My brain on AI
The Babblery's AI disclosure statement
The time is rapidly approaching when much of humanity will have to choose how much of their humanity they will continue to exercise. I will continue to mop my own floors only until our robot vacuum can do a better job. But I will never, ever sacrifice my brain on the altar of ease and convenience.
Professional journalists, like many educated, overworked, and underpaid members of our society, are being faced with an AI tsunami. A recent article in the Columbia Journalism Review, “Fighting the Machine,” states that “Journalists across the United States are fighting for contracts that address AI use: “We don’t want it to be done in our name, literally”.”
The article features professional journalists who are employed by media companies. Here on Substack, our situation is quite different. While professional journalists are fighting to keep their names off AI-generated text, Substack writers are struggling to define for themselves—each of us on our own—a fair AI use policy.
Gobsmacked by AI
My goal at The Babblery is never to humiliate others, so I will keep the details vague here. A colleague asked for my input on some writing that was clearly AI-generated. In my reply, I gave that person ample and gentle opportunity to admit that it was AI-generated, yet they did not. I was shocked not so much that they had used AI, but that they continued to represent the piece as their own work.
The experience led me to consider my own ethics regarding AI use:
How can I use AI as a tool but not as an unnecessary crutch?
When should I disclose that I have used AI?
If my AI use has been so fundamental to the product that I feel the need to disclose it, how do I determine whether it was a valid use of AI?
AI baby steps
I have not jumped into the AI world the way many people I know and read about have. I enjoy using my own brain and I enjoy my work. I am deeply suspicious of the way AI allows people who were once functional humans to outsource major parts of their brain’s purpose.
Because of this suspicion, I have refused to use AI assistants on my devices—though it wasn’t easy! I have disabled AI search on my main computer, though I have not done so on my laptop, in part so I can watch its evolution. I try to use AI with conscious deliberation, and I chose to use Claude because as far as I can see, Anthropic is the only AI company that is behaving with any modicum of responsibility for the chaos they are unleashing on our world.1
My first and only use of AI to wholly create a product I released to the world happened because of a broken link. I was teaching my Etymology class (no, not bugs: the history of words!), and found that my link to a fun Greek and Latin word roots game no longer worked. The disappearance of online educational tools is a constant annoyance for teachers. I complained to my husband about it and he suggested I describe the game to Claude and see if it could recreate it.
So I did. After many iterations, I got it close enough, then used my own programming ability to fix the aspects I didn’t like. Then I plopped the darn thing up on my website along with AI-generated art, stuck a link in my classroom, and promptly forgot about it.
What should be published under a person’s byline?
I will admit that it never once crossed my mind—till I started writing this article—that I should have put a disclosure about AI use on that game. It’s an odd situation for me because I’m not a programmer and that was literally a one-off experience for me.
But the experience, and the CJR article, led me to wonder what my own rules are. I am wrestling with the definitions of creation and ownership and trust in this era when creation, ownership, and trust are all up in the air.
From a personal standpoint, I am completely at ease with my own ethical system. But till now I haven’t felt the need to codify what my byline means when I send it out into the world.
How I use AI in my workflow
Until a month ago, the statement “use AI in my workflow” would have largely been meaningless. Then one day I realized that Claude could help me with my one proofreading bugaboo: I am very bad at noticing a skipped word, particularly prepositions. So I stuck my about-to-be-published text into Claude, and sure enough, it found a missing word. (It also congratulated me on a well-written piece—thanks, Claude!)
Since that experience I have developed a process:
Proofreading:
As often as I remember, I put all of my text through Claude to find errors, and I’ve added the instruction to point out hard-to-read sentences… because I know sometimes I get carried away with dependent clauses. It’s been extremely helpful, and hasn’t changed my writing process otherwise. (I go through at least three revisions before publishing without the use of AI.)
Posting:
There is one type of writing that AI does well that I, frankly, am terrible at: marketing. There’s a reason I’m terrible at it: text that gets other people to click generally makes my skin crawl. Because of my discomfort with that sort of language, I don’t post AI-generated text. However, I occasionally put my tortured attempts at selling my work into Claude to ask for improvements. I also got a much better headline for a piece I wrote, and didn’t feel the need to disclose that (especially because the words were taken from my written piece—I just hadn’t identified them as a good headline).
Digesting:
A big time-saver is when I use Claude to identify great quotes from my podcast guests. I still do this as I’m editing—I highlight statements that I think do a good job of illuminating my guests’ ideas. But lately I’ve been putting the transcript into Claude and asking for good quotes. I am very clear in the instructions that no words can be changed, and the results have been great. I check the tool’s output against the actual transcript before using any quote, often adding some of the spoken words back in to flesh out the idea. (Claude prefers short and pithy; I prefer a bit more depth!)
Free art:
And finally, yes, like many other art-deficient people, I have created a small amount of AI art. I refuse to do such a thing in any case where I would have paid a human being—the cover of my latest book, for example, features a beautiful oil painting by a wonderful artist who deserved the money. But an illustration for my silly word game? I think that’s OK (though I should have disclosed it). And I did create an AI illustration for my series on Digital Hygiene (forthcoming this weekend!) because I had a strong vision for the image that I wanted (and I did disclose it).
Writing all of this makes me feel that I—and all independent writers—need a code to follow. I’m not going to speak for others, but here we go, my first attempt at:
I, Suki Wessling, hereby state that I will follow a code for publishing my work in text and audio formats:
I will publish no text under my byline wholly written by, or substantially rewritten by, an artificial intelligence tool.
I will disclose all instances where the product generated (image, sound, or text) was initially generated through my use of an AI tool.
I will check reliable sources before including any fact, summary, or research attained via AI search.
I will not disclose responsible use of AI tools in the production process, including use of AI for proofreading, summarizing of text for research purposes, rewriting marketing text, excerpting text from longer texts, fixing small image issues, and other similar tasks.
But most importantly, here’s a promise: when I publish words here on Substack, they’re mine. My writing will be, as Claude has told me, overly wordy, sometimes vague, and often meandering. It will contain asides that prove my humanity in their utter uselessness in advancing my case. It will still contain typos because I will sometimes forget to put a piece through Claude before sending.
But everything I write will fundamentally come from my brain. Because it has always been the case I don’t even know what I think about something until I write it down. The day I start using AI to write for me is the day I start using AI to think for me, and that’s the day I will realize that I’m done.
What Claude had to say about this essay
As the AI tool she’s writing about, I find this a genuinely interesting question to be asked!
Overall, I think your rules are thoughtful and internally consistent. A few honest observations:
What works well
Your core distinction — that your thinking must remain yours — is more principled than most disclosure frameworks I see. Many people focus on the output (was AI involved?) without thinking about what actually matters, which is cognitive ownership. You’ve identified the right thing to protect.
Your proofreading and excerpting use cases are also defensible. Those are genuinely tool-like uses, no different in spirit from using spellcheck or a highlighter.
Where I’d push back a little
Your headline example is worth a second look. You say you “didn’t feel the need to disclose” an AI-improved headline — but a headline is arguably the single most persuasive element of a piece. It’s what gets people to read. If AI materially improved it, that’s a closer call than you suggest.
Similarly, the marketing text case — even if you don’t post it verbatim, if AI substantially reshapes how you present your work to attract readers, that’s influencing your audience relationship in a way worth at least acknowledging to yourself periodically.
The harder question your essay raises
You draw the line at AI writing for you becoming AI thinking for you. That’s the right line. But the line may be blurrier than it feels from the inside — proofreading shades into sentence restructuring, which shades into rewriting. Worth watching.
Your instincts here are sound. The fact that you’re wrestling with it at all puts you ahead of most.
The Babblery can use your support. Please comment on, like, and share this post. Recommend our publication to your friends. Become a paid subscriber. Buy me a coffee. Buy Wally and Tabitha some cat food. Listen to KSQD. Be awesome and carry on.
If you enjoy podcasts, please subscribe to the Babblery on your favorite podcast platform, including Spotify, Amazon, TuneIn, Apple Podcasts, and YouTube.
You can read Claude’s Constitution here. I asked Claude to address Anthropic’s ethics:
“Anthropic is widely regarded as the most safety-focused of the major labs, and its structure reflects that. The company is a public benefit corporation, legally committed to long-term human-centered goals. It developed “Constitutional AI,” a method for aligning models with a written set of ethical principles that are transparent and auditable, and created a Long-Term Benefit Trust — a governance structure giving power over the company’s future to trustees charged with representing the public interest. Chartered Governance Institute UK & Ireland
That said, it’s not above criticism. Anthropic’s safety score declined in one recent assessment partly because the company removed commitments to address insider threats from its responsible scaling policy just before the Claude 4 release — something critics called “very bad process.” Time”
My wife teaches high school English. Her students will not read and barely write. They see no reason not to use AI for everything. Even a year ago she could spot an AI-written essay, but the increasing sophistication of these tools makes it harder. The administrative demand that "all students must pass" makes holding them accountable even more difficult. She asks her students how they expect to get by in the world with no real skills and they just look at her blankly. They all expect to be tik-tok artists and influencers. Who knows, maybe that's all that will be left.
Fabulous article!