Gardener, Baker, Weaver, Architect: How Authors Use AI Today

Over the last year, I’ve listened to and participated in countless panels about AI use in writing and publishing, but only a handful of discussions (like this) feature authors actively using the technology. Even then, it’s rare to hear anyone speak openly about using AI models to generate and publish new material, for good reason. Ethical and legal concerns persist, and it’s fair to say AI-assisted book writing remains a controversial practice. Earlier this year, when a chapter of the Romance Writers of America announced a class on AI-assisted writing, it had to be canceled due to pushback from the community.

That said, you can find plenty of thought leaders—particularly on the self-publishing side of the industry—who aren’t afraid to talk about AI use and have long promoted it. One such person is Joanna Penn.

Another author repeatedly referenced among AI advocates is Elizabeth Ann West. She’s published more than 25 novels, has spoken on AI at major writing conferences (such as 20Booksto50K), and worked briefly with AI writing software Sudowrite in 2023 as community and education lead. She is one of the co-founders of Future Fiction Academy, which offers not only online education on AI writing tools but also an AI prompting tool designed just for writers, Rexy.

I spoke via Zoom with West in early July about the use of AI to write and publish and how she sees these tools changing the industry. This interview has been edited for clarity and condensed.

Jane Friedman: How long have you been using AI to assist with your own writing?

Elizabeth Ann West: Since 2021. Before that, I was using Dragon Dictation, which is natural language processing. A lot of people don’t realize that the technology of predicting the next word is actually more than 10 years old. It wasn’t just what it heard—it would also try to predict what word was being said based on the context of what it was already processing.

The first book I published with AI writing inside of it was A Test of Fire, and that went out in December 2021. I was using Davinci-003 as a model and Sudowrite. [Editor’s note: Davinci was an early version of OpenAI’s GPT-3; Sudowrite was founded in 2020 and uses models like ChatGPT to function.]

Why did you start Future Fiction Academy?

I started teaching people how to write with artificial intelligence in December 2022. At that time, I was mostly teaching OpenAI Playground, a little bit of ChatGPT, and I would teach Sudowrite. Sudowrite really needed a community manager. So I signed up, and I was a community manager for them for a couple of months. I left in April 2023, and I started Future Fiction Academy. I really am appreciative of my time at Sudowrite, but I learned firsthand that authors’ interests and the interests of these companies building AI tools do not necessarily align. I was the only woman on the team. I was the only author on the team.

I would love for you to say more about that, not that I want you to burn any bridges. Why was there a misalignment?

There are a lot of things about writing a book that people don’t know if they’ve never done it. There’s a lot of mythology, a lot of stereotypes still out there. A good example—and I’ve been public about this—is when Sudowrite launched the outlining tool. It had the Hero’s Journey, and it had, you know, a Save the Cat, but there was no romance outline. And I’m like, we need a romance outline. And they said, “Do we really?” It’s the biggest genre, so if you want to tick everyone off, yeah, go ahead and release it without a romance outline.

And there’s certain things they didn’t understand that authors would want. All these tools come out, for example, and none of them give an export option. They think we want to keep our words in these programs, and that we’re happy to copy and paste. I got tired very quickly of constantly advocating, and I was kind of looked down on because I don’t have a degree in computer science. I’m not a coder. I’m just somebody who’s highly technical. And I write really good books. I live in that Venn diagram.

So when I left, I knew I wanted to start Future Fiction Academy, and I wanted to teach authors to go straight to the source, go straight to the AI. [Our prompting tool], Rexy, is not designed to have authors write books with AI in only one way. It’s designed to let authors write books with AI however they want to. So a lot of software coming out is saying: Fill in this box for your characters, fill in this box for your settings. Rexy doesn’t do that. [Editor’s note: Sudowrite is AI writing software built on top of existing tools such as ChatGPT and Claude; the material you generate at Sudowrite is seen and accessible by Sudowrite. Rexy allows writers full control of their prompting and requires them to have their own private accounts and API keys for access to the AI models they prefer to use. It does not store or see anything the writer is working on.]

How would you characterize the people taking advantage of your tools? Are you getting traditional authors?

We don’t ask if anyone is traditional or not, especially because that’s a sensitive issue right now. Agents can’t get any publishers to accept AI books right now. I don’t think publishers plan to involve authors at all [in AI-generated work]. I think they’re going to want that to be internal. The majority of our members, though, they have published more than five books. So these are professional authors.

So let’s discuss traditional publishers for a moment. I think they’re cautious about this because, technically speaking, you can’t get full copyright protection for an AI-generated manuscript. And there’s the legal gray area surrounding what these models produce because of their original sin of training on copyrighted material they did not license. But it seems like you believe publishers aren’t accepting AI books for different reasons, more related to cutting out the author, or that’s the implication I’m hearing.

I have reasons for that. In 2023, when I was working for Sudowrite, I went to a luncheon that was about AI. There was a member from Penguin Random House (who was a machine learning engineer on staff), there was an agent from New York, and there was a woman who specialized in AI for video games (non-player characters, basically). And Penguin Random House was admitting that they were using Bloom for backlist translations. They were using ChatGPT for marketing, and they were exploring other applications for it.

In my brief time with Sudowrite, I also witnessed multiple IP holders—major companies—approach Sudowrite and ask them to create for them a content generator. And these are big names that, if I said them, you would recognize them. And so I knew it was happening behind the scenes; I literally watched it happen. And to this date, you still don’t have a publisher who has come out and said they’re not going to use it to generate text. But they are being clear that they’re using it for marketing and translations and things of that nature.

Following that to its logical end, if publishers do use it to produce texts wholesale, do you think they’re going to do it with brands like James Patterson, or are they just going to create fake authors who don’t exist?

I think both. I think they’re already doing the first one, I can’t say for sure. But if you read the Look Inside for some recent releases, those of us who write with AI all the time, we see the tell-tale signs that they’re using AI, particularly New York Times bestsellers. There’s one in particular, the first paragraph is like 15 sentences about boats, boats, boats inside of New York harbor. And when you compare that to this author’s previous work, that doesn’t even match.

AI also has a tendency to put four ideas in one sentence. You will open up a book and it will say, “Susie Q walked down the path, chewing her gum, her phone rang, and the scent of jasmine was in the air.” Most humans write in threes. Another big tell is echolalia. In the dialogue, you’ll see, “Jane, how are you feeling today?” And Jane says, “I feel fine, Elizabeth.”

In any case, there is always going to be a human interaction aspect. Whether you’re trusting a publisher to curate books—it’s a Sourcebooks Casablanca book, or it’s a Harlequin book, whatever—some readers like that. But most readers gravitate toward I’m going to love this book because it’s a Stephen King book. There is always going to be a brand.

I know prolific indie authors use ghostwriters, hire on Fiverr, and so on. Are you seeing changes there as well?

If you think about it, prolific indie authors are more akin to a small publisher than a traditional author because they are putting out usually two or three titles a month, and they usually have a bunch of co-writers. … You’re managing multiple pen names, you’re managing multiple lines. And I think that’s more possible now with AI. I don’t think authors are really aware that [using] these AI models is cheaper than anything you could ever imagine.

Recently you posted an interesting comment on Facebook after someone asked about the different ways that writers use AI. You described four types. Would you talk about those?

I also run the AI Writing for Authors Facebook group with Steph Pajonas. While Future Fiction Academy is definitely for professional authors, we also have this group that’s everybody; there’s 6,000 people in it. And humans just naturally have this tendency when something new comes along: If you don’t use it, one of us is wrong. For AI, we want authors to have the vocabulary to talk about how they work with AI and not be contentious with each other by accepting that the way you write and the way I write, they’re both correct, but they’re different. So we went with job types.

A gardener is someone who doesn’t necessarily want to use AI for writing. They’ll use it for brainstorming. We’re thinking like a gardener who goes out and chooses what vegetables or fruit they want to grow. Everybody who’s a gardener grows different things in their gardens, and everyone still goes to the same gardening club. You and I are the same, but you might grow peppers and I grow strawberries and it’s okay.

A weaver takes their own writing and AI writing and they literally weave it together. Weavers are most often pantsers or people who can’t see the story for very far, which limits your use with AI, because if you’re going to use AI, you have to give it very specific instructions.

The third type: bakers. Imagine you have a recipe for muffins. And you test those muffins with strawberries and strawberry rhubarb and chocolate chip and blueberries, but the core recipe is always the same. A baker often writes multiple drafts. They’ll run the same scene [through AI] five or six times because they’re convinced that there’s the perfect [result]. They just click the button one more time, and they have five versions of the scene. Then they do what we call kit bashing. That was a term I coined because of model airplanes: You can take different models and kit bash them together and make a whole new airplane that’s like a Frankenstein airplane. So if somebody says kit bash, everybody knows, oh, you took text from different models and you bashed them together, and you turned that into its own prose.

The final one is architect. I’m an architect. Architects use very specific instructions. I just did this last week with Stacey Anderson. We wrote two 17,000-word women’s fiction novellas in 21.25 hours from start to finish. [Editor’s note: That figure includes three hours of brainstorming, plus multiple hours of making outlines and character lists used for prompting the AI.] My mom read it yesterday. And she didn’t know it was 95 percent AI. She said, “Where’s the rest of it? I want more.” Architects are really involved in story structure, then they’re using AI basically like, in times past, before AI, as a publisher would use ghostwriters or co-writing.

So let’s go back to some of the legal questions. People are worried that they won’t get protection for whatever it is they’re doing with AI.

In March 2023, the US Copyright Office gave guidance. Specific selection and arrangement of AI writing will qualify for copyright. This means the individual passages themselves are public domain. So anyone could use probably individual sentences written by AI. But the specific sequence that I have, the kit bashing, that is my copyright. Since none of these AI models can write more than about 2,000 words at a time, any piece of writing you have that’s longer than 2,000 words, you had to select and arrange pieces of AI. Elisa Shupe was just awarded copyright for the selection and arrangement for her AI-generated book. So the work is protected. It’s definitely a gray area. But I still see movement in finding ways to copyright protect this work in some way, shape, or form. [Editor’s note: Things move fast. Since our conversation, West says current AI models can now write up to 12,000 words at a time.]

I often hear writers saying they don’t want to feed their work into an AI model because it’s going to plagiarize or steal it. Do you think that’s a valid concern, whether it’s an author or publisher? Should you feel worried about putting your work into these models?

Seven or eight years ago, that was a very valid concern, and we’re even seeing that now when there are certain phrases or names that the AI uses over and over again. If you’re writing fantasy, you’re gonna get a Luna [from the AI], you’re gonna get an Ava. But today, we are talking about models that are trillions of parameters or billions of parameters. The ocean is so deep, you would have a really hard time making a significant impact on a training dataset.

There’s a trend developing where authors add transparency statements to their books about what AI use went into it. Do you? Would you say that’s a good practice?

My books don’t have those statements. I did come out to my readers. My readers asked, “Well, can you write faster?” Yeah. That’s what they cared about.

We are actually starting an AI-forward press, Future Fiction Press. We have our books getting ready for launch in August and September. We’re gonna get a lot of hate at first. But I think that long-term transparency is the right idea. And I think, eventually, it won’t matter. But I think it’s good for people who don’t want to read AI for whatever reason that we honor their wishes.

Do you imagine that these tools will eventually be seen as, if not benign, something that’s in every writer’s toolbox? Meaning: There will not be bifurcation in the market between authors who never use AI and those who rely on it. It will just become this thing we all use, like a word processor.

I can remember the stigma about whether or not you use Grammarly or an online editor versus a human editor. And there was a trend where people were listing their human editor as a contributor on their books on Amazon. But most readers don’t care if the human editor is listed on a book or not. I think it’s going to be kind of like that. Unless you’re going to pull out a typewriter in a cabin in the woods, I don’t know how you get away from it.

Some writers accuse AI of spitting out really mundane, lowest-common-denominator work, and I admit I’m one of them. But I also see how some writers can get results that are truly unique. Why the discrepancy?

When you first start working with AI, and you’re a baby prompter, you don’t know what’s possible. And one of the things that we’ve been teaching at the Future Fiction Academy is that there are entire settings not available in ChatGPT. If you go to the Playground, now you have sliders of temperature and presence penalty and things like that—that literally changes the randomness, the selection of the words. So now, if you prompt with “I want a list of character names inspired by 1880 settlers from Germany in small-town Pennsylvania,” and you take that temperature slider and you bump it up, you’re going to get less-expected names on that list. These are components of AI that authors don’t really know about—there’s a lot more you can play with than just what they’re giving you in the chat models.

---

Note from Jane: Not long after this article was first published, Future Fiction Academy released Raptor Write, a free software that runs in a web browser and lets any author use a single key to access every model available.

---

Sudowrite Responds to Elizabeth Ann West interview

After publication, James Yu, a co-founder at Sudowrite, reached out to correct inaccuracies that appeared in this interview with West.

West spent a short period of time working for Sudowrite in 2023 and said in the interview she was the only woman and only author on the team at that time. In fact, Vanessa Panlasigui, on their customer support team, worked for Sudowrite and still does. (More women have been hired since West’s departure.) Also, co-founders Yu and Amit Gupta both published short fiction before starting Sudowrite, although neither have published novels.

West also mentioned she witnessed major companies approach Sudowrite and ask for a content generator. Yu says, “I’m not aware of any major companies or IP holders approaching Sudowrite, asking us to create a content generator. We’ve never had these conversations, but it’s possible Elizabeth did and didn’t tell us.”

While material generated at Sudowrite is seen and accessible by Sudowrite, Yu says, “This is standard practice in order for an app to provide syncing and cloud backups (like Scrivener, Google Docs, Microsoft Office, etc.).” He also adds, “We’ve been upfront and clear that we have never trained on author’s works without their explicit permission. Apps that ask you to plug in your own developer keys can’t provide this guarantee. We have direct relationships with the providers via enterprise contracts that have explicit protections on the content sent to LLMs.”