A Note to Readers About How I Use AI

Substack has recently unrolled a partnership with Pangram, an AI detection software that is supposed to help readers and writers discern who is using AI to write, and who is not. The idea is to improve transparency and give readers a choice about whether they read AI-produced writing or solely human-produced writing.

It’s a good idea in theory. In practice, however, all AI-detection software is notably bad at accurately predicting the presence and percentage of AI usage. There is already evidence with Pangram specifically that it is biased against writers who are neurodivergent, writers of color, and writers who speak English as a second language. On Substack this morning, there are writers of all kinds sharing their experiments with their own essays that they wrote entirely themselves, only to discover that Pangram has marked it as 100% AI written.

So, I’m going to take a minute to talk to you about how I use AI. Then, from here on out, I’m going to self-disclose before every essay. Since it’s not likely that Pangram is going to get more accurate any time soon, I’d rather tell you how I use it (and open a discussion if you have questions) then have you guessing or making assumptions based on a broken AI detector.

Here’s how I use AI: I draft at least 70% of the article before I ever open a chat window. I’m happier if I get to 90% but I admit that AI is most helpful when I get stuck. A couple of years ago when I was first teaching myself how to work with AI as a writer, I learned the hard way that there was a limit to how much I could let it draft for me. My creative muscle started to atrophy if I leaned on it too heavily, so I pulled back and created some rules for myself. This is one of them. I don’t let AI come up with the ideas or write the rough draft of the essay, and by the time I bring a draft to it, I’ve got the direction and the majority of it down.

I use Claude more than any other AI. The prompts I use with Claude are pretty specific and repetitive. I ask Claude for help in understanding what holes I have to fill, what I’m missing, or where I make logical leaps that don’t work. Often, I start an essay with the intention to braid topics together, only to discover that I’m actually writing two essays. Since this is a common habit for me, this is also part of my prompt. Sometimes I ask it to test my theory or point out examples that prove me wrong.

Then I take the suggestions I get from Claude and continue writing the draft, adding, shaping, or editing whatever it is I need to do. Sometimes Claude points me in a direction that requires some research, so I need to put the writing aside and do some reading/digging online.

Claude and I don’t always agree. Sometimes it takes me in a weird direction I don’t want to go. Sometimes Claude’s ideas are pure shit. Even that feedback clarifies and sharpens for me what I really think. In that sense, Claude acts the way an editor would act, pushing me either to strengthen my stance or strengthen my evidence for the stance.

I rarely let Claude write for me. However, aside from the weak spots as a writer I mention above, I also struggle with transitions between sections and ideas and occasionally I need help with a conclusion. That’s where I let Claude suggest a sentence for transitions or a paragraph for the conclusion. Then I take it and make it my own, usually rewriting the text in a style that’s much closer to mine or that drives home the angle I want to drive home.

I’ve had a love-hate relationship with AI. When it first came out, I was thrilled. I’m a futurist in some sense, and I could see the opportunities this created. I could see how it would one day shape writing and how we put together books, even though it was crap in the beginning (and still has major, major problems).

However, ChatGPT was probably three months old when I had a deep understanding that this would one day rock the writing and publishing world (and it is now—the writing and publishing is going through a prolonged dark night of the soul that will continue for many years). I’m a writer, editor, and book coach, as well as a healer. I used to ghostwrite. Last year was the first year that I looked at my writing business and knew without a doubt that AI was affecting it negatively. Last year alone, I turned down $30K+ in editing business because the authors brought poorly written AI manuscripts that were essentially uneditable, in my opinion. AI is one of several reasons I took a full-time job outside of the writing industry so that I could have a financial port in the storm.

I’m also aware of, and think often about, the environmental impact of AI. I don’t have answers for that, as I don’t have answers for all of the other dark parts of our world that keep me up at night. Along with my belief that AI is not going anywhere comes a faith that humans will figure out how to use AI in a more responsible way. We have to — there is no other choice. The genie is not going back in the bottle; AI is not going away no matter how people protest or abstain from using it. That said, I try to use it sparingly and if I find myself pulling open a chat window, I stop to ask why I’m doing it and if it’s really needed. I ask myself if it would be better for me to seek the answer now, puzzle it out by sitting with it longer, or seek other sources.

I think every writer has to determine for themselves what writing means to them, and where AI does or doesn’t come into their writing. Writing for me is a way to think, emote, transmute, sort through messy life experiences, and to create, as well as to connect. I never want to give that function to AI. Without AI, though, my partially complete essays might languish unfinished for weeks, or months… or longer. AI helps me see my blind spots as a thinker and a writer, and as a teacher. It’s also pushed me in beneficial directions I may not have pursued. But this came after a lot of trial and error, including times in the early days where I gave over way too much of the work to AI. This is not something I can tell you how to do for yourself; only you can determine what works for you.

As for reading the work of those who use AI, only you can determine what you think about that. My guess is that lots of publications will see a shift in their numbers. There are those who don’t care whether the publications they read are AI-written or not, but there are those who will feel some way about it.

As for the readers of this publication, maybe some of you will unsubscribe from this newsletter, and that’s okay. Maybe some of you feel disappointed; you might wonder how a spiritual writer can use AI. My answer to that is complicated and requires a lot more time to unfold (and an entirely different essay).

If you feel disappointed, or any which way, I sincerely ask you to share your thoughts in the comments below. I invite you to comment below if you have questions, concerns, or comments.

P.S. Pangram’s scan of this article says “fully-human written,” which is true.