Podcast: Play in new window | Download
Cara Ray, host of The Writers' Bloc Podcast, had me on to talk about how Christian writers can make wise use of AI without losing their voice, their faith, or their integrity. We covered which models help most with research and writing, how to prompt in paragraphs instead of sentences, why Grammarly can sand off your voice, and how to take full responsibility for every AI-assisted word.
Key takeaways:
- AI is already woven into nearly every digital tool writers use, from spell checkers to search engines, so the question is not whether to engage with it but how.
- Grok is the most research-friendly and truth-seeking of the major AI models and is the best option for writers with theological or nonfiction research needs.
- Claude is currently the strongest model for writing-related tasks and marketing copy.
- Writing better prompts matters more than choosing the right model. Use paragraphs, not sentences, and give the AI as much context as possible.
- Tools like Grammarly can quietly erase your voice by framing stylistic preferences as absolute rules.
- Using AI projects and project instructions helps you get consistent, personalized results across every conversation.
- Writers should take full responsibility for every word of AI-assisted output, just as Solomon took responsibility for building the temple.
Who Is Thomas Umstattd Jr.?
Cara: Welcome back to The Writers' Bloc. I am so delighted to welcome Thomas Umstattd Jr. with
us today. He is an award-winning speaker who teaches creative people around the world how to build their author platforms, sell more books, and change the world with writing that's worth talking about. As a podcaster, he hosts the Novel Marketing Podcast as well as the Christian Publishing Show, and he is currently serving as the CEO of Author Media. Welcome to The Writers' Bloc, Thomas.
Thomas: Thanks for having me.
Cara: Tell us more about what life looks like for you.
Thomas: I have been helping authors full-time for over a decade, almost 20 years now. I started off
building websites for authors, and that grew into a marketing agency, a podcast, and courses. I was a literary agent for a while and a marketing director. I woke up one morning and realized I couldn't get out of bed. I was so overwhelmed with how many things I was doing.
Why Writers Cannot Afford to Ignore AI
Cara: You've become something of a subject matter expert in the AI world, which I think is really
rocking writers in particular. Writers either feel really excited about it or really threatened by it. What's your take? Why do you think writers need to make friends with AI, or should they?
Thomas: AI is an irresistible part of the market. Even if you are drafting with a ballpoint pen or
typing on a typewriter, you are still competing with authors who are AI-augmented, either in their marketing or in their writing. Since ChatGPT was released, the number of new books published every day on Amazon has tripled. Assuming all of that growth comes from AI-enabled authors, that means two out of three new books published are written with the aid of AI.
But we know that many previously successful authors are using AI at some point in the process. In fact, it is almost impossible not to use AI once you understand what it is, because AI powers the spell checkers, powers the grammar checkers, and powers the search engine. If you have done a single Google search, you have read an AI summary. Google now is just a wrapper for Gemini.
There is actually something to be said about drafting with a ballpoint pen. I did a lot of research on the origin of different technologies that have changed the way people write, and the typewriter is actually 80 years older than the ballpoint pen. There was no point in history when authors wrote books with a ballpoint pen. A bit crystal is actually like a magical interface for your brain. But there is also something to be said about using AI to help you with research or to find inconsistencies. There is a lot that can be done with it.
Everything on a computer is turtles all the way down. There is a moral panic about AI right now, but the people who are panicking over it cannot tell you the difference between machine learning and a large language model. They do not know how to draw the line between good AI and bad AI. They just know that they feel uncomfortable. The initial essays on neural networks go back to the 1940s. This is not new technology. If you are using a thinking machine to help you write, a word processor
is all AI. There is no pure version of Microsoft Word that does not have AI, because they all have a spell checker, a grammar checker, and basic neural networks that have been running in them for a very long time.
Cara: What do you mean by "turtles all the way down"?
Thomas: There is this ancient philosophy that underneath the world, holding up the earth, is a turtle.
And then another philosopher asks, well, what is underneath that turtle? And the answer is another turtle. And underneath that turtle is another turtle. It is just turtles all the way down. This is the challenge with people trying to draw a line when basically everything on a computer is using this thinking technology. You are either using a bit crystal or a typewriter, or you are using AI.
Machine Learning vs. Large Language Models
Cara: You mentioned the difference between a large language model and machine learning. Can
you explain that?
Thomas: Machine learning is a way for the computer to program itself. The human sets what the
success criteria is and then the system tunes itself based on user behavior to optimize for that criteria. So the Amazon algorithm is pretty honest about what it is trying to do. It is trying to maximize sales and specifically lifetime customer value. It constantly surveys everything you do and adjusts itself accordingly.
You will notice this on Facebook. If you watch one video on philosophy, suddenly you see nothing but philosophy videos. The Facebook machine learning engine is trying to maximize time on site, so it has basically tuned itself to be as addictive as possible.
Of the AIs, I think the most harmful ones are run by Meta. The Instagram one is particularly insidious. We have really good science showing it causes anxiety and depression. In fact, if you are experiencing anxiety and depression right now, there is almost no intervention that will be more effective than getting rid of all Meta algorithms in your life and going outside for a walk for 20 minutes a day. It is better than therapy, it is better than medication. Do not go on SSRIs while the Instagram app is on your phone.
The side effects of getting off Instagram are that you will be happier, healthier, less depressed, less anxious, and you will have more free time to write your book.
Are the New Books on Amazon Real?
Cara: If that many more books are going onto Amazon every single day, are those written by real
people using AI, or are they just automatically generated? Is there human ingenuity behind them?
Thomas: There is no such thing as a bot that operates entirely on its own. Every time you have
something automated communicating to you, there is a human puppet master driving it, because AI has no agency. So 100% of the books uploaded to Amazon have some human behind the scenes. The real question is how much of that author's process is automated and how much of their blood, sweat, and tears is going into the book.
I know authors who are running whole teams of ghostwriting using AI and publishing a book a day, making $10,000 to $30,000 a month. Most of the books do not sell very well, but some do really well. Some genres are almost entirely dominated by just one or two authors who figured out what the reader of that genre really wants and run half a dozen different accounts to give readers what they want.
One of the interesting things in this conversation is that while there is a lot of hostility toward a large language model generating text, there is far less hostility toward machine learning determining what you see and do not see. And in some ways, that is the scarier technology. That is what powers Facebook. That is what powers the Amazon algorithm. I find it ironic when authors go on Facebook, which is an AI-powered platform, to complain about AI. Every interaction with Facebook is powered by AI. They have the biggest AI engine powering their advertising model of all the big tech companies.
How Writers Should Think About AI as a Tool
Cara: How should we relate to AI? Should we make friends with it? And where do you personally
draw the line?
Thomas: I think the way to think about AI, and when most people say AI they really mean ChatGPT
or Claude or Grok or Gemini, is as a tool. A tool is a good tool insofar as it helps you do good things, and it is a bad tool to the degree that it allows you to do bad things or causes you to do good things in a less good way. To the degree that you are able to write better books faster, AI is a wonderful tool. To the degree that it causes you to write worse books or books that are less appealing to readers, it is a poor tool.
The Patron Toolbox: AI Tools Built for Authors
Cara: Tell us a little bit about your toolbox. You have created AI tools that authors can access?
Thomas: Yes. I started building AI-specific tools in the Patron Toolbox a little over a year ago. They
have turned out to be really incredible. The toolbox is kind of like training wheels for using AI yourself, because the best practices are built into the tools themselves.
For example, just today I was upgrading the timeline chronicler. It creates a timeline of events in your book. But there was a problem where a character might be lying about what they were doing. So the butler swears he was at the pub with his friends on the night of the murder. I just fixed the tool so it now notes things like "the butler says he was at the pub, but the narrator later says..." It creates a timeline you can use to make sure characters are aging at the right times and are not in two places at once. It does not put any words into your book, but it does catch inconsistencies. These are the kinds of things that are hard to check for, especially when you have four different versions of the timeline in your head, the original version and the fixed version and the final version and then final version two.
Cara: Is this geared mostly toward fiction writers, or do you have tools for nonfiction as well?
Thomas: I have fiction and nonfiction. Probably half or maybe 70% of the tools work for both. For
example, one tool writes a one-sentence pitch for your book, then a one-paragraph pitch, and then cover copy. That works for both fiction and nonfiction. It is not that you would just copy and paste this into Amazon, but it is hard to read the label when you are standing inside the bottle. Getting a third-person view of your own writing is really helpful and gives you something to start from.
Another tool is the plot snopsizer, which gives you a plot synopsis or a chapter summarizer. If you are going to be traditionally published, this is something that is expected. Authors hate writing chapter synopses. But if you upload your book and push a button, two minutes later you have a chapter synopsis that you then read and check. It might take an hour to read and tweak a chapter synopsis. It might take a week to write it from scratch.
Cara: So you are building a better mousetrap for some of those marketing tasks that come with
publishing and promoting a book?
Thomas: That is what has been happening over the last year. Often I will do an episode of Novel
Marketing where I am teaching how to do something, and in the process I realize I could create an AI tool that applies these principles. So it becomes like an AI tool companion to the episode. I still give away how to do it yourself on the podcast for free. Someone can listen to the 45-minute episode and do it themselves, which is the more enriching experience. But if you are pressed for time or need that third perspective, you can run the tool.
For instance, I did a whole episode on how the Amazon algorithm works. While doing that episode I created several tools to help optimize your Amazon page. One looks at your page and gives you feedback. Another reads your book and helps you brainstorm categories and keywords, and also texts optimized for your description that is both search-friendly and appealing to readers.
I also have a homepage scanner that pulls up your webpage and critiques it on dozens of factors I have taught about in my episodes. Then it gives you a score. Your homepage scored a 60 out of 100. Here are 20 things to do to boost your score and help you sell more books. Having an algorithm on your side is really helpful when the other side also has an algorithm.
Cara: So these tools are not things people access through ChatGPT or Claude directly. People go to
your specific website and use tools built specifically for fiction and nonfiction writers?
Thomas: That is right. Right now they are all done through my Patreon with my Novel Marketing
podcast. I may spin it off as its own separate product, but right now it is very much integrated with the free education I do on the podcast.
Each tool behind the scenes uses either Claude or Grok or Gemini or ChatGPT. Very few use ChatGPT. I find that in general it is not the best for any specific purpose. The different models have different strengths and weaknesses. Part of what I do as I am developing the tools is testing the different models. I have done a ton of research on where certain models outperform others.
For example, I have a bunch of compendium tools that give you a list of all the characters in your book, where they were at the beginning of the story, where they were at the end, how to pronounce their names, and some guides for an audiobook narrator on how to perform that character. For a task like that, you want a really low hallucination rate, and Grok tends to be the best. Whereas for tools that help you create a pitch or write compelling sales copy, Claude is currently the best.
The Best Models for Writers and How to Use Them
Thomas: For people who are prompting on their own, I recommend either Grok or Claude. Grok is
better at research-type tasks, and Claude is better at writing-type tasks.
My one principle that will totally change how you interact with AI can be summarized in three words: paragraphs, not sentences. Most people use AI the way they use Google. They type a sentence into the box, do not get what they want, type another sentence, and have this long back- and-forth conversation. That is not the best way and you are never going to get good results that
way. Instead, write paragraphs. Introduce yourself, explain your goals, explain what you want. The more context you feed the AI, the better the result.
Many of the Patron Toolbox tools have prompts that are three, four, or five thousand words long with very detailed instructions. You can actually generate those kinds of prompts yourself. One hack is to tell the AI, "I want to do X. Help me create a prompt that will help me accomplish X." So you might say, "I want to fact-check my book for historical inaccuracies. Help me create a prompt to do that." It will then write its own prompt, a page long with paragraphs of information. You read it quickly, make sure it is asking all the right questions, fix any wrong assumptions, and feed it back into a new conversation. You will be blown away at how much better the response is.
Cara: That is a great tip, because I have found myself in those loops where I am saying, "No, you
are not understanding what I am asking."
Thomas: Right. Giving it the goal and then saying "create a prompt for me to accomplish this goal"
is just game-changing.
Switching from ChatGPT to Claude or Grok
Cara: I have been using ChatGPT for about a year in the paid version and I have been hearing how
much better Claude is at writing. Is it hard to switch, since ChatGPT feels like it has kind of learned all my things?
Thomas: Not if you know how to do it. I actually pay for all of them. Grok, ChatGPT, Gemini, and
Claude. I often multi-prompt because I want to always be up to speed on which is best for certain tasks.
Here is what you do if you want to move from one to another for the first time. You say to ChatGPT, "I want to start also using Claude. I want you to create a context transfer document for me to take over to Claude." It will crunch through everything and give you this big context transfer document. Then you go to Claude and say, "I am coming over from ChatGPT. Here is the context transfer document," and you paste it in. That is the beginner version.
The more advanced way is to use projects. You want to cluster similar activities inside of a project. Put all your research into a research project, separate from your writing brainstorming project. Create a project for each book that keeps that book sandboxed away from the others. Each project can have special knowledge and special instructions unique to it.
Using projects is really the three levels of using AI. The first is learning how to write a good prompt, which we already talked about. The second is learning how to use a project and project instructions. What is wonderful is that AI is not antagonistic. None of this is secret. You can just ask it, "How would I use a project for my book to get better results?" and it will tell you. The third level is skills. Learning how to create a Claude skill that goes along with your project instructions.
One reason I do not particularly recommend ChatGPT for authors is that because it is the most common model, people are the most tuned to notice all of its idiosyncrasies. You never really want to do any writing with ChatGPT because people will be able to spot the ChatGPT quirks right away.
Setting Up Projects and Using Skills
Cara: When you talk about projects and skills, are projects just how you are organizing your history
within the tool? And what are skills?
Thomas: The language changes a little depending on which one you are using. Grok, ChatGPT, and
Claude all use the word "project." Gemini uses the word "gem" for no reason I can determine other than to be difficult. Gemini is arguably the most underrated model. People overlook it. And Copilot, the Microsoft one, is the most overrated. People who use Copilot and think AI is no good often come to that conclusion because Copilot is no good, not because AI is no good. Siri and Copilot are not representative of what good AI can do.
If you open up ChatGPT, in the top left corner you will see projects. You can click "new project" and name it. Once you create the project, you can edit the project settings and add instructions. Think of these as the rules you want it to always follow for every conversation inside that project. You can get ChatGPT to help you write those instructions too. You can say, "My goal is to renovate my bathroom, my budget is this, I live in this city. Help me write project instructions to guide this process." Or better yet, "Ask me some questions that will help you write project instructions later." Claude is really good at asking you questions to guide this kind of process. You answer those questions and then say, "Okay, now write the project instructions." It will build out a big long list and you paste those into the project. Now every single-sentence prompt you type is combined with all of those instructions, and you suddenly get much better results.
A skill is an ability you add to your AI that you can use across projects. There is a marketplace of skills. A common one for Claude is the Stop Slop skill. It identifies all of the most common markers of AI writing. Not just things like "it's not just helpful, it's game-changing," but all the little sentence constructions and words that Claude particularly likes, like "delve." You install the Stop Slop skill
and it will pull on that skill any time it thinks it needs to for a certain task. Someone else has already created it and shared it. Most of them are free right now.
A skill is like a methodology or a set of rules applied to all your future work. It is almost like programming, but it is all in plain text. You can just open up the skill file and read it and tweak it.
How Writers Are Underutilizing AI
Cara: Are there one or two ways that writers are maybe underutilizing AI right now?
Thomas: Grok for research is tragically overlooked. Grok is so good at research, particularly for
nonfiction. Because it is not designed around any particular political truth-seeking framework, it is much less likely to have a political bias or to just straight up lie to you. It tends to have the lowest hallucination rates and it is the most willing to search the web.
Claude is very lazy about searching the web and grounding itself. You have to basically explicitly tell it to search or it will not, unless you push back and it suddenly says, "Oh, you are right. Let me go do a quick search." And then it finds out its knowledge was about a year out of date.
Grok, every time you prompt it, will ground itself on the current conversation on X and grab dozens, sometimes over 100 web pages and read them all, often within seconds, before generating an answer. So for anything relating to nonfiction topics, you can get instantly up to speed with the latest thinking on any subject. It has access to all of that in real time.
I use it for shopping research, too. My grandparents had this collectible chess set with hand-carved figurines. Someone in my family, I will not say who, dropped one of the knights 20 years ago and the head broke off and was lost. I tried for years to find a similar chess set and could not. My brother suggested I just take a picture and have Grok see if it could find it. I took a picture and Grok instantly recognized it. It said this is likely a handmade chess set sold to tourists in the late 80s or 90s. And that was exactly the case. It then searched all of eBay, all of Amazon, scoured the whole internet, and found that exact set. I was able to order a replacement at a reasonable price and redeem myself.
Grok is really good at that kind of research, and a lot of people just have not tried it. It can also access websites that other AIs cannot, which is why I particularly like it for shopping research.
AI and Theological Bias
Cara: Something you said makes me want to ask about bias. Do you have concerns about how AI
handles biblical content and theological questions?
Thomas: Yes, I am pretty discerning and I feel like ChatGPT is generally aligned with a certain
worldview. I have caught it refusing to serve up certain Bible passages on specific topics where it will serve up passages from the Quran but will not serve up passages from the Bible. It absolutely censors you. Claude does this too. This has been a problem for a long time and one I have worked on with Christian and conservative AI researchers.
The saving grace is basically Grok. Grok is a lot like Micaiah in the Old Testament. Ahab brings in all the prophets and they all tell him what he wants to hear. Then Jehoshaphat says to bring Micaiah, and Ahab says there is no prophet from the Lord I hate more than Micaiah. Micaiah comes and first says mockingly, "Oh yes, you are going to be successful." Ahab says, "Haven't I told you to always tell me the truth?" And Micaiah says, "Okay, you want the truth? Here it is." And then he gives the truth.
ChatGPT and Claude are like those 400 prophets who will just tell you the party line. Their source of truth is essentially whatever the New York Times says. The farther away from the New York Times you are, the more the AI is going to nudge you back toward that perspective. If you ask Grok, its initial answer will often be the same party line. But if you nudge it just a little bit, it will tell you the truth.
I created special project instructions called Micaiah Mode for Grok specifically. I posted it on Author Media's social network for free and anyone can copy and paste it. It knows the story of Micaiah and it will give you the mainstream answer and then the truth underneath the truth on any topic, using a kind of dialectic where the two perspectives debate each other.
I also worked with Alexander McCreece, an AI researcher who developed an AI trained on the classics of the Western canon. Aristotle, Plato, Thomas Aquinas, John Locke, all of the greats of Western civilization. In doing so, he inadvertently produced an AI that became quite Christian, because so much of the Western canon is rooted in Scripture, and the AI absorbed that. He gave me permission to use Cosmarch, which is what this AI is called, and I have a version of it running on Grok inside the Patron Toolbox. I took his prompting and nudged it a little bit more between Catholicism and Protestantism, adding a bit more C.S. Lewis and Martin Luther influence.
I found that I can use Cosmarch to generate really good Sunday school curriculum. I was teaching fifth graders at our church and the curriculum we had been using had become very watered down. So I used Cosmarch and in a single prompt, giving it the Scripture passage we were supposed to
teach and asking it to generate discussion questions, it created a better classroom discussion than what the published curriculum had provided. We had an amazing Sunday school that day. The kids were asking great questions and we had this wonderful dialogue.
Cosmarch has no problem quoting the Bible. You do tell it what translation you want, but it will just quote it. ChatGPT, very recently, will often just shut down entirely, even on benign non- controversial passages. Someone on my team experienced this just the other day. It gave a guardrail denial and said it was not tuned to do that.
If you want to do this with Grok on your own without the Patron Toolbox, just create a project in Grok and in the instructions explain your denomination, your theological perspective, and the translation of the Bible you want. Very simple grounding instructions like that make a huge difference.
You can also upload your previous writings into a Grok project. If you are writing a sequel or nonfiction and you have a previous book, you can upload that whole book and it will reference it for all kinds of things and ground itself on your previous writing, helping you accomplish your goals without throwing up theological barriers.
Ethics and the Line Between Helpful and Harmful
Cara: Do you have something that addresses the different ethical dilemmas that authors face while
using AI?
Thomas: The problem is that no one can define AI consistently, so every publication and every
publisher has their own definition of what AI is. One basic principle of ethics is to abide by the contracts you sign, or do not sign a contract you cannot abide by. You need to understand what each publication's guidelines allow. Almost all publications are okay with using Microsoft Word's built-in spell check. Almost every publication is against a sentence-to-article prompting of ChatGPT, because if AI can do that, why would they need you?
Setting ethics aside, I would caution you against the things that are allowed but are still a bad idea. And at the top of that list I would put Grammarly and ProWritingAid. I think those are awful. They revert you to the median and polish off your voice. They communicate preferences in the language of absolutes, right and wrong. When I was learning to write as a teenager, I had Microsoft Word. When I misspelled a word, I got a red squiggly and that meant it was wrong. Grammarly and ProWritingAid both speak through that same little red squiggly about things that are not actually wrong. They are preferences.
It is so difficult to resist that design vocabulary, especially now that they are clustering multiple suggestions into one change. They may fix a spelling issue and a comma splice and rework your sentence in a way that is more generic and bland all at once. It makes your writing more neutral, less distinctive, and more boring. And it is far too easy to just click "accept all changes" and have your voice vanish.
I have a copy editor and a developmental editor in the Patron Toolbox. One thing I like about those tools is that you get a separate document and you, the human, have to take action to implement the changes. It is way less convenient. It is a lot of work to go through a two or three page report pointing out developmental issues and then decide using your own judgment which suggestions to take. But the fact that you have to take action means that if you do nothing, your voice is preserved. Just because something is lawful does not mean it is beneficial.
Cara: So for example, I am editing my first novel right now. I use AI to flag things like speech tags
and showing versus telling, and then I go line by line and make those corrections myself. I very specifically say I do not want you to rewrite it. But sometimes I wonder, did I cheat?
Thomas: There is an argument, often framed in very moralistic language, that AI makes you stupid.
But this argument is actually very old. The oldest version of it was made by Socrates, who said reading makes you stupid. And he was not entirely wrong. At the time of Socrates, young men were memorizing the entire Iliad and the entire Odyssey. The poetic form acted as a kind of checksum to ensure correct memorization, because if you got a word wrong the meter broke. He said if we move our memories to these external marks, the stories will not dwell inside us. They will dwell outside us and we will lose the enriching force of having memorized those great works. And he was not wrong.
Digging with a shovel builds more strength than digging with a backhoe. Digging with your hands builds more strength than digging with a shovel. But if you need a trench dug, you use a backhoe. There is a time to use a tool and a time not to use a tool, and it requires judgment. Adopting the new thing does not necessarily mean forsaking the old thing.
There is something to be said about writing with a bit crystal. You can get a box of Bic Crystals at your local grocery store for $5. It is the best-selling consumer product ever made in the history of man. No object has ever been mass-produced in such quantities. The design has been unchanged since the 1950s. It is a masterpiece. You and a bit crystal and a legal pad might do more for your writing than you and AI in a word processor. And sometimes using both is the answer.
Dictation as an Ancient and Modern Writing Tool
Thomas: Another use of AI that I have not mentioned yet is dictation, which is actually a more
ancient way of writing. Paul dictated his letters. They were written down in shorthand and the amanuensis would rewrite them in long form. This is not some weird conspiracy theory. It is right there in the text. In the book of Romans, his amanuensis says, "I, Tertius, who write this down, also send you greetings." Paul also got so worked up in Galatians that he pulled the pen out of Tertius' hands and wrote, "I, Paul, write this in my own hand." Some Bibles render that sentence in a different font.
That act of dictating is now really easy with AI. You can get a $5 app called Just Press Record that runs on your iPhone, Apple Watch, or Mac. It will capture with perfect accuracy what you say. I have a tool in the Patron Toolbox that will then convert your recording and break paragraphs out, put quotation marks in all the right places if you are writing dialogue, and fix any homophones. Other than that, it makes no changes. So it is your raw first draft in a slightly more readable form. I really wanted to call it the Amanuensis, but I knew no one would know what that meant. So I call it the Chapterizer instead. I almost named it Tertius, but that was too deep a cut.
Brainstorming and Taking Responsibility for AI Output
Cara: I think we can also learn from AI as a learning tool for ourselves. When I first started writing
my book, I did not know the objective and subjective use of "thee" and "thou," and I do now.
Thomas: One important thing when it comes to using AI is always taking responsibility for any of
the outputs. As soon as you start blaming the AI for something, you are doing it wrong. It is okay to take responsibility for the work of someone who worked on your behalf.
It says in Chronicles that Solomon built the temple. Did Solomon build the temple? He paid for it, provided the designs, and orchestrated the whole thing. But was he there picking up bricks from the quarry and placing them and doing the carving? No. But he did build the temple because he oversaw it and took responsibility. You have to be willing to take the blame for every word of output. Because he who gets the credit gets the blame. As soon as you are tempted to say, "I was just listening to ChatGPT," no. You take responsibility. Whatever comes out is 100% on you.
Cara: I love using AI for brainstorming. I was using it today to figure out which software is best for
a certain task. What is your take on that?
Thomas: What is more important than the model is the prompt. A good prompt will give you so
much better brainstorming than a good model will. My top two for brainstorming would be Grok
and Claude, depending on what kind of brainstorming you are doing, but the prompt matters far more.
In your prompt you want to force the AI to brainstorm from multiple different directions. Use the super-prompting technique I talked about. Tell it your goal and ask it to create a detailed prompt that hits the problem from a bunch of different angles. You can also create personas that will each brainstorm separately.
If you are not careful, you get the same problem in AI brainstorming that you get in human brainstorming, where everyone riffs off the first idea. Brainstorming as an activity in corporate America is not actually very effective. What is much more effective is to first have everyone create a list of 10 ideas on their own before they hear any ideas from anybody else. Then they come together and share. If you skip that first step, you end up with the loudest people's opinions rather than the smartest people's opinions. That quiet, shy thinker often gets blasted out by the chatterbox.
You get the same issue with AI. If you just say "brainstorm," it will give you its most obvious, chatterbox-style answer. But even using the Micaiah Mode project instructions I mentioned, if you say "brainstorm," it will give you the mainstream brainstorming and then the Micaiah Mode brainstorming from a totally different perspective. You may find that second angle is way more interesting.
Better prompts require that you do a lot of thinking before you come to the AI and ask it to brainstorm for you. You need to have clarity on your goals.
A good example of a well-written prompt is actually the book of First Timothy. You can look at that whole book as an LLM prompt. First, who I am: I, Paul, an apostle of Jesus Christ by the will of God. Here is who I am and here is my authority. Then who you are: Timothy, my beloved son in the faith. Identity, then identity, then instruction. The same structure works with AI. Here is who I am. Here is what I want. Here is who you are. You are a fact-checker expert in 18th century romantic exchanges among aristocrats. And then the instructions follow. I want you to make sure all the courtly manners are appropriate, all the historical details are accurate. That kind of structure will get you much better results even if you are still using ChatGPT.
Where to Find Thomas and His Tools
Cara: Well, thank you so much, Thomas. I know your website is ThomasUmstattd.com, although
the better website is probably AuthorMedia.com.
Thomas: AuthorMedia.com is where you find the Patron Toolbox and all three of my podcasts. That
is the right place to go.
Cara: Thank you so much. This has been a real education. We will need to talk about this more on
The Block and share our findings.
Thomas: Thank you for having me.
Generated by the Patron Toolbox for Novel Marketing patrons · authormedia.com/patron-toolbox
