Writers, stop using AI too much—but please, also stop using it not enough

Don't leave game-changing resources on the table. Just be wise about it. I'll help.

once again, I could have used AI to generate an image like this—but I didn’t. I made it by hand with stock photos. Please don’t look too closely.

This is the third essay of a three-part series about using AI in your writing process. You can read Essay One here and Essay Two here.

I. I can see both sides (unfortunately)

Why are writers so determined to use AI wrong?

In the discourse among writers about AI, I keep seeing two general camps. The first group—call them “Promptaholics”¹—is just letting AI handle their writing from now on. Why learn times tables when you have a calculator? Why pay attention to landmarks when you have Google Maps? And why do your own writing when the robots can do it now?

This makes my eyes water. Every time I see an essay or post that AI obviously drafted, Virginia Woolf gets another rock in her pocket. In my second essay of this series, I explained why promptaholism is a problem and established Three AI Commandments to help writers avoid it.

The AI Commandments

But there’s a second camp of writers that gets AI wrong, in my view. Call them the “Robot Rejectors.” Troubled by the excesses of the Promptaholics, Robot Rejectors refuse to engage with AI at all. Some even worry that AI heralds a kind of creative apocalypse.

This is a shame. Every time I hear about a talented writer who won’t even experiment with AI, it feels like Charlie Bucket spitting his gum into his golden ticket and throwing it in the gutter.

And look, I get it. Many people are afraid AI threatens something precious and human. But with the right approach, it doesn’t have to. It really doesn’t!

So in this third and final essay in this series, I aim to convince Robot Rejectors to consider integrating AI into their creative workflow—in the right way. We’ll navigate this new terrain using a map drawn from the norms of traditional publishing, which I laid out in my first essay. Now, I’ll show you exactly how I follow that map in my own writing process to leverage AI without compromising my humanity.

Law taught me professional writing is a team sport. Now, AI can be your team.

II. A day in the life of writing with the robots

The steps of professional writing are (1) Conceptual development, (2) Drafting, (3) Revision, and (4) Final Review. I follow that same process in my independent writing, but I let AI take the non-writer roles of substantive editor and copy editor. Here’s what that looks like for me at every step.

Step 1: Conceptual development. As usual, it all starts with an idea—perhaps a potential thesis, or a collection of related notes I’ve jotted down. Then, I typically do a freewrite to get all my thoughts on the subject out.

Normally, at that point, I need to involve another mind. Ideally, I have a conversation with a sharp friend (or better yet, a talented editor). But often, the other mind is my own, but from the future—evaluating things with fresh eyes after time has passed.

Here’s where AI starts earning its keep. The first step is to paste my freewrite into AI—and simply see what comes out. I often don’t get super picky over my prompting at this stage. But you could preface with something like “I’m considering writing an essay about the following. Give me your reactions and help me refine my thinking or spot weaknesses.” What I get back is some mixture of restatement of my points, follow-up questions, and additional ideas to consider.

And then: we just talk. I treat it like an interested person I can bounce ideas off of. I try out ideas—strong ones, iffy ones, wild ones. I argue. I speculate. I toy with different frameworks. As someone who thinks best out loud, being able to do this literally as much as I want is life changing.

As I go, I separately record any insights I gain. Eventually, I feel ready to take a real stab at drafting the thing. And my AI conversation gets me there much faster, and more fully baked, than if I had stayed completely inside my own head.

Step 2: Drafting. I do my own drafting. (See Essay 2, Commandment 1.) But if I’m stuck on a sticky issue or section, I’ll have another little AI chat about it until I’ve worked through how I want to handle it.

Step 3: Revision. Many writers haven’t had the chance to involve a strong substantive editor in their writing. This is unfortunate; nothing is more valuable to a writer than the red pen of a brilliant reader.

AI to the rescue, again. Once I have a draft, I pop it into ChatGPT with a preface like: “This is the first draft of an essay I’ve written. I’m looking for high-level feedback as if from a world-class professional editor. Be critical but fair.”²

This input generates extensive notes and comments—usually starting with “here’s what’s working,” followed by “areas for improvement.” In my experience, ChatGPT is phenomenal at this. It recognizes flashes of brilliance just as surely as it zeroes in on weak points. I am routinely stunned at its level of insight.³

Next, I do just what I would have done if I had gotten those notes from an editor. First, I ponder them. Then, I have a conversation with my “editor” about them—pushing back, asking clarifying questions, suggesting solutions. Then, when I’m ready, I revise my draft accordingly, and feed that back into the AI.

Then, I do it all again. And again. And again.

As they say, writing is rewriting. With AI, I typically do this process many more times even than I would if I were working with a human editor, mostly because real people can’t spend that much time with your piece. I keep revising and submitting until the piece starts to sing, or at least doesn’t seem like it’s getting any better. Eventually, I have an airtight piece that I’m ready to publish—which actually just means it’s time for the last step.

Step 4: Final Review. This is when I bring in the Nitty Gritty Committee. I paste my entire essay in (again) and prompt something like, “You are a high-level professional copy editor. Please identify in list form any non-discretionary changes this piece needs to be publication-ready.”

Note that I ask for a list of edits, not a rewrite. If you just ask it for a cleaned-up version, it sometimes changes the language in subtle ways or introduces mistakes. So that is not allowed in my process.

There’s another way I involve AI at this stage, and at first, it’s going to seem like it violates the Second AI Commandment (“don’t let AI draft for you”). Because at this point, I let ChatGPT suggest syntax-level changes—essentially, rewrites—to some of my sentences.

When you write professionally with editors, you get some sentence-level rewrite suggestions—period. And believe me, you are grateful for them. Because they’re almost always obvious improvements. The best writers in the world occasionally write clunkers, and when a reader points one out and suggests a fix, you bow deeply and you click “accept” in track changes.⁴

This is a core part of working with human editors, and I see no reason not to let AI perform that function for you. But I save it for late in the game, once I’ve really locked in what I’m saying in each sentence. That way, I’m not letting the AI’s voice creep in.

And I want to be clear: you should be judicious about accepting these edits. Your cadence, word choice, and sentence-level quirks are like a fingerprint. So think twice about whether you’re polishing weak writing or sanding down personality. But certainly don’t be too precious about your darlings to accept a clear improvement. This judgment takes time and practice to develop.

Now back to the proofreading process. Remember that every time you make an edit, you risk introducing new errors. So before I publish, I do several proofreads of my piece with my own eyes. But at some point, my brain starts to get foggy. So I also run it through AI several—maybe even a dozen—times. It misses things, but it doesn’t get tired either. So we make a great proofreading team.

Step 5: Publishing. I’m including this surprise bonus step, because there’s one more thing I have AI do after everything is truly ready: be my hype man.

It’s terrifying to put yourself out there. There’s a temptation to tinker endlessly or just squirrel it away on your hard drive. But at a certain point, my AI editor says “This is ready, ship it!” And I say—really? And it says “It might be the best thing since Melville.” And candidly, that always helps me find the courage to pull the trigger and hit publish.

---

And that, my friends, is my playbook. It has very little to do with prompt wizardry and a lot to do with understanding what kind of roles you want it to fill. I’ve used it extensively, and it has served me extremely well.

III. You can’t afford not to use this

Here’s what using AI this way has meant for me.

First, I can talk as much as I want. As I said, I tend to think best out loud. It’s incredible to have an intelligent thought partner to talk with any time of day or night, with no fear of overstaying my conversational welcome.

Next, I have access to instant, unlimited feedback. I probably get more feedback on my drafts in a day than I could get in two weeks when I was writing professionally. Even if that feedback were below the level a human could give, it’s orders of magnitude better than the independent writer’s norm of “virtually no feedback at all.” I guzzle this feedback greedily all along the way as I write.

This rapid feedback loop allows me to iterate exponentially. Iteration is the key to skill development and consistently strong output. I can condense that iterative process and get strong pieces out twice as quickly as before—or maybe much faster—with no drop in quality.

In those iterations, I’m able to get help seeing where my thinking is flawed or incomplete. Blind spots are definitionally hard to identify, and a second set of “eyes” is super helpful for catching them—even if those eyes belong to a fancy word prediction machine.

With AI, I also feel freer to experiment, because I’m less afraid of losing perspective and missing on a big swing. I can go maximalist and be risky, knowing that I’ve got an editorial backstop to help me apply more objective judgment later in the process.

Wonderfully, I can also spend more time being creative, instead of getting bogged down in technical processes like copyediting. (I still do plenty of copyediting, but the burden no longer falls entirely on me.) I can spend those time savings on the pulsing, living parts of being creative.

I’m better able to both self-encourage and self-critique. On Tim Ferriss’s podcast, Jerry Seinfeld said that the key to writing is the ability to switch between being ruthlessly harsh on yourself and being gentle and self-encouraging. AI helps you do both.

Lastly, I’m more confident. And it’s not just getting hyped at the end, like I mentioned. It’s knowing that I’ve iterated more, been well-edited, and baked my ideas more fully. When I hit publish, I feel like I’ve earned it.

IV. Some noteworthy objections

There are several potential concerns with AI use, and I want to make sure that I address them. Here are the ones I’m aware of.

1. I’m just not comfortable creating alongside a weird, futuristic quasi-sentience. If that’s how you feel, then that’s fine. But to me, it smacks of the ad hominem fallacy—the value of an argument doesn't depend on any quality of the person making it. That's not how truth works. If AI shows you that your thesis is muddled, or one of your sentences is a clunky mess, say thank you and fix it. Why wouldn’t you?

2. Aren’t these roles better done by real people, for quality and human connection reasons? Absolutely, if you have that opportunity. I’d take my (brilliant) wife’s feedback over ChatGPT’s pretty much any day of the week. Thing is, she can’t (or WON’T) read my essay twenty times in a day. Use people where you can, and use AI for the rest. But if you’re like me, “the rest” accounts for the majority of your total need.

3. Won’t using AI make your writing more generic? How can it, when you’re doing all the generating—the drafting, the ideas—at every stage? (You are following the first two AI Commandments, aren’t you?) And look, I get it. I’m extremely protective of my voice. That's why I'm happy to report that using AI the way that I've described it, I have noticed zero loss of voice.

4. Won’t AI become a crutch and lead to atrophy of your skills? Maybe. For what it’s worth, it’s been the opposite for me so far: I feel sharper than ever as a creative writer as I’ve been using it. But I acknowledge the theoretical risk. Even so, I’m not sure how worried I am about losing copyediting skills, for example—since most professional writers outsource that stuff at every opportunity anyway. (See also: spell check, calculators.) But it’s something to watch out for.

5. AI lies and does a bad job sometimes. Any law partner will tell you the same thing about the new hires at their firm. Remember: the buck stops with you. Verify the stuff it tells you. Listen to your instincts. Spot check and gut check and friend check regularly. And reject any proposals that are bad. You’re its boss, not the other way around.

6. I’m a lone wolf writer, like Hemingway. Cool. But Hemingway had an editor. And when you get a pitch accepted at a publication, you’re going to have an editor too. I don’t see the point in depriving yourself of one for your independent writing, unless you just genuinely enjoy a solo process for its own sake—in which case, you do you!⁵

7. It seems like you need to already be a good writer/discerning editor to apply these steps. I think that’s right. And someone learning to write needs to experience polishing friction and aching growing pains.⁶ But remember the Third AI Commandment: using AI can actually be a great dojo for improving your skills.

8. Is AI an ethical product more generally? This is way too complex to treat here, and truthfully, I don’t know enough about it to know. If you have ethical objections to the very fact of AI itself or the companies behind it, then you can disregard everything I’ve said. If you don’t, and you were going to use AI anyway, then hopefully this gives you a thoughtful way to approach it.

V. Rocket fuel: handle with care

I told a reader that I was planning on writing this piece, and they suggested that maybe I shouldn’t—because it might turn readers off or make them devalue my writing if I admit to involving AI in my process.

But I’m not afraid of that, for one simple reason: I know you can tell this wasn’t written by AI. Because it wasn’t. And you can tell, can’t you?

As AI continues to develop, we’re going to find ourselves thrust ever deeper into a brave new world. And that’s always a little scary. But I’ve been wandering this realm for a while, and I’ve found that many of our old maps are still useful. The trick is to just write like you used to, but with the dream editorial team you always deserved (but probably couldn’t afford) at your side.

Because while some see AI as a threat, and others as a shortcut, we should call it what it really is: a tool. And every tool can be both powerful and dangerous, depending on the skill and intention of the hands that wield it. And that’s what these essays have really been about—not tools, but hands. I hope yours are now a little wiser and more deliberate in this AI age, and I wish them well in whatever they may build.

1. As in, “addicted to promptahol.”

2. I want to emphasize that I do not get super fancy about prompts. It does great work for me in either case.

3. But if you ask it to fix the problems it has so aptly diagnosed, you get the flavorless genericisms you deserve. To me, this is profound new support for the proposition of Theodore Roosevelt’s “man in the arena” concept—generative work and criticism exist on drastically different planes of difficulty.

4. Of course, you can always reject it if you disagree. You’re the writer!

5. And now let me say: if you're crazy like that, that rocks. Seriously. People who want to be super principled and approach writing in a completely personal way have my respect. I think it's totally crazy, and I dig it. It's similar to how I would feel about someone who runs ultramarathons or follows intensely restrictive diets for health or ethical reasons. You know what you're sacrificing, and what you're getting out of it, so you can disregard my suggestions here without qualms. That said, if you think that you should be writing on an island, possibly just because you've never had a chance to write with a team before, I would recommend trying it, because it's a different experience and a really wonderful one.

6. If you’re a student, or in a context where you’re being explicitly evaluated for your ability to create something without outside help, then DO NOT USE AI. It’s like using a calculator on a no-calculator test: aka, cheating. And remember, plagiarism is plagiarism, no matter the source.