calvinist side – as opposed to salvation based on our own free will choice. Maybe it’s not obvious, but that very much changes the dynamic of what it looked at as pools of potential disciples.
   • That led to a complete misunderstanding of the selection of disciples and apostles – the mess it had with multiple tiers of followers – the complete miss of the Parable of the Sower – and on and on.
3. Then, once it decided to synthesize those varying study methods, objectives, and theologies into one – there was no hope of return.

Verifying what AI says

You probably figured out – I have most of the referenced sources in my eLibrary. More than enough of them to figure out the items addressed above. Because of that, I could determine the emphasis of the authors and determine that they were quite varied and never should’ve been merged into one.

I was also able to read the actual quotes, see the context within which they were written – contrast the context within which they were used by LOGOS AI – and determine that they were misused. Most commonly, statements in the books that were meant for a single instance were turned into trends by the AI. My dog, Mikey, can take a one-time event – like a Starbucks “pup cup” and turn it into a trend, expecting one even if I go to Starbucks without him and bring back a coffee but no pup cup. It’s cute – he just stares at my cup and you can tell he’s wondering, “where’s my pup cup!?”. But what happened here wasn’t cute.

If you ask a question like this – don’t have at least enough knowledge to realize something’s not quite right – don’t have the resources to be able to check what it said – it’s misleading, at best. It’s disastrous at worst.

Will I continue to use AI?

Yes, I will. But this experience will lead to some changes.

• Even if I get the reply I’m expecting, I’ll still ask if there are alternative thoughts on what something means.
• There won’t be the same level of trust when it quotes an author I’m familiar with, realizing that it can misuse the excerpt so badly.
• I already use one AI to help double-check another. There will be more of that. The only issue there is the price of Claude. It’s the most expensive – also appears to be the “deepest” and “best trained” for what I do. Unlike other AIs, you can see its analyzing process displayed on the screen as it works through questions, so you have an idea what it’s doing. Plus, it can deal with multiple scholarly methods/theologies better – and give split analysis more often.
• For full disclosure – I use MS Copilot for analysis of my writing and helping with things to be found by things like Google search and various AI bots. I also use Google Gemini for most of the images (at this time), although it changes as they leapfrog each other on capabilities. And then, I also use Copilot, Gemini, and Claude to double-check LOGOS and each other.

I am disappointed in this set of answers from LOGOS. To that end, I submitted the entire “conversation” to their support group. They said it would be sent to the appropriate group to analyze. According to the AI itself, it’s not capable of making the distinctions I asked for. However, I know that’s not entirely true. If asked to – it can give one answer from a Calvinist point of view – then another answer to the same question from an Arminian point of view. The catch is that we need to ask for it that way – it won’t do it on its own.

As of June 20th, 2026 – this is where it stands.

But I do have hope they’ll make it better. I expect they will, since it’s already improved over its short life.

So – yes, I’ll use AI – but with care.

As for you? You need to decide whether you’re in a position to use it safely – and if you want to take the time and go to the expense to do so.

---

Final thoughts and what’s next?

One of the clearest examples of things going off the rails was when AI began describing Jesus as if He were scanning crowds for “potential disciples.” That’s simply not how the Gospels present Him. Jesus didn’t sift through a pool of possibilities – He called the people He intended to call, after a lengthy period of prayer with the Father. The Twelve weren’t discovered; they were chosen.

I’m not going into the full theological explanation here – that’s for a later article – but this moment shows how easily an AI can merge scholarly viewpoints, leadership theory, and sociological language into something that sounds plausible but doesn’t match Scripture.