Ghost, meet machine
What can AI do for writers?
For months now, I’ve been stuck on a complicated piece. I keep accumulating notes without a clear idea of how the whole thing will eventually fit together. Sitting in front of the file, which is now huge, I struggle to find my way in.
A few days ago, instead of opening that file again, I started a new Gemini chat. A friend recently started using the Google chatbot and has been raving about its usefulness in matters of daily life.
“I have an essay to write that is long overdue,” I typed. “It is based on interviews I conducted as well as secondary source research. Can you help me come up with a plan for finishing it?”
As always, the chatbot responded first with support. “It is completely understandable to feel stuck, especially when a project is already late and the pressure is mounting,” it reassured me. “When a feature relies on both deep secondary research and hours of interview transcripts, the sheer volume of material can feel paralyzing.”
I felt … seen.
It continued, “The goal right now is to stop looking at the article as one massive mountain and break it down into high-velocity, manageable sprints.”
I know this, obviously. I have given the same advice to students countless times! But when I’m struggling myself—paralyzed, if you will—it’s hard to put the strategy into action. Perhaps this is especially because I work mostly at home, by myself. My pets do the best they can, but they can’t give me professional support.
As I watched, Gemini unspooled an eight-day plan for finishing my article. I would spend Day 1 and 2 doing a “triage” of all the material. First, it told me to write down in a few sentences the core question the piece is trying to answer (the “nut graph”). Then I would go through all the interview notes and transcripts, but rather than reread everything start to finish, I would scan for “golden quotes” that directly relate to my theme. Everything else could be considered secondary for now.
On Day 3, I would write a “scaffold” outline, setting out the major beats of the piece and then dropping all the quotes in where they belong.
Days 4 through 6 are the big writing days where I try to get the whole messy first draft down, leaving TKs when necessary (placeholder text where I need to look up a fact or check a quote). And on Days 7 and 8, I fix the TKs, polish the language, and finally read the entire thing aloud to check for awkward moments and pacing problems.
None of this advice was groundbreaking. I know how to write an article! But to be honest, part of what was holding me back was the low-level depression that can set in when I’m not making progress. In that mode, to see all the steps spelled out like this gave me an odd sense of relief. It felt as if someone was cheering me on—or maybe just holding my hand—while I got myself moving again.
Until now, I’ve been loath to use AI for any work-related task, though I have tried asking it mundane questions about travel, dog training, household repairs, and the like. That reluctance is mainly due to ethical qualms. In addition to the primary concern about AI’s environmental consequences, nearly a dozen books that I either wrote myself or contributed material to are listed in the Anthropic lawsuit, which means they were scraped to educate an LLM without my consent. The thing sounds like a human because it has been trained on text by humans like me.
And my very first experiment with ChatGPT, in which I asked it to come up with restaurant suggestions in a Brooklyn neighborhood I don’t usually frequent and it produced two “hallucinations” in response—was sufficiently discouraging to keep me away for a while.
But in the professional circles where I spend time, I’ve seen the conversation shift over the past year or so from “AI is a monster that will destroy our livelihoods” to “AI is here to stay—how can we best harness its power?” A publication I regularly write for recently issued ethics guidelines for using AI during the reporting and writing process. (The essence: don’t use it to generate text or ask it to edit your writing, but some help with brainstorming or structure is permitted.) Last summer, I tried out a few AI apps for transcribing interviews and was pleasantly surprised by the speed and accuracy of the results.
After my experiment using Gemini as a project manager, I posted on my Facebook page to ask for opinions on whether writers can ethically use AI. Some of the responses were vehement. “Please don’t feed the machine,” one person wrote, while another called me a traitor to humanity. Most were concerned primarily about the environmental costs, although others suggested that there can be no ethical way to use an LLM since the system they are built on is too deeply flawed. Why not look to a writing book for advice? one friend wondered.
But when it comes to technology, we are all ethically compromised. This debate was conducted, after all, on Facebook, which everyone who participated in it is still using, despite the damage it has done to humanity. We all make tradeoffs in our behavior based on our values and our circumstances. If you abstain from using AI but drive a car 50 miles a day or eat red meat, are you ultimately harming the environment more? Olga Tokarczuk, one of the most environmentally conscious writers I’ve ever encountered, recently discussed using A.I. for her research.1
Meanwhile, have you ever evaded a paywall? If so, you’re taking money out of the pockets of journalists like me and contributing to a social culture that doesn’t value paying for media, resulting in its devaluation. Part of the argument against AI is that its “thoughts” are simply a dressed-up aggregation of search engine results, without attribution. But we’ve been rewarding aggregation over argument for years now in the websites we choose to click on. Remember the days when only Buzzfeed and its ilk had “listicles”? Now they’re everywhere.
Still, I don’t plan to consult an AI chatbot as a regular part of my writing practice. In addition to all the reasons above, I fear growing psychologically dependent on it, like the New York Times writer who just used AI to help sell his house and wound up feeling that “it was now almost impossible … to make a decision without getting A.I.’s opinion.” But at the moment, I’m hopeful about the possibilities of using AI to support my work rather than replace it.
Are you in the “never A.I.” camp, or do you use it for help with writing, research, or other tasks? If so, what have you found most useful? Let me know in the comments.
I couldn’t resist telling Gemini I was writing about using it as a writing coach. “What other ways do you think you can be helpful to nonfiction writers like me?” I asked.
After the obligatory sycophancy (“That is incredibly exciting to hear—I’m honored to get a mention in your newsletter!”), this is what it suggested:
Interactive digital whiteboard: “You can dump a messy outline or a complex set of themes into our chat and ask me to analyze it from different perspectives.”
Devil’s advocate: “You can feed me a thesis statement or a specific historical interpretation and command me: Act as a skeptical reviewer or a specialist in [X subject]. Tear this argument apart. Where are the weak points? What assumptions am I making that require stronger documentary evidence?
Information synthesizer: “You can upload or paste dense historical texts or trial transcripts and use me to quickly locate specific needle-in-a-haystack details, summarize dense institutional histories, or cross-reference timeline events to ensure chronological consistency.”
Admin manager: “I can take over the administrative cognitive load. You can ask me to draft pitch letters to editors based on your notes,2 format your bibliography, craft engaging social media threads or newsletter teasers about your upcoming work, or break down your massive macro-deadlines into micro-tasks for the upcoming week.”
Sample text generator. [FOR THE RECORD: I WOULD NOT USE AI THIS WAY OR ADVISE ANYONE ELSE TO DO SO.] “I am completely immune to writer's block because I don't care about making mistakes,” Gemini writes. (Interesting phrasing: Isn’t it supposed to care about not making mistakes?) It continues, “If you are stuck on a transitional paragraph or a tricky introductory hook, you can tell me: I need to get from Point A to Point B. Give me five completely different, rapid-fire ways to transition this paragraph. They might not be your final prose, but breaking that initial friction gives your brain something concrete to react to, edit, and transform into your own voice.”
I REPEAT: I HAVE NOT USED AND WOULD NOT USE AI TO GENERATE TEXT, WHICH SEEMS DIFFERENT IN NATURE FROM THE OTHER POSSIBLE APPLICATIONS.
Do any of these suggestions strike you as useful? What else would you consider asking A.I. to do?
Where to find me
New York, July 26, 2 p.m.: I’ll be in conversation with my friend Heather Clark, author of Red Comet: The Short Life and Blazing Art of Sylvia Plath and, most recently, the novel The Scrapbook, which deals with the aftermath of the Holocaust two generations later. Heather’s writing is definitely 100% human generated! Details here.
Online, July 25, 1-3 p.m.: I’m hosting a single-session workshop, “Write Like a Mother: Finding Your Voice After Having Children.” It reframes motherhood as a creative opportunity rather than an obstacle, as well as offering practical strategies for organizing time and work. We’ll look at passages on parenting and art from brilliant mother-writers like Shirley Jackson, Audre Lorde, and Ursula K. Le Guin, and discuss how to keep our work moving forward when life is unpredictable and exhausting. Sign up here.
What I’m reading
Thinner by Stephen King, in which a man falls victim to a curse that causes him to waste away. Also 100% human generated, since it was published in 1984. (Maybe someday books will bear a sticker like Ivory Soap: “99.4% human generated!”) This book has not aged well in terms of its stereotypes about “Gypsies,” a major plot point. In that respect at least, we’ve come a long way in the last 40 years. But I was intrigued that its horror consists in inverting the desire to be thin. A great number of people surely still believe that losing too much weight would be a good problem to have rather than a literal curse.
Note that Tokarczuk didn’t “use AI to write her latest novel,” as the LitHub headline claims. Meanwhile, no one seems to have picked up on the truly shocking thing she said: that her current novel will be her last, because she believes people are no longer interested in reading complex literary fiction. If Olga Tokarczuk thinks she no longer has readers, I’m not sure how much hope there is for the rest of us.
I’m pretty sure my editors would not appreciate this.
I'm writing creative nonfiction now, so it's a little different from what you're doing. I use it for feedback only. Please give me your harshest assessment of this essay in terms of [fill in the blank] (literary value, voice, pacing, diction and word choice, structure). It's helpful! Like you, would never use it to write anything.
Rather than have honest, deep conversations about AI and the points you raise, many writers take the shouting from the rooftops path. Even with what you placed in CAPS above, I'm sure there are people reading this that think you will that, as if using AI at all means one is firmly planted on the dark side.
I appreciate you brining an honest and authentic perspective to the conversation...though we are in a world dominated by algorithmic bubbles and strident self-righteousness, so few will probably see it as a chance for discussion.