Bertie Fielding, the dean of science at Stellenbosch University in South Africa, made a remarkable statement this week.
“I am a fraud,” he wrote in a commentary for Nature Africa.
He’s not really a fraud, but he says he felt that way after a two different AI detectors flagged his writing as 100% AI-generated.
ROUGH DRAFTS now has a paid subscription tier! Subscribers get access to a four-week course on human writing that I’m offering in September. Details are at the bottom of this newsletter.
Fielding doesn’t dispute the detectors. He uses AI extensively in his writing process, which he says starts with a “mess” of fragments, voice notes, and disconnected ideas. He then feeds those into an AI tool and asks it to generate a first draft, which he then works on, with the AI, through extensive revisions -- a process that he says takes hours, days, or even months -- “until I can recognize myself in it and until I am prepared to stand behind every claim.”
Calling AI his “ghostwriter,” Fielding calls for a more nuanced approach to disclosing AI use in writing. His argument: subject matter expertise and writing expertise are two separate things, and one can be an expert (a “thought leader”) and still not be very good at writing -- or vice versa.
In other words: You found my ghostwriter! Please don’t obliterate it with your proton pack.
As Fielding notes, the distinction between subject matter experts and writers is not new. Scandals occasionally erupt when it emerges that some famous person has hired a human ghostwriter rather than putting all the words on the page themselves. Eventually, though, that outrage subsides. We now frequently see ghostwriters getting credits and sometimes even shared bylines.
Similarly, speechwriters have been putting words into their clients’ mouths for decades, and no one is scandalized by that. We do, however, appreciate a speech more when it seems to be extemporaneous. We also like speeches when they reveal something characteristic about the person, not just through their subject matter but through their very phrasing. When Michelle Obama says, “You gotta look at me now,” addressing her husband, who is embarrassed by the honors she is heaping on his head, we know that moment might have been scripted But it’s no less moving for that.
As someone who works as a ghostwriter, speechwriter, and editorial consultant, I live in this world every day. I help my clients find the stories and words that will make their writing more powerful. Sometimes I write the words for them, sometimes I’m editing what they wrote. It’s always a collaborative process involving conversations, discussions, revisions, and iteration. At the end of the day--with a client who’s truly engaged in the process--we have co-produced something that they can recognize themselves in and stand behind.
That engagement is the key to successfully using a consultant, and I believe it’s the key to working with AI, too. You need to engage with the AI, interrogate its answers, push back, demand revisions, and get your hands on the copy yourself. Make it your own. Put yourself in.
Two things about the article make me uncomfortable, though.
First, Fielding is making a sharp distinction between being a subject matter expert and being a writer. Fair enough: That’s why there’s a market for people in the second category, to help people in the first. But the distinction is often not so neat. A writer with no knowledge of cybersecurity would be a poor choice to help a cybersecurity expert write a book. Familiarity with the field helps close the gap between writer and expert, yes. But also, the expert may discover that their ideas are not as clear and sound as they thought they were only through the process of trying to express them in words.
Second, an experienced academic like Fielding has enough writing skill to recognize what’s good when he sees it. His article proves that: it’s well-written, personal, and compelling. But how do you develop skills like that if you using AI from the start?
Students who rely on AI to write their term papers will never develop the writing skills they need to differentiate good and bad writing. Worse, their critical thinking skills may atrophy, as a New York Times feature story this week discusses:
“...there is no part of the writing process that can be handed over to A.I. without some cost. Intellectual growth occurs at every stage, from brainstorming to fine-tuning a final draft. If writing is thinking, then any part of the struggle that is outsourced to technology amounts to relinquishing some freedom to perceive.”
That’s why, as a Financial Times article put it earlier this year, “AI-native” college graduates are turning out to have alarmingly shallow ideas. “We want critical thinking, not just AI,” one finance executive told the FT.
Also worth noting: The Economist recently published results of a study from China showing that AI helps students who use it get their homework done faster and better. However, the students who used AI do worse on tests. If the goal of education is to get your homework done, AI is working great -- but if it’s to learn something, the test results show that AI might not be so good at that.
We might as well get used to the struggle. Half of Americans use AI services, and more than half (59%) are using it for writing or editing. At the same time, there’s a surge of interest in AI detection tools and AI watermarking. LinkedIn even added a button so people can report posts that they suspect of being AI slop.
I’ve struggled with AI myself. In my experience, the writing it produces isn’t very good. Using AI to create drafts limits the possibilities and flattens out the voice I’m looking for in a piece, making it more generic and less interesting. On the other hand, I’ve also found it to be invaluable at analyzing data, extracting insights from long transcripts, and organizing my ideas. Like Bertie Fielding, I sometimes arrive at a writing project with a “mess” of notes and half-baked ideas, and a conversation with Claude can help me sort it out.
For me, writing the first draft myself is crucial. Without that very human start, it’s too easy for the final draft to lack the essential human elements that make writing worth reading.
If you do use AI, don’t lose sight of the goal. Writing should still be a process of discovery, of learning what you think. Test one way of phrasing an idea against another, one sequence of argument against a different structure, one word against a synonym with totally different connotations. Get your hands dirty. And remember that at the end of the day, it’s you who will be responsible for what you put out there. Own your words.
Notes
I’ve turned on paid subscriptions for this newsletter, and I’m delighted and amazed that some of you have already signed up to support it. Thank you for your support!
In September, I’m offering a short writing course for paid subscribers. Here’s a preview of what subscribers will get:
I’ll show you how to write with more authenticity and power, whether or not you use AI
I’ll explain the seven essential elements that make writing persuasive and inspirational
I’ll host group chats and live meetings where I’ll demonstrate these principles -- and you can workshop some of your writing in real time
If you want to learn how to write in a more human way, this course is for you. And if you sign up now, you’ll get 20% off the already cheap annual subscription rate of $50.
Bonus: If you join as a founding member, you’ll get a 1:1 session where I can help you with any writing project you’ve got.
Note: I am donating all of my Substack revenues in 2026 to literacy and poetry nonprofits, so you not only get the benefits of subscribing, you’ll also be supporting literacy!
I am looking forward to seeing you in the course!
Until next time,
Dylan