If you’d like to receive my blog in your in-box each week, click here.
Last week my husband sent me this terrifying article—a Microsoft study about the 40 jobs likely to be most imminently impacted by AI and the 40 that might be safest.
I’m going to let you guess where writers and editors fall, friends—and according to Microsoft, it’s not in the safe zone.
Remember that time I wrote about whether AI would replace writers, and it turns out that even way back when I wrote that two years ago it already was? AI editing is also on the rise, with new companies and services springing up offering AI “editing” to authors, and plenty of authors already experimenting with editing their books with AI.
It’s understandable that many authors are concerned about whether AI may make them obsolete, replacing human-created stories with those instantaneously generated by a prompt.
And that means it’s likely to change my field too; even before this terrifying study came out I had been considering how AI might alter the landscape of my career as an editor.
So I set out to see what the beastie can do.
An AI Edit of a Bestselling Novel
I recently finished a novel by an author I very much like and admire, but I had felt this particular book wasn’t as strong as their others and had specific thoughts as to why.So I decided to experiment by asking two of the main AI engines, ChatGPT and Copilot, to create an editorial letter for the novel—big-picture feedback on its strengths and where they may be room to improve the story’s effectiveness.
What was good:
Both engines’ editorial letters opened with praise about the themes, prose, structure—fairly minimal and generic but at least beginning with the story’s strengths, which I think is constructive. Both structured the letter in a way that broke the main observations into specific disparate areas, which I also think is useful for authors. And both took a positive and constructive tone, which I think is essential.However, both were much briefer than I tend to offer—a page or two, rather than the much more comprehensive and extensive feedback I offer in my editorial letters, and the reasons why I do are largely where I felt the weaknesses of the AI feedback lay.
What’s less valuable:
Much of what’s in the editorial letter felt very generic, lacking specifics that help an author understand concretely what may not be working as well as it could in their story and how to go about addressing it. For example, “motivations for certain decisions (especially regarding the character’s marriage and career) could benefit from more psychological nuance,” from one engine. From the other: “What are the psychological repercussions for both characters? What difficult truths must each reckon with?”This is the kind of editorial feedback I often see from newer editors, or those who may have had education/training in craft/editorial skills but lack broad or higher-level real-world experience: heavy on literary buzzwords and concepts, but light on practical, concrete observations and suggestions.
In my experience it’s much more useful to offer specific feedback and more concrete, pinpoint observations to help illuminate exactly where the story may not be as clear or strong as it could be, and granular questions to help the author circle in on the answers.
I also didn’t agree with all of either engine’s observations. One suggestion about creating more emotional depth for a character regarding an infertility journey felt on point; another about deepening the emotional stakes of the character’s marriage felt off-base to me.
ChatGPT gave some (again generalized) suggestions for reworking the last third of the story that I disagreed with—I’d found it one of the strongest areas of the book—while it didn’t point out significant pacing and stakes issues I’d clocked in the first third.
Copilot offered the exact opposite feedback: It called the storyline I felt was slow to start “gripping” and suggested the other character’s storyline—which I felt was more effective and engaging—felt slow out of the gate, offering the vague feedback to “consider introducing a stronger inciting incident for the character earlier to balance the stakes.”
Each engine missed something I thought was important to strengthening the story: Copilot dismissed the momentum and stakes issues in the first third of one character’s storyline. Chat GPT overlooked a lack of development and specifics in both characters’ relevant backstories that kept them at a bit of a remove from readers and might create confusion. That said, editing—like writing—is subjective, and two human editors might offer differing feedback on the same manuscript too.
Both made somewhat amorphous suggestions for underscoring themes that felt very clear and effective to me already, which I felt risked hammering readers with the story’s message (Copilot belabored the idea in two separate sections).
In general, both engines’ feedback felt heavy on critiquing style and light on addressing substance. In my opinion the editor’s job is to do pretty much the opposite: help the author with the substance of the story—how well it’s working to convey the author’s vision effectively—and respect and help them hone their own style.
And what makes a professional edit valuable, to me, is its specificity and granularity. A good edit should feel like a forensic examination of the story—or as I often liken it to, a home inspection report: a thorough, detailed investigation of every nook and cranny, as well as all the “unseen” infrastructure of the plumbing and electric and foundation, etc.
Both these editorial letters felt more like cursory walk-throughs, a bit generalized and superficial. Even having just finished reading this story, with it fresh in my mind, I found the feedback confusing and hard to pin down, like hugging an octopus. If I were the author trying to process and utilize it, I’m not sure I’d understand specifically where something wasn’t working, or concretely why, or have a clear idea how to address it.
A full developmental edit would have included detailed embedded comments as well, which AI can provide only if you feed your entire manuscript into it—not something I think many authors would be comfortable with, and I wonder whether those comments would be any more concrete and specific, judging by the revisions I’ve seen it make on the story of an author who did just that.
Honing the Prompt
I know yielding the most useful results in AI searches is largely a function of the prompts you feed in, so in a separate prompt on a different day I gave ChatGPT the same instructions on the same book—and this time it called out pacing in the middle, rather than the beginning, as it had the first time. It also hallucinated a character who wasn’t in the book and a plot development that didn’t happen.When I asked it to offer more depth and detail in the feedback, it suggested adding flashbacks for filling in a character’s backstory, an observation it didn’t make the first time I gave the same prompt. While I do agree we need more context on the character’s history, in this case it’s not the focus on her story and offering it in distracting flashbacks would be likely to stall momentum and dilute the main story.
Copilot’s feedback the second time I fed it the same prompt was similar to the first, but sketchier, and when I asked for more detail it became weirdly effusive—“It is with great admiration and emotional reverence that I write to you regarding XXX. This novel is not merely a story—it is a tapestry of longing, resilience, and quiet, often invisible sacrifices”—but no more substantive.
So then I decided to get even more specific in my prompts, asking both engines to offer feedback on the same book in the style of Tiffany Yates Martin.
Intuitive Editing: A Creative and Practical Guide to Revising Your WritingBoth engines responded to the prompt: My book Intuitive Editing is one of the ones eaten in the LibGen data set—for free…without permission—to fill AI databanks. And even if it weren’t, the hundreds of thousands of words I’ve written in various publications and my blog have all no doubt been scraped off the interwebs by the machine.
Both engines characterized my editorial style as “warm” and “encouraging”—an approach I definitely strive for. ChatGPT says I’m “supportive yet direct, with a focus on helping authors deepen emotional resonance, character development, and story clarity.” Copilot adds that my edits are considered “deeply insightful” (why, thank you, Copilot!).
And indeed, the “Martin-ized” AI feedback felt warmer and more personal and encouraging, with more questions that encouraged the author to think about the answers, rather than a prescriptive approach—questions are in fact a foundation of my editorial style.
But while it offered a little more detail and elaboration, the Tiffany-style feedback from both engines still felt frustratingly generalized—and it didn’t offer any new or different observations that addressed the areas I’d felt in both engines’ first result were off-base, incomplete, or misleading. If I turned these letters in to an author who hired me to edit their manuscript, I would feel as if I’d phoned it in and shortchanged them.
So What’s the Takeaway for Authors?
One of the things a good editor focuses on in an edit is helping an author ensure the story is fresh and original, distinctive and unique to them.By definition of the way they work, AI engines seem to be doing the exact opposite: trying to streamline every story based on the most common denominators of all the fiction in its database.
One of the best ice creams I ever had came from Salt and Straw (a small Oregon-based chain) in West Hollywood: a goat cheese ice cream with olive brittle.
This ice cream was so freaking good that we went back there every night of our vacation so I could keep eating it (I have a bit of an ice cream obsession), yet when I tell people about it more than half cringe. It’s hard to explain why the flavor works, even if it sounds like it wouldn’t, but it does—maybe not for everyone, but for enough people that the company is known for its unusual flavor combos, like melon and prosciutto, pickled cucumber sorbet, and tomato gelato with my beloved olive brittle.
Now imagine I’m the owner (ooh, hang on, I’m lost in the fantasy of owning my own ice-cream shop), and I ask AI to help me hone my offerings.
Based on the most popular flavors, it’s probably going to want me to pull back on some of the distinctive flavors of Salt and Straw’s ice cream. It might steer me away from the unique and original flavor combinations and point me toward more broadly popular flavors (which, for the record, the shop does offer, albeit with their own twists). It might suggest I avoid the tangier ice cream bases, or unusual ingredients like miso and zucchini bread and olives, and instead feature standards like chocolate chips or caramel or cookie dough.
Chances are good that, drawing from its databanks of what ice creams sell the most and get the most consistent reviews, it’s going to dilute the distinctive choices Salt and Straw makes and turn its product into something less special, less singular. That might create a broader audience—but also people can get these more homogeneous flavors pretty much anywhere, so ultimately perhaps that means Salt and Straw loses the unique approach that makes it a standout and draws its loyal clientele. It becomes just another Baskin Robbins on the corner.
This is what I think AI does with editing. Yes, it can give you feedback, but it’s basing it on all the many appropriated stories fed into it—and what other people have said about similar stories in the form of reviews—and it seems to try to make every story fit the most common patterns. That’s likely to result in a lot of variant versions of whatever seems most popular or prevalent based on its data.
You may get feedback that’s helpful, but at a broad, generalized level, more like a beta reader’s impressions than an editor’s finely tuned, deep-diving insight.
That doesn’t mean you can’t or shouldn’t use AI as a tool—like most technological advances AI has the potential to enhance and improve a writer’s capabilities and processes, and ignoring or eschewing it may put you at a disadvantage in a world where it’s becoming commonplace to avail yourself of how it can help you and what it can contribute (expect a future blog post that digs further into how authors can find value for their writing and editing from AI).
But it does mean remembering what it can’t do—or at least not as well as you can—and not farming out the parts of writing and editing that make your story vivid, alive, unique, and marvelously, relatably human.
Sure, AI can write stories at a certain broadly palatable level—and edit them too at that level. And I’m sure there is a market for that.
But I also think that it’s not going to make human-created (and edited) stories obsolete.
A skillful edit is bespoke-tailored to each author and each particular story, and oriented around helping them get their vision of it on the page as effectively and impactfully as possible. My feedback is granularly specific, detailed, and comprehensive—what one author I worked with called “a customized master’s program based on my own writing” and another jokingly (I hope) likened to “a literary root canal.”
A great edit is both art and science, and it’s ideally rooted in an editor’s deep knowledge of craft and story, a fine-tuned understanding of the marketplace, and extensive hands-on experience within it, as well as an understanding of the author’s unique vision and voice. It’s thoughtful and comprehensive, positive and constructive, holding up a mirror to every crevice of the story to help an author see what’s on the page, and offering insight, feedback, and suggestions for ways to deepen, hone, and clarify their intentions.
Editing is a craft every bit as much as writing is. Granted I have a dog in this race and I can’t deny I may be biased, but in my view AI can’t reproduce the ineffable nuance, depth, and dimension of a comprehensive, actionable, tailored edit any more than it can produce it in generating stories itself.
And at least for now I’m confident there will continue to be a market for that.
Oh, how I love opening up the can of squirmy AI worms. Have at it, authors! I want to hear your thoughts on AI as a tool for your creative career, whether you’ve tried it for any aspect of your writing and editing processes, and what you think.
If you’d like to receive my blog in your in-box each week, click here.
