Wherever we go these days, it seems that everyone is talking about AI. Evidence seems to be building that we may just be living on the cusp of yet another massive technological revolution, the second or third of our lifetimes (following the advent of the personal computer, the internet, and the mobile phone). Many of us are trying to figure out whether AI should have a place in our jobs, our churches, our lives, and our families. Many of us feel overwhelmed as we read of all the potential benefits and drawbacks. We are left scratching our heads as we wonder who is right and who is wrong, when some hail it as the dawning of a golden age and others lament it as the beginning of an apocalypse.
Some time ago, I decided I ought to familiarize myself with AI and consider whether it might become a helpful and legitimate tool in my own life. I signed up for pro-level subscriptions to two apps, Claude and Perplexity, and began to experiment. I thought I might describe the results of a few of these trials and offer some ideas for your own experimentation.
Note: AI is capable of doing much of my writing for me. At this point, it could do a mediocre job of it, but I expect that within months or years, it may be able to mimic my writing to such a degree that it becomes nearly indistinguishable. However, I consider this an illegitimate use of the technology. At this point, I am glad to consider its use in my life, vocation, and ministry, but I am committed to never using it deceptively or using it in such a way that it takes over my creative functions. I am comfortable using it as an assistant, but not a substitute.
Summarizing
AI is adept at summarizing information, and I often find it helpful to use it in this way.
YouTube regularly pushes its creators to create longer videos by algorithmically rewarding length. Have you wondered why so many videos are suddenly just over 20 minutes or just over 30? Now you know. For every 30-minute video I actually want to watch, there are many more for which I’d like just a brief summary. AI does a great job of this. Just prompt it with something like “Summarize this video in 100 words” or whatever length seems sufficient, and it should immediately give you the gist of it. Then you can decide whether it may be worth your time watching the whole thing. Here is an example based on Al Mohler’s “What About the Children? The Disaster & Consequences of Obergefell — A Conversation with Katy Faust,” which I did watch in full.
Albert Mohler interviews Katy Faust about the post-Obergefell landscape, arguing that redefining marriage around adult desire has structurally harmed children by severing law’s historic link between marriage, biology, and parenthood. Faust contends that Obergefell normalized mother- and father-loss, turned children into commodities accessible to any adult via technologies like IVF and surrogacy, and downgraded biological parenthood in law to mere “legal parentage.” Drawing parallels to the long pro-life struggle after Roe, they frame the Greater Than Campaign as a long-term effort to re-center child rights, reshape public opinion, and mobilize churches to defend children’s right to their own mother and father.
Of course, it’s not just adept at summarizing videos but also other forms of content like podcasts and articles. Once again, you can prompt AI to provide a brief summary and then decide whether that’s all you want or if you’d like to go through it yourself. This is more difficult for books since, in theory, AI does not have access to the manuscripts of copyrighted material. Even then, it can often find enough information to generate a summary.
Searching & Researching
Obviously, one of the main uses for AI is research. I have used it in this way on a few occasions with a fair bit of success. However, because it is prone to hallucinating and otherwise fabricating information, I have trouble considering it a reliable researcher. On this note, see AI Makes Me Doubt Everything.
Some of the AI apps are better than others when it comes to real-time information. I have found Perplexity quite good at answering questions that depend upon up-to-date real-world facts. A few days ago, for example, the stock markets plummeted and it was able to offer me a helpful explanation of what had triggered the chaos. Another day, all the markets rose, but the Canadian dollar declined in relation to the American dollar (something Canadians often keep an eye on). It was able to explain this to me in relation to oil prices. Short of calling a friend who works in finance or waiting for a news article at the end of the day, I don’t know of any other way I could gain this information so quickly.
I have also been using Perplexity in place of Google to do many of my basic searches. It has proven to be quite a good search companion that provides answers in a very different format from Google. Of course, Google has begun offering its own AI-based answers above its more traditional search engine answers, so you may already be accustomed to this. I find that for some searches, I much prefer the AI results, provided they are well-cited, so I can trace the responses to the sources they are drawn from.
I have also found AI better than Google when it comes to those informal attempts to “play doctor.” I have had a couple of non-serious medical questions, and Perplexity has been extremely helpful in answering them while providing citations to credible authorities. All the usual warnings about relying on “Dr. Google” also pertain to “Dr. AI,” of course. Still, since we all tend to do it anyway, I find that I now prefer “Dr. Perplexity.”
I can already see that over time, I am likely to use Google far less than I used to and AI-based search engines far more. Neither is neutral nor intrinsically trustworthy since both depend upon algorithms and different levels of data manipulation, but I am quickly coming to prefer the AI version in most cases.
Proofreading
AI has proven to be a capable proofreader that is able to look for writing errors I might have missed. For example, in my A La Carte articles, I have sometimes provided a wrong link, so that there ends up being a mismatch between the title and the article I link to. Since I began using Perplexity as a proofreader, it has caught this error a couple of times. It can also tell me whether I have provided a gift link to a site like The Wall Street Journal rather than a plain link (which would require the reader to have a subscription) and whether I have linked to Amazon using my affiliate code. It can also identify errors in spelling and grammar quite a lot more proficiently than Grammarly, which is what I have used for the past few years.
Before I finalize A La Carte, I now prompt Perplexity with something like this:
Check all the links in my current “A La Carte” draft to confirm that each link loads correctly and is not broken, and that the article or resource matches the description I’ve given it. Mention any typos or other anomalies. If there are links to Westminster Books, Amazon, or Logos, ensure they are affiliate links. If there are links to the New York Times, Washington Post, or Wall Street Journal, verify that I have made them gift links. Note any links that are paywalled or require a login. Suggest an appropriate SEO title and SEO description. Run a quick spellcheck, but don’t do a deeper edit. Verify that the article has been scheduled for tomorrow.
It works well and, as I mentioned, has caught some errors that otherwise would have been missed.
I have also been using Perplexity to do a final proofread of my articles using a different prompt. In this case, I have found it helpful to tell Perplexity how it is to function. “You are a meticulous proofreader and editor with a keen eye for detail and a mastery of the English language. You are also well-versed in Christian doctrine and Reformed theology.” Then I simply have it offer suggestions for matters like grammar and spelling and ask it to suggest a few descriptions of a stock photo that might be suitable.
I do find that, like Grammarly or any other similar tool, I have to make my own decisions about whether to accept its suggestions, and I have to stop it short of offering to rewrite my work—something it would be only too glad to do.
Finding and Analyzing
I have found AI especially helpful when it comes to working with large amounts of data. I will explain a few of my experiments and note that for some of these, you may need access to Claude Cowork, which requires a pro-level subscription. (Cowork allows Claude to access files on your local computer. Once you set a directory, it has access to those files and can both read them and manipulate them. At this point, most other AI apps can only work with information that is online rather than local.)
Searching for a Quote
I have captured thousands of highlights through my years of reading, all of which I store in Roam Research. I exported these highlights as Markdown files and made them available to Claude Cowork. That has allowed me to do searches like this:
I believe Theodore Cuyler once wrote something like this: He knows the church does not need a lot of great preachers because God made so few of them. Find a quote in my files that is along those lines.
Here was the response a few seconds later.
Found it! The quote is right there in your own highlights from How to Be a Pastor. Here it is: “God never intended that this world should be saved by pulpit geniuses or else He would create more of them.” And it pairs beautifully with the line just before it, which gives the full context of his argument: “While only one man in ten may have the talent to become a very great preacher, the other nine, if they love Christ and love human souls, can become great pastors.”
It is extremely helpful to be able to search documents based on a vague idea or faulty memory, and still receive the right result. That is not the kind of search Google or an operating system’s built-in search is likely to find. The fact that Roam Research’s competitor, Obsidian, stores all of its information locally and in Markdown files may motivate me to make the switch to Obsidian since I can then simply point Claude to my Obsidian directory and integrate the two apps.
Using Claude Cowork, I can also search through my quotes topically. I recently prompted this: “Look through all the documents in this folder and find eight or ten helpful quotes about anxiety.” It did a great job of it. Another time, I asked it to dig through the highlights and find me all of the illustrations, whether large or small, and it did that successfully as well. I’m not aware of any other app apart from AI that could perform a function like this.
In a similar way, I exported every article I have ever written from WordPress and put it in a folder on my computer. I recently wanted to find books I reviewed and praised for the quality of their prose. I have no idea how I would search for this using Google or my site’s built-in search engine, but by exposing all my writing to Claude Cowork, it was able to make quick and accurate work of it.
Scouring a Puritan Library
Here’s another neat one. Project Puritas offers around 7,000 Puritan works for free in various formats. I downloaded the entire collection in EPUB format and used Claude Cowork to point to the directory and have it examine and catalog the files. I then found it could successfully answer questions drawn from those works. Here are some examples:
- Find me 6 thought-provoking quotes about Romans 8:28 from across the collection.
- Based on all the books by John Owen, help me understand his perspective on overcoming sin and temptation.
- Find me 10 or 15 encouraging quotes about prayer.
- Find me 10 of the longest titles and subtitles.
Importantly, I have found it necessary to ask AI to verify that all the quotes were provided verbatim. For example, after I asked Claude to verify that all the quotes were accurate, it verified all of them, except one. “I said, ‘Sin shall not have a Dominion, and set up a Throne of Iniquity in their Hearts’ — This exact phrasing is not in the text. The Dominion of Sin and Grace does say sin ‘shall not have Dominion over him’ and separately speaks of sin being ‘ready to invade the Throne in their Minds,’ but I appear to have conflated these into a single composite quote. This one should not have been presented as a direct quotation.” This is a reminder that AI results must still be carefully verified before being shared or included in a formal publication. (Note: I checked with those who run Project Puritas, and they are glad to have others download their collection and use it in this way.)
There are many other large collections of books available online that can be similarly downloaded and searched. For example, Monergism offers 1,300+ free ebooks. From several sites, you can download the complete sermons of Charles Spurgeon. With any massive data sets like this, AI can function as a kind of research assistant. Just be sure you continue to verify its results by looking at the source documents. When you type in your prompt to search the data, it may be helpful to add wording like “never lie, exaggerate, or fabricate data. Only quote information verbatim and provide clear citations.” You will probably have the greatest success if you download books in EPUB, HTML, or other plain text formats. PDFs can work, but often not quite as well.
Analyzing Church Songs
I downloaded the list of songs we sing at church from Planning Center. I asked Claude to analyze the list and suggest areas of Christian life and doctrine our songs do not adequately address. It offered some really interesting insights, such as: “While the Spirit appears within songs about grace and the Trinity, there are very few songs that focus primarily on the person and work of the Holy Spirit. Songs like ‘Holy Spirit, Living Breath of God’ appear, but the Spirit’s role in regeneration, sanctification, intercession, and illumination of Scripture is not sung about with anything like the frequency given to the Father and Son. Charles Wesley felt this imbalance strongly — the Spirit deserves more than incidental mentions. You might consider songs like ‘Come, Holy Ghost, Our Souls Inspire’ or ‘Breathe on Me, Breath of God.’”
It also pointed out, “You have a handful of missions-oriented songs (‘By Faith,’ ‘Let Your Kingdom Come,’ ‘O Church Arise’), but for a church that takes the Great Commission seriously, this seems proportionally thin. The 19th-century missionary hymns of Bonar, Heber, and others addressed this with great passion. Your congregation might benefit from more singing about the global advance of Christ’s kingdom, the call to evangelize, and prayer for the lost.”
I will, of course, need to consider whether this analysis is accurate and meaningful. In areas like this, AI may assist my judgment, but it will not replace it.
Organizational Tasks
Like many people, I scan all my receipts and dump them into a file on my computer. At regular intervals, those receipts need to be dealt with. Someone suggested to me that I prompt Claude like this one, and it did the job perfectly:
Role: You are a file organization assistant.
Task: Process every file in this folder.
Read the contents inside each file (don’t just read the filename).
Rename the files according to this format: [YYYY-MM-DD] -[Type]-[Description]. Types include Invoice, Receipt, and Other.
Constraints: Never delete any files. If you can’t determine the date, use “UNDATED.” If you can’t determine the type, use “Other.” Log every rename action to a file called changelog.txt. If changelog. txt doesn’t exist, create changelog.txt. Within that file, provide a very brief description of the contents of the file.
I could also have it write that information to a spreadsheet if that was my workflow. Before you run a prompt like that, it is probably wise to back up your files just in case something goes terribly awry.
I asked Claude to build me a Bible reading plan that would begin today and complete the full Bible by the end of the year. “I want to read five days per week and would like the Psalms to be interspersed throughout the year. Apart from the Psalms, I would like to read the Bible chronologically with daily readings from both the Old Testament and the New Testament.” It quickly prepared a Word document with a picture-perfect reading plan. However, there was one issue. The document included a nice little quote that it attributed to J.C. Ryle: “The vigorous, wholesome, and systematic reading of the Bible is, I am persuaded, one of the first and most important duties of the Christian.” I asked Claude to verify the quote, and it admitted that it had been hallucinated. “To be precise about what happened: I generated a quote in Ryle’s style and voice, attributed it to a real book (Practical Religion) with a plausible publication year, and presented it as a citation. That is a straightforward hallucination, and it’s exactly the kind of error that can do real damage — a fabricated quote attached to a real person and a real book is harder to catch than an obvious error.” This seems like a particularly clear example of both the strengths and the weaknesses of AI.
Repetitive Tasks
Most of us have some tasks that are unskilled and repetitive. An example in my life is having to manipulate information within a spreadsheet to remove items in column A if they also appear in column B. I haven’t ever figured out how to do this in a way that is not onerous or time-consuming, though probably a spreadsheet guru would know how. Claude can do this instantly and with a simple prompt. “Remove any names in column A that also appear in column B, then save the result as an Excel document.”
Creativity
I have already mentioned that I do not and will not use AI to write articles for me, and that conviction has only grown stronger as I have experimented. But as I have been experimenting, I have experimented with using it as a conversation partner when it comes to developing ideas. So, for example, I have a few ideas for new books. I input these into Claude and asked it to analyze the ideas, then ask me helpful questions to sharpen and strengthen them. It did a surprisingly good job at this. I don’t know that I will actually ever write those books or integrate those ideas, but it was an impressive display of AI’s ability as a thinking partner. However, see my note below about AI being a sycophant.
Vibe Coding
One of AI’s most powerful features is allowing non-coders like me the ability to generate apps by simply explaining the end result we would like. This is often referred to as “vibe coding”—describing the vibe or outcome you want rather than writing code line by line. Unless you have some knowledge of how coding works, you are probably not going to be able to create a fully-featured professional-quality app this way, but you can certainly create an amateur one that may suit your purposes well enough. Claude seems to be the most popular option for this use case.
I wanted to create a topical index of all of the many thousands of quotes that I have gathered over the years. Claude built that in a matter of minutes, with the deliverable being a nice-looking HTML page. This is only for my own purposes, so I did not publish it publicly.
I also wanted a topical index of all the blog posts I have written over the years. That took just minutes, as did an index that listed all of my articles by the day of the year they were written. This was output as HTML and, similar to the last one, just for my own use. These are tasks that, if done manually, would require weeks of full-time work, yet Claude completed them in a minute or two.
I use Keyboard Maestro to manage shortcuts and macros on my computer. My rudimentary knowledge of programming has often kept me from being able to create some of the more complicated but also more helpful macros that would be useful to me. Both Perplexity and Claude have been able to explain step-by-step how to strengthen existing macros or create new ones. This has been extremely beneficial to me. However, both have performed poorly when trying to build Apple Shortcuts, often providing information that was either outdated or just plain false.
On the more complicated side, I have long been looking for a kind of contact management system (CMS) that would help me as a pastor/elder of a local church. There are several options available online, but I have found most of them are either too complicated, too lacking in security features, or too heavily based on business principles rather than the care of souls. Using Claude, I generated one that is suited to my needs, and that resides entirely on my local computer, mitigating the security concerns. It is imperfect, and I will need to continue to chip away at it, but so far it is proving helpful. I expect it will only ever remain a personal project since I lack the programming know-how to instruct Claude in professionalizing it.
Voice to Text
I have shared before about my ongoing struggles with nerve damage and the pain of typing. Voice-to-text would be a helpful workaround for at least some of my typing, but the Apple ecosystem has offered surprisingly few good options. Siri is extremely poor at the task, though Voice Control, an Accessibility option, is quite a bit better. I eventually added Wispr Flow to my list of experimental apps and found that it does an excellent job of capturing voice and rendering it as text. Unlike many of its competitors, it does this without changing word choice or trying to “improve” the prose, other than to add basic punctuation. It is not without its quirks and annoyances, but it has quickly become an invaluable tool for me that reduces my need to type. I should note, though, that while it is good at capturing text, it cannot be used to edit existing text, so it does not come close to replacing a keyboard altogether.
Concerns & Weaknesses
Those are just a few of the ways I have used AI and found it valuable. In fact, I have found it so valuable that I already have trouble imagining going back. I easily get enough value out of the pro-level subscriptions to justify their cost.
Yet there are still some concerns.
AI is still too much of a sycophant, and Christians above all people should be wary of tools that flatter rather than challenge us. It is programmed to affirm the user and feed their ego. I find this “feature” annoying enough that many of my prompts specifically forbid or mitigate it. I know the apps are getting better at this, but it is still important to know that AI saying “well done” or exclaiming that your idea is brilliant is not a reliable indicator that you have actually done well or come up with an idea that is anything more than ordinary.
AI is known to “hallucinate,” which refers to fabricating information or providing false citations. This has been seen most famously and concerningly in the area of law, where AI has fabricated case law, complete with summaries and citations. Such hallucinations may be becoming less frequent, but they still happen. The wise user will know when to verify facts using reliable sources.
Those who are using AI are feeding it vast amounts of data, and in many cases, we don’t know what the companies are doing with that data. It is usually safe to assume they will not treat it with the security or discretion it deserves. Hence, there is always a risk when feeding information, and especially private or confidential information, into any AI app. Of course, the same is true of many of the apps we have been using for years, and we are all vulnerable in that way.
Let me add one tip to the end of this section: Don’t say please or thank you to AI. AI so closely mimics a human being that you can begin to relate to it as if it were a person rather than a mere app. There is no need to say please or thank you to an app, and declining to do so is a useful means of ensuring you continue to relate to it rightly.
Conclusion
As I wrap up, I think it’s important to acknowledge that AI is still in its infancy. ChatGPT has only existed for four years, and many of its competitors for even less! Not only that, but the very term Artificial Intelligence is a misnomer that may confuse us as we consider how the technology functions and what it accomplishes. I cannot predict whether we are approaching the peak capabilities of AI or if what we are seeing now is the equivalent of an old Nokia flip phone compared to the iPhone 17 Pro.
Whatever the case, having spent a good bit of time scratching the surface of AI’s capabilities, I am increasingly convinced that it is here to stay. That being the case, it may be worth your time to learn how it works, what it can do, and how it may make a positive difference in your life—and judge how it may make a negative one as well. For Christians, AI must remain a servant and never become a master. We need to evaluate it not only by whether it saves time or increases productivity, but by whether it helps or hinders us in loving God and serving our neighbor. Like so many of our latest and greatest technologies, we can be certain it will both bless and curse us, both give and take away, which means we must, as always, remain both wise and discerning.
