This open letter first appeared in https://thatdougsmith.com

Like millions of Bible students, I’ve been a grateful student of your Bible study resources for many years. I became a paid subscriber to support your mission. I’m a fan of your clean user interface, the quick ability to navigate from verses to original languages, the parallel Bible versions, and the helpful commentaries.

But while researching my new AI book, I noticed that your Topical Lexicon smelled like LLM-generated word sausage. I searched to see if you admitted to using GenAI, but found no authorship of any kind on these passages. I emailed your support team and received an email confirming that you are using GenAI for most of the Topical Lexicon entries.

I’m grieving the toxic poisoning of another trusted Christian resource by the scourge of generative AI. My heart breaks for the tragic loss to Bible scholars around the world.

I find your use of GenAI a betrayal of a sacred trust. Do you not know that millions of people come to BibleHub.com for clarity, precision, scholarship, and timeless wisdom from authoritative commentators? We don’t come for ultra-processed word sausage, no matter how plausible sounding the words may be.

And I especially feel betrayed that you didn’t declare your use of LLMs plainly. You show copyrights on Bible versions, commentaries, and several other resources, as you should. But the LLM content just stands with BibleHub.com’s authoritative name behind it. And since your name has meant so much to us, your readers are led astray.

On “Word Sausage”

The statistical, confabulated output of LLMs has been labeled by many colorful metaphors. I’m coining what I believe is a new one here. “Word sausage” captures a quality of LLM-generated words that we can almost taste when we think about it.

Most meat sausages are ultra-processed, and contain ingredients we don’t want to know about. What entrails, scraps, and questionable parts of various animals were squeezed into the grinder? Just put them into the machine and grind them into an unrecognizable form, so that after it’s all mixed together with special spices and wrapped in a bun-friendly size, all we get is the yummy taste. Pass the mustard!

When I asked BibleHub.com support whether you use LLMs, your representative confirmed that you do. But he said you also proofread and edit everything before it is published. Unfortunately, that simply doesn’t work. It’s like trying to extract bad ingredients out of a sausage. Once unsavory ingredients are put through the grinder, there’s no way to sort them back out again. The only remedy is to throw away the tainted sausage and start a new batch with good ingredients.

It’s the same with LLM-generated word sausage. Once the words are generated, they frame the conversation. They sound good. They set the tone, style, direction, and especially the objective. So your proofreaders are already at a disadvantage. Plus, I imagine that they’re tired, overworked, and limited people. They can’t possibly notice every detail, especially when the word sausage grinder adds plausible sounding words that are just a little bit off.

In response, some people say, “human authors make mistakes too, and proofreaders can’t catch them all either.” But that dehumanizing comparison between humans and machines comes from Big Tech’s materialistic world view, not the imago Dei taught in the Bible. Humans are not machines, full stop. Importantly, human authors know what they’re talking about, and Bible scholars are (hopefully) inspired by the Holy Spirit. LLMs have no understanding of the meaning of a single word. So the comparison between human and machine fails, and the word sausage metaphor remains.

My wife is an author with several stories published in various Chicken Soup for the Soul books. She recently received this addition to their submission guidelines:

Do not use AI. We do not accept AI-generated or AI assisted submissions. We have observed that even running your story through Grammarly or any other AI tool after you have written it does not make it better. It makes it worse. Please do not use AI to generate or “improve” your story. Do not use AI at all.

They get it. AI “does not make it better. It makes it worse.” AI has made your service much worse, too. I hope you’ll take their advice.

Examples From BibleHub.com

Here’s part of the Topical Lexicon entry where I noticed that BibleHub.com was using LLMs to generate their content. The familiar Proverbs 22:6 starts with, “Train up a child in the way he should go…” I wanted to know more about the Hebrew word for “train.”

BibleHub.com linked to the Strong’s concordance number 2596, chanak. Good start. Your page shows that the word isn’t used very often in the Old Testament, but its semantic range includes “dedicate,” “dedicated,” and “train.”

The exhaustive concordance entries helped me understand that the word properly means to narrow, but figuratively means to initiate or discipline. The cross-references showed that the other usages for chanak involve dedicating houses, whether their own house, or the temple of the Lord. All that was great. And if it stopped there, I wouldn’t have written this letter.

First Example

But then came your Topical Lexicon. It starts with this “Essential Idea”:

חָנַךְ portrays the act of setting something—or someone—apart for an intended, God-honoring purpose. Whether the object is a house, the Jerusalem temple, or the heart of a child, the word gathers up the ideas of initiation, consecration, and disciplined preparation. This sounds really plausible and uplifting. That’s why it’s so dangerous. It’s precisely this kind of ear-tickling that Paul warns against in 2 Timothy 4:3-4 (NASB):

For the time will come when they will not tolerate sound doctrine; but wanting to have their ears tickled, they will accumulate for themselves teachers [or AIs?] in accordance with their own desires, and they will turn their ears away from the truth and will turn aside to myths.

What really struck me in the LLM-generated “Essential Idea” was the dehumanizing comparison of a child’s heart to a house or even the temple. It calls a child’s heart an “object.” A machine wouldn’t know the difference, would it?

The problem with all LLMs is that they have no idea what they’re saying because they have no ideas. They do not know what one word means. They only know which token most probably follows the last, but that doesn’t lead them to truth.

And for Bible study, that’s not good enough, not by a long shot. (LLM word sausage isn’t good enough for anything that matters, but especially, not Bible study.) So yes, chanak does involve dedicating a house, or initiating a child, depending on the context. But blending all of that to objectify a child’s heart is dangerous, because it sounds so good on the surface but it’s so wrong as a concept.

But the LLM linked the words together, grouped them into its predictable, algorithmic groups of threes, with, “house, temple, heart” and “initiation, consecration, disciplined preparation.” The winsome-sounding but exaggerated, imprecise, and wrong verbiage fills the entire Lexicon entry. But I’ll pick just a couple more examples to keep this as short as possible.

Second Example

Under “Lifelong Training and Moral Formation,” the machine quotes Proverbs 22:6 again and then says, “Here חָנַךְ shifts from dedicating buildings to shaping lives.” The word “shifts”? The word itself is taking action or changing its meaning? To me, on examination, it’s meaningless but also slippery. Words don’t act by themselves, but change in context according to the meaning intended by the author.

But most problematically, again, the LLM-generated sentence includes a dehumanizing link between buildings and human lives. What happens when generations of Bible students are discipled by your LLM and regularly compare humans to inanimate objects, just like when AI boosters compare LLMs to humans? Mass dehumanization never ends well.

Third Example

Under “Theological Patterns” (because we want LLMs doing theology?), the word sausage grinder extrapolates broad platitudes that seem far afield from the meaning of this word. Sure, again, they sound good, but with your name on them they imply a scholarship that simply isn’t there. The trajectory is away from clarity, not closer to it. Like the third point (because there are almost always three),

True initiation anticipates perseverance. Proverbs 22:6 moves beyond a ceremonial moment to a lifelong trajectory, foreshadowing the New Testament theme of perseverance in grace (Philippians 1:6). This sounds deep, but on reflection it’s about as profound as a fortune cookie. And the imprecision abounds. There’s no “ceremony” in Proverbs 22:6. Yes, the “way they should go” may imply perseverance, but that’s not the focus of chanak.

But what about the claim of “foreshadowing the New Testament theme of perseverance in grace.”? That’s a huge, unjustified leap. A pleasant thought, no doubt, but not for an authoritative reference claiming to produce “Theological Patterns.” And Philippians 1:6 isn’t about Christians “persevering in grace,” but about God continuing his good work in us. So that cross-reference is just wrong.

Can’t Unmix This Word Sausage

Again, every word generated by a chatbot is confabulated. They sound good, even educated. But they have no grounding in truth. After scratching the surface, we can see that we should not be passing LLMs off as authoritative guides anywhere, let alone on one of the most popular Bible study websites in the world.

It took a lot of time to read through this single Lexicon entry to figure out why it seemed so “off.” This is what I mean by word sausage. You can’t unmix the recipe. I can’t possibly find all the ways this single entry goes sideways and leads people to questionable and wrong conclusions. And no editor or proofreader at BibleHub.com could possibly make LLM-generated content worthy of the time of the students who depend on your site for its (previously) dependable and authoritative scholarship.

Pastors and teachers may riff on Bible words under the inspiration of the Holy Spirit to bring needed truths to their congregations. But LLMs do not have the Holy Spirit. And your readers aren’t looking to BibleHub.com or any other authoritative Bible study resource to serve us with confabulated word sausage.

You should only offer scholarship written by Bible scholars who know what they are talking about. If you don’t have authoritative, human-written content for a word or verse, then please, don’t publish anything at all.

I’m particularly concerned about the harm to next-generation Church leaders. Young pastors, Bible students, and upcoming scholars all trust you. I almost quoted your LLM in my book, until I smelled the tone and pattern of LLM-generated word sausage.

With young leaders trusting your now untrustworthy content, they’ll propagate the tainted LLM-generated words instead of the Word. Bible teaching quality will degrade as it spreads from teacher to student until millions of Christians come to believe things about the scriptures that simply aren’t true.

My Plea

I’m asking you, my brothers and sisters at BibleHub.com, to remove all LLM content from your website. I know that you’ve spent a lot of money and time in hope that GenAI would yield the promised efficiencies and utopian dreams of Silicon Valley elites. But you’ve been taken in. You’ve been duped. Deceived.

You’re not alone. Many Christian ministries have also been duped by Big Tech. It’s the spirit of the age. But you can rise up, do the right thing, and admit your mistake. Lead the church by your example and leave GenAI in the dustbin of history where it belongs.

Christians are people of grace and truth. We become Christians by repentance—by turning around. I’m asking you to turn away from your use of LLMs. If you do, I’ll happily subscribe again and promote you as far and wide as I can. Thank you for reading and considering my concerns. God’s best to you all.