What I can't make tidy about AI

Yesterday Substack launched AI detection in partnership with Pangram. You can now scan Notes, replies, comments and posts published from July 21 forward and get an estimate of how much of the text was written by hand versus with AI assistance. Substack also added a “How I make this” field, where writers can describe their process in their own words.

Consider this my version of that field but written in tribute to Substack via a long-form post that takes a few twists and turns.

Let’s start with establishing a few basics. My general work in this publication is to help writers build publishing skills in service of what they’re building on the platform. In my off-Substack life, my career has read “a little like a map of everything the media industry has been through in the last two decades.” I’ve had a curious front-row seat to AI since 2023, because my professional work shifted me through software spaces as the editorial and comms counterpart.

Today I want to share a few thoughts on AI from these vantage points. Because watching the machinery of artificial intelligence move through our experience as humans and the software world, specifically, has made me want to vomit more times than I can count — queasy because it is deeply uncertain and impossible to pin down an easy stance.

I think there are lessons and deep complexities to consider for anyone who, like me, has related to language as both creative lifeline and livelihood.

I think it’s worthwhile to start by noting that AI has been gobbling software production norms at an astonishing rate. I watched a CRM be built from the ground up circa 2015. It was a production event that lasted more than a year with a full remote team.

In comparison, fast forward to today and let’s briefly consider the launch of a tool I built, (an admittedly much smaller product) SubSight — a data-driven supplement to your Substack dashboard and publishing advisor tool. Even just three years ago, I probably would’ve had to hire a team of three developers, at part-time hours, just to build that tool. Instead, I used Lovable (powered by AI) to build this tool on my own, using my own writing, experiments and Substack data audit patterns, within just a few weeks.

Every new shift and development in software often pushes me to confront a stark reality of human progress (that it is imperfect) and to also consider what is coming for editorial and audience work (it is being gobbled and I don’t know how we stop that).

Three years in, I also see how useful AI is inside my small business and I still get a queasy stomach when I think about its impact on the job market writ large. I worry for humankind as we know it, and I worry that greedy organizations won’t improvise their own production processes quickly enough to re-tool and retain their workforce.

Three years in, I worry that original writing and literature as we know it will be lost. That the art form my great-grandfather, grandmother and myself all devoted our lives to will become another space that’s industrialized and therefore barriered. I understand that all great inventions create barriers in society. And I do not want a world in which original, human art is something that only the ultra wealthy get access to while everybody else gets the robot churn.

Three years in, I do not have tidy answers about AI, but I would like to look at its impact on Substack production norms and creative editorial processes. I’d also invite you to indulge me in considering how AI is asking us to confront consciousness and inner belonging itself.

If I had drafted and edited this piece alone, as a one-woman show, it would have reached you next week, after the conversation had already moved on.

But because AI works as my production assistant, it reached you today instead within a news window that will hopefully continue to help my publication rank in AI summaries (which, by the way, if you’re a small business owner, you need to be watching for how SEO norms are being upended).

Using AI inside my production process has taken some time to polish. As someone who has worked in traditional publishing and media roles for nearly 20 years, I fondly remember walking into the newsroom to a sea of turquoise, blob-shaped iMacs. This was the super cool tech of the day: Adobe InDesign and InCopy software (that talked to each other!).

Colorful all-in-one computers (and the software installed on them) aside, everything else about getting the news out the door was done by humans, piece by piece.

But things are different now.

AI is making things different.

I am different.

My access to and understanding of my brain is different, and I have had to make peace with the fact that some parts of the publishing process were destined to evolve — I just had no idea how quickly and in what manner.

When someone asks, “Is it wrong that AI helped me write this?” I’ll be straightforward with you and say that I don’t try to answer that question at all. Instead I sink into the production norms I rely on and my own values around what I promise readers of my work.

When it’s time to draft a new piece for my publication, I instead ask: is getting this post quickly out the door in service of my work and my audience? Or does it merit the attention of a more manual process, and therefore letting the writing be delivered at a later time?

For me, this is a daily values scan I run. Sometimes AI production efficiency wins, especially for transcriptions, live stream summaries or short announcements where my cognitive reserves are better spent in service of creative language work than the rote production of a transcription summary. Other times, still, especially for work around poetry and personal storytelling, my stomach votes quickly with a quiet churn. I won’t risk my “fresh thought” space for poetry because I decided to let a robot push me out the door too quickly. More on that a few paragraphs down.

In the early years of my career, I worked as an editor, a copywriter, a ghostwriter. I have lost count of the times I handed a writer, in the margins of their essay, an original idea or conclusion that was truly mine — but was also deeply intertwined with their original work.

Should my name have gone on that paper beneath theirs because I contributed a few refining positions?

Should I have even offered that as an editor?

Was it ethical for the author to use my original ideas because I was, after all, paid an editor’s wage to expand and broaden their thinking on the page? Did the wage function as consent, and consent for what, exactly?

These kinds of questions tend to get twisted up on their own, but they are what I hear every time someone tries to vilify a writer for using AI.

If we are worried about source attribution, then what about every author I read and studied in college to become an excellent editor — whose ideas and cadences alchemized into how I edit on the page naturally?

How do I own my own thoughts so thoroughly that I could no longer point to the seam?

Who do I owe? What would I even pay to Betty Friedan, whose work (among countless other giants) changed the way I understand female liberation and informs how I sniff for gender-based constraints on the page inside authors?

I don’t ask these as questions for you to argue with me in the comments, though feel free to do so respectfully.

I ask because I’ve never had a straightforward, satisfying answer that will let me put my editor’s pen down and place my head in the sand. Ultimately, I don’t think AI created an attribution problem. I think it made a deeper problem impossible to keep ignoring.

I hinted at this above, but it deserves its own point in this essay.

Right now I’m in the middle of designing a workshop. Something like Audience Nurturing Bootcamp for Substack Publishers. It’s built from the foundation of all my editorial and audience work, and the application/design process is massively resource-consuming, partly because I’m working with a lot of my own original data sources inside the Substack space. I’m building the plane in flight and trying to create tools people will find useful.

So let’s say I design a series that is mine — the architecture, the sequence, the argument, the reason it exists.

It’s not enough to just design the series and then people show up.

First I have to market it, which is fine in premise but massively expensive cognitively to shift toward. Which insights am I allowed to use? Is AI fair to recruit in this scenario?

It’s been 10 years since I was working inside a marketing agency on client accounts and marketing strategies. I did that work, and it’s alchemized into other viewpoints (mainly publication design and audience stuff on Substack).

There are fundamental marketing best practices that I know, but do not have expedient access to inside my daily business operations. And if I put the “no AI” purity test inside this production process, some purists might say that I’m untrustworthy if I use AI to write a first draft of a marketing sequence in service of my originally designed series.

The double, triple and quadruple binds galore.

Yet, still, I am sensitive to publishing anything in The Publishing Spectrum where I assert myself as an authority on a topic in which I have only read a few books and had no real world application. I feel uncertain about asking AI to take a marketing book I devoured myself 13 years ago and applying that book’s insights into my own marketing prompts. I also don’t have fresh access to every layer of marketing strategy at any given time.

Do you see what I’m trying to illustrate here? The purity test of AI use wreaks a lot of havoc, and if I weren’t a newly single mom building an advisory practice, and were, instead, independently wealthy, I’d just opt out of AI. But I don’t have any of those things. I have my thing, my life, my work. And I have to approach all of it with what’s in front of me and accessible to me, within reason.1

Despite all the inner consternation around AI, I am quietly optimistic about its potential to invite us to greater depths of mindfulness, inner belonging and creative output. Not because I think anything will ever replace the need for human connection, but because AI has — with the right inputs — the powerful ability to create mirrors for humans to see the self more clearly.

I often notice that AI is mimicking something we sometimes encounter in small supply in life: the seeingness of self that occurs when humans witness one another. AI cannot move energy between two hearts. But it can, with the right data inputs, take your work and hand it back to you more clearly.

This is the work happening right now inside Substack’s new MCP connector, and it’s the same work that’s been happening inside SubSight.

Writing online in the pre-Substack, pre-data age used to require a writer to shift out of that inner seeingness and into marketer, producer, strategist, networker (ugh). With AI tools, those roles are immensely more manageable, and they hand the writer back to her creative work more quickly when managed well.

What’s important to notice here, in the context of publishing on Substack, is that publishing and editorial judgment systems are available to you now — and you don’t have to abandon your own writing process for 18 to 24 months just to get up to snuff on the basics.

So here’s where I’ll land this: it’s not that all AI is bad, and it’s not that all AI is good, and it’s not that AI belongs in every process you have with your work. It’s about where it can actually help you stay in the arena week after week. If AI — and dare I say data — helps a writer work through the uncertainty of showing up with her voice week over week, I count that as a good for humanity.

Which brings me back to yesterday’s announcement.

You may not think AI is a good thing, and you may be glad to catch anyone who uses it. The challenge is that Substack names the inherent limitation of this kind of scanning tool themselves: it cannot see the care that went into the work, or whether AI was used somewhere upstream as a source rather than as a writer.

So what is true attribution in the age of AI? What is true attribution in the digestion of human thought?

We’re all mid-experiment in this together. AI is already touching every facet of this work — how we publish, how we understand ourselves as writers, how we access insights about our own writing. The jury is still out on what impact it will have on us over the next 10 to 20 years. The best thing I know how to do is acknowledge a real need and set a real value boundary.

Walk lightly, keep writing and carry a big stick.