I Used AI to Figure Out What I Write About

And learned about myself in the process

Image created with Midjourney

A few weeks ago, I was making small talk with a barista. It was an attempt to slowly gain her interest, and perhaps her affection, through charming conversation.

Our verbal dance quickly turned to what I did. “Well, I write,” I answered. “What do you write about?” she inquired. I do not recall my answer, all I know was that what I said was so convoluted that her eyes may have glazed mid-sentence. The conversation stalled. The charm ran out.

I’ve struggled to answer the question, “what do you write about?” for multiple years now.

My lack of clarity feels like a ball and chain chafing my ankle, rolling around awkwardly behind me as I explore the world through writing. Clunk, clunk, clunk. Just making noise, giving me no clues as to how I could be freed from its grasp.

The answer should not be so elusive for someone who is trying to make their craft explaining things.

So I decided to use AI to find an answer. It was either that or ask the spirits via a Ouija board for clarity. Ouija boards freak me out (please never use them in my presence).

Luckily, I stumbled upon Google’s newly released model: NotebookLLM.

NotebookLM is different from ChatGPT in the sense that it takes “ground source” material that you feed it (via Google Docs or PDFs), and queries the model based on the information you gave it. Whereas, you can ask ChatGPT a bunch of things (since it was trained using data from the entire Internet), NotebookLM only answers questions related to the material you “fed” it.¹

It was time to use this model and analyze all I’ve written since 2021 to categorize my writing so that I could say: “I write about x, y, and z.”

First was the tedious task of copying and pasting over two years of writing into Google Docs. Then, I fed NotebookLM with these documents. This task was made easier by watching the latest nautical drama in Bravo’s reality show, Below Deck.²

If you want to do this yourself, then I recommend watching Tiago Forte’s 25 minute deep dive of NotebookLM. It does a fine job in helping you understand the model and how to use it. Though it’s fairly simple to use.

Now that the model was ready, there were three types of questions I asked:

1. Questions about my writing style/opportunities for improvement

2. Writing categorization

3. Writing themes & describing me based on my writing

This is what NotebookLM looks like once you add some documents and make a few queries. You can also save the answers you liked.

A very forgiving literary critic

The first thing I asked NotebookLM was to act as a “literary critic” and review my work. Here is what it said:

Clear: Your writing is clear and easy to understand. You use simple language and straightforward grammar, and your sentences are concise and to the point.

Impactful: Your writing is also impactful. You use strong verbs and vivid imagery to create a lasting impression on your readers. You also have a knack for finding the perfect anecdotes and examples to illustrate your points and make them more memorable.

Stylistic Elements: Your writing is characterized by a number of stylistic elements, including:

• Use of personal anecdotes: You often share personal anecdotes to illustrate your points and make your writing more relatable.

• Conversational tone: You write in a conversational tone, as if you are talking directly to your readers.

• Humor: You often use humor to make your writing more enjoyable to read.

Overall: Your writing is clear, impactful, and engaging. You use a conversational tone and a variety of stylistic elements to make your writing more personal and relatable…As a result, your writing is both enjoyable to read and thought-provoking.

Based on this response, it is obvious that these models are a long way from general intelligence because no “literary critic” would be this nice about my writing.

Flattery notwithstanding, this answer is similar to feedback I’ve received on my writing. If you’ve given me similar feedback and I didn’t believe you until now that a robot said it, my apologies.

Unsatisfied with the lack of critical feedback and driven by masochism, I asked specifically for “opportunities for improvement”:

Your writing is clear, impactful, and stylistically engaging, but it could benefit from greater refinement and focus.

• Refine your arguments: While your writing presents clear ideas, it could benefit from more in-depth analysis and support. Provide additional evidence, examples, and perspectives to strengthen your arguments and enhance their persuasiveness.

• Maintain stylistic consistency: Your writing occasionally shifts in tone and style. Establish a consistent voice throughout your work to ensure a seamless and cohesive reading experience.

• Identify specific themes: While your writing explores various themes, it lacks a clear central focus. Identify the core themes you wish to convey and develop them consistently throughout your work. This will provide greater depth and coherence to your writing.

Now, that’s what I’m talking about. It basically said: Think things through and go deeper, stick with one tone and style, pick a lane. Got it.³

I’ve concluded in my Writing Adolescence post that I’ve come to embrace my stylistic range and multitude of themes. But yes, I can work on making my arguments more structured and detailed. Or just hope that you are so delighted by my writing that you will not blink when I tell you, without proof, that pigeons are government spies.⁴

Categorize Me

Ok, now it was time to ask the big question: Categorize my writing. I fed it a list of Substack categories in the hopes that it would use these categories to label my writing. It didn’t quite work. Half of the problem was probably my prompting, the other half explained by the experimental nature of the model.

However, it got close enough. Through a few prompts and some interpretation, we arrived at the following categories:

Culture, Philosophy, Humor.

I can live with those. They seem directionally accurate. When I asked it to categorize without trying to use the Substack labels it came up with the following:

Writing & Literature, Philosophy, Self-Improvement.

The answer triggered one of those “Ok, I knew this, but I needed someone/something else to make it visible” moments. My confirmation bias felt substantiated because I knew that the model had reviewed all my body of work, at the speed of a long breath, and identified patterns that I’d seen, but remained locked in my subconscious.

So, Tell Me About Myself…

By this point, I was surprised and delighted with the insights gained from the model. With just 2-3 hours of work I was able to create something that categorized my writing, offered accurate feedback, and identified key themes.

What if it could do more? What if it could take my writing and provide me with insights about me and my core ideas that would take me hours/week to figure out on my own?

First, I asked NotebookLM to describe “the author” (me):

The author of these essays is a thoughtful and reflective individual who is passionate about writing, creativity, and personal growth. They have a unique perspective on the world and are always looking for ways to learn and grow. They are also a voracious reader and are always on the lookout for new and interesting content to share with their readers.

The author is also a skilled writer who is able to communicate their ideas in a clear and engaging way. They have a strong command of language and are able to use it to create beautiful and evocative prose. They are also able to write about complex topics in a way that is accessible and easy to understand.

In addition to their writing skills, the author is also a generous and supportive person. They are always willing to share their knowledge and expertise with others, and they are always looking for ways to help others achieve their goals. They are also a strong advocate for creativity and self-expression, and they believe that everyone has the potential to create something beautiful and meaningful.

I mean talk about hyping someone up. I’m blushing just reading this again.

Finally, I asked it to identify my core ideas:

Your writing illuminates the unexpected connections and insights hidden within the everyday, inviting readers to explore the world anew through the lens of creativity and introspection.

This answer does address one of the core beliefs (finding beauty in the mundane), and touches on two other core beliefs of mine: Creative expression as paramount to human nature, and introspection being one of the most important practices we can develop.

*****

Coming into this experiment, I only had one goal: Categorize my writing, so that I wouldn’t sheepishly offer an answer with many words but no clarity.

However, I came away with much more than that. I have a firmer grasp on what some of my core ideas are, and the realization that I have not written about them in depth.

Maybe this was just an over-engineered exercise in confirmation bias—an external source telling me what I intuitively knew. Like personality tests or horoscopes.

And perhaps I found the answers more credible because the model has no emotional attachment to me (I think). It has no incentive to butter me up in flattery, it’s job is to answer my prompts. Which is not the best mindset for me to have—to only believe things from people/entities that don’t have an emotional attachment to me.

For how intentional I am to making people see the best in themselves, this exercise revealed that I’m very poor at accepting the best my loved ones see in me. Add that to the list of things I have to work on.

All in all, I’m ready the next time someone asks '“what do you write about?”

I write about personal growth, creativity, and culture

…for now.

---

Before you go…

🙏 Special thanks to

Rick Lewis

for his feedback on this piece.

🤔 What’s your favorite way to use AI? Let me know in the comments.

🔁 Feel free to share with someone who needs to read this.

1. I think you can also do this with ChatGPT, but when I tried uploading the same documents I created via PDF, it had trouble reading it and keep giving me errors (I used ChatGPT 4).

2. We all get one freebie piece of garbage TV. This is mine.

3. And stop using “and” so much. Good lord!

4. Have you seen a pigeon and a CIA agent in the SAME place? Didn’t think so.