AI DIY
Using an AI Writing Assistant Without “Cheating”
Everyone knows AI is bad, except for the times when it’s not—right? AI-generated video games are bad except when we enjoy them. Using AI to do your work for you is cheating, unless you successfully explain that you’re not using AI to do your job but using AI as a tool to help you do your job, in which case you are working smarter, not harder.
On the application formerly known as that bird app, Joanna Maciejewska famously twote: “I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes.”
You might suggest that Maciejewska is talking about robotics and not AI but I think it takes a tremendous amount of intelligence to do laundry correctly. You have to think about colors and fabric content and water temperature and stuff. Even if you think about all those things you might still ruin some clothes.
Rayon exists to keep us humble.
Something AI is not very good at is recognizing when something has been written by an AI versus a human. “AI checkers” are not very good at what they do. To test this, I pasted some text into a variety of AI checkers. I got mixed results with AI checkers flagging human-written text (mine) as AI as well as passing AI-written text as human-written. The checkers had varying levels of success but were wrong as much as they were right.
Humans are not always great at recognizing AI-written content as AI-written content but humans are pretty good at identifying low-quality content that may have been written by AI or may have been written by a bored copywriter phoning it in, an intern on their first day, or a new writer just feeling out the limits of their skill. This is to say, human reviewers may flag human-written text as AI-written when the human-written text is poor, but are less likely than an AI checker—in my experience, anyway—to pass AI-written work as human-written.
Because AI is so bad at identifying its own work, and humans so decently mediocre at it, people trick themselves into thinking they can use AI to create stuff. You can ask a generative AI engine like ChatGPT to write you a story, for instance, and then paste that into an “AI Checker” and get the all clear. But you’re not in the clear. A human would probably identify your AI-written story as AI-written, or else just very poorly written, and in either case would not give it a good grade.
I’m writing about this because I see a lot of writers advise other writers to try AI writing assistants to write faster, better, and more. It’s true that using AI will make more content, and faster. However, this content will likely not be better than something a human wrote. And if there’s one thing the world does not need it’s more writing.
I do realize I contribute to the world’s surplus of writing, but that’s fine. At least Shelf Life is not terrible, written by AI, or expensive.
It’s not wrong to be skeptical of the use of AI in the arts. There are a lot of executives out there trying to figure out exactly how to replace human workers with GPTs and LLMs, and they are wrong to do so. While humans and AI can both put out a lot of mediocre, bland work, only a human can put out something truly incredible.
Approaching AI from a place of fear, proscribing its use in the arts and letters entirely, is not the right way, either. People are already using a lot of human-trained algorithmic tools to improve the writing process. Word’s grammar checker, Grammarly, Hemingway Writer, and other grammar helpers like those all do the same thing. Publishers, or those publishers’ vendors, are using sophisticated pre-editing algorithms to correct a lot of what copyeditors and proofreaders used to correct before we had these tools. As much as the finance people might like to replace humans with these tools, they are just tools—they need a human hand to use them.
Or possibly and orangutan hand or a crow using its beak. Humans are not the only tool-users on the planet. I don’t mean to be speciescentric.
Listen: I remember when WordPerfect launched for home PCs and it had “spell checker” built in and many of my teachers believed that using spell check was “cheating.” Some of those teachers would only accept handwritten essays because they didn’t want us to use computers to help us cheat at essay writing.
They were genuinely concerned that young people wouldn’t learn to spell if a computer helped with the process. Personally, I don’t see how having a teacher cross out your word in red pen and write it correctly will teach you to spell a word right but seeing Word underline it in a red squiggle and then give you the correct spelling won’t teach you to spell the word right. But that was a very real concern for the middle-aged white ladies who taught at my school.
Today we know that spell check is one tool in the writer’s arsenal to improve the quality of their work and the speed with which they can turn that work out. Before spell check, if you didn’t know how to spell a word you had to physically look it up in the dictionary or else run and ask Mom how to spell it, both of which were time-consuming.
I was unable to find solid evidence that spelling acumen has declined over the past thirty years since spellcheck has been in common use. There’s no reason to believe that spellcheck has made humans worse at spelling. It has only helped us become faster at writing.
AI, likewise, is another tool that writers can use to improve their writing, write faster, and work smarter. The trick is knowing how to use the tool. Just as with coding, the garbage in, garbage out philosophy holds: You have to know how to chat to the GPT if you want to get a good result.
Having spent some time fooling around with some short stories and novel chapters in Open AI’s ChatGPT engine—which you can use for free—I have five ways you can use AI to analyze your writing and help you improve it.
Information security suggestion: Be careful what you plug into an AI engine, and which AI engine you use. Read their privacy policies first. Know that anything you put into an AI engine will be used by the engine for training and maybe for other stuff you hadn’t thought of. Guard your intellectual property.
Create a Summary
ChatGPT would be happy to summarize your short story, chapter, novel, or screenplay for you. Why is that important? Well, for one thing, it can give you a good sense of whether your story is concise and sticking to your plan for it.
I plugged some content from my trunk (a couple short stories and the opening chapter of a novel) into ChatGPT and asked for a summary of each. I noticed that the summaries of the short stories, which were better written than the chapter, were fairly accurate. The chapter, which was more poorly written, returned a summary that didn’t follow the purpose of the chapter. The wrong things were highlighted as important, and things I thought were important were left out entirely. This indicated to me that the chapter needed tightening up to serve its purpose in the overall novel.
The chapter is in the trunk because I was able to cut it wholesale off the beginning of a longer work and chose to do so specifically because it meandered and didn’t add much.
Analyzing the ChatGPT-provided summaries helped me plan a second pass or revision to bring the material much closer to its intended form.
Write a Logline
I absolutely hate writing a logline. A logline is a brief (ideally one-sentence) summary of a book, short story, play, or movie that captures the main conflict of that story.
ChatGPT knows what a logline is and would be happy to write one for you based on the text you put in. For example, here’s a logline it wrote for Shirley Jackson’s “The Lottery”
In a small, insular town, an annual ritual that involves drawing lots reveals a shocking and brutal tradition that challenges the veneer of civility and exposes the dark side of human nature.
And here’s a logline for Casablanca:
In World War II-era Casablanca, a jaded nightclub owner faces a moral dilemma when his former lover and her husband, a resistance leader, arrive seeking his help to escape the Nazis, forcing him to confront his past and make a sacrifice for a greater cause.
Are these ready for prime time? No. Neither are the ones it came up with for my short stories or my novel (based on the one discarded chapter it read). But if writing a logline is a nightmare for you like it is for me, AI can provide a starting point that you can work on.
And if the logline comes back from your AI engine and entirely misses the point, you might need to critically reread your work to make sure the conflict is coming through loud and clear.
Provide Literary Analysis
Now with four fewer years of going to college to learn how to do this. I’m just kidding. You’re never going to get a Dr. Fischer’s 10am MWF English Literature 201–level of literary analysis out of ChatGPT. But it can analyze your writing and identify issues therein if you ask it the right questions.
Some of the prompts I supplied that returned helpful answers were:
Analyze this text for writing style.
How’s the pacing of this text?
Does this text have a lot of exposition?
What percentage of this text is exposition?
Is this text boring?
What are the stakes in this text?
ChatGPT offered up a lot of useful information. I was impressed that ChatGPT was able to identify a moderate-paced story versus a fast-paced story when asked to tell me about the pacing of each. It’s not perfect, but the more pointed the question you ask the more useful an answer you’re likely to receive.
Suggest Fixes
If you think your work could be better but you’re having a hard time figuring out what you need to edit, and if you’ve exhausted the basics that Word and Grammarly have to offer, you can ask ChatGPT or another AI engine for some help. Here are some questions I asked ChatGPT about my texts:
How could I structure this text better?
How is the grammar and syntax in my text?
How can I improve the pacing of this text?
How can I eliminate exposition in this text?
Where are some places that I show instead of tell in this text?
The information it spat back out was fairly generic but tailored to the work I showed it. For instance, it might give generic advice like “increase tension gradually” but then it provided examples from my text and made suggestions on how the text could be changed to improve the pacing. When I asked about showing versus telling, ChatGPT offered several examples of “telling” from my text and drafted examples of how they might read if they were “showing” instead. This use alone would be a big help for those who are always hearing “show don’t tell!” but have difficulty identifying the places where they should put that advice into action.
Draft a Query Letter
Finally, I asked ChatGPT to draft me a query letter for each of the texts I had given it. The query letters came out completely generic—but, they followed established best practices for agent queries and provided a solid foundation on which I could build a usable query. If you’re not sure where to start, you can ask ChatGPT to draft a query letter for you and it will hand you a starting point. If you give it a little information about yourself, it can draft an author bio for you to work on, too. ChatGPT also promised to help me work on my nonfiction book proposal. It can’t do worse than I did, I’m sure about that.
If you have questions that you'd like to see answered in Shelf Life, ideas for topics that you'd like to explore, or feedback on the newsletter, please feel free to contact me. I would love to hear from you.
For more information about who I am, what I do, and, most important, what my dog looks like, please visit my website.
After you have read a few posts, if you find that you're enjoying Shelf Life, please recommend it to your word-oriented friends.