Why You Will Always Write Better Than ChatGPT

Let me begin with an admission: I’ve experimented with machine-generated art. No matter the AI model, I find computer-generated images to be aesthetically unappealing, but there’s something charming about their jankiness. Sometimes it’s necessary to post a low-resolution jpeg of the Power Rangers sharing a pizza with Godzilla to the group chat, and it’s best that no actual human artist is harmed in the creation of this image.

Still, I was uninterested in ChatGPT, which seemed less like an amusing toy and more like an annoying tool used by content creators to manufacture endless pages of trash for search engine optimization. My casual dismissal of ChatGPT was challenged when I started to notice a distinctive writing style crop up in undergraduate student papers in early 2023. Later that summer, I began to see the same style appearing on AO3 – and even in professionally published writing. When I finally bit the bullet and signed up for ChatGPT myself, there was the exact same writing style, ready and waiting to produce anything I wanted. I hated it immediately.

At the end of this post, I’ll provide a small selection of essays written by experts about the myriad problems surrounding computer-generated art and writing. Before that, I’d like to offer my own analysis of the “style” of text produced by ChatGPT in order to support an argument for why our human writing will always be better.

(1) Stylistic uniformity

ChatGPT produces sentences of uniform length and structure, which it uses to build paragraphs of similarly uniform length and structure. Sentences always follow a “Clause 1 + Conjunction + Clause 2” pattern and contain a roughly equal number of words. Without exception, paragraphs either have three or five sentences, and the words in the final sentence of the paragraph will mirror the words of the first sentence. This visual uniformity produces a smoothness that makes writing produced by ChatGPT easy to skim. While it’s easy to run your eyes over frictionless paragraphs, however, the lack of texture makes it difficult to find a foothold.

(2) By-the-book formulaic structure

Whether writing fiction or nonfiction, ChatGPT will use the same organizational structure to present information. It relies on a set pattern of transition words and expressions to map this structure, and it always concludes with a moral or take-away point. In all fairness, this is how students are taught to write in English and Rhetoric classes. The clarity of this style must be taught and enforced because its precision doesn’t follow natural human thought patterns, which tend to meander as a writer draws from personal experience and cultural context. A dry organizational precision may be suitable for textbooks and instruction manuals, but this type of writing is infamously boring to read.

(3) Inability to surprise the reader

ChatGPT cannot lie. Granted, it’s prone to hallucinate. It invents citations, misuses specialist terminology, and doesn’t understand basic math. Still, it can’t deliberately create doubt or uncertainty for artistic effect. Moreover, as it can’t acknowledge its past mistakes or correct its assumptions, it can’t create multilayered arguments. The same lack of depth is characteristic of the fictional characters it generates. It’s precisely this lack of a flawed perspective that makes ChatGPT uninteresting and unable to provoke surprise or reflection in the reader.

(4) Lack of a distinct voice

When ChatGPT writes nonfiction, it has no sense of a target audience, so it can’t regulate its tone. Various commenters on social media have pointed out that ChatGPT reads remarkably like mansplaining, a type of address that comes off as condescending because it refuses to acknowledge the pre-existing knowledge of the reader. When it comes to fiction, ChatGPT can’t create distinct character voices, which results in generic characterizations. This inability to understand tone, culture, and context can be downright offensive when ChatGPT is used to mimic a voice grounded in a specific time, place, or identity.

(5) Failure to achieve insight

Even in the most utilitarian nonfiction writing, the process of putting words on the page with your own hands is important. This process generates a distinct mental space, and the physical act of typing enables the time necessary to analyze information properly. As you write, you make connections with your life and past experiences that create insight, and these connections create a sense of depth and meaning. Without a specificity of voice or the serendipity of insightful connections, ChatGPT has nothing to draw on save for a database of clichés.

Many professional writers spend their lives working to achieve a smooth and frictionless style, but every human has stylistic quirks, including favorite words and preferences regarding punctuation. ChatGPT has no texture or individuality at all. It’s this lack of a distinct voice (especially when combined with stylistic uniformity) that alerts me to the use of ChatGPT in student work. One might argue that a tonal flatness is the defining characteristic of ChatGPT’s “voice,” and I would agree. Even if you’re not trained in linguistic or literary analysis, ChatGPT’s generic style of writing is distressingly easy to identify simply because of how boring it is.

Instead of being a pleasure to read, the writing generated by large language models feels like the worst sort of homework. Whether you specialize in fiction or nonfiction, this is why you will always write better than ChatGPT.

If you’re interested in pursuing this topic in a broader cultural context, please allow me to recommend six recent essays written by fellow humans.

Confessions of a Viral AI Writer, by Vauhini Vara
https://www.wired.com/story/confessions-viral-ai-writer-chatgpt

AI Art & the Ethical Concerns of Artists, by Vicki Fox
https://beautifulbizarre.net/2023/03/11/ai-art-ethical-concerns-of-artists/

My A.I. Writing Robot, by Kyle Chayka
https://www.newyorker.com/culture/infinite-scroll/my-ai-writing-robot

AI-generated art raises tricky questions about ethics, copyright, and security, by Melissa Heikkilä
https://www.technologyreview.com/2022/09/20/1059792/the-algorithm-ai-generated-art-raises-tricky-questions-about-ethics-copyright-and-security

How Hollywood writers triumphed over AI, by Dani Anguiano and Lois Beckett
https://www.theguardian.com/culture/2023/oct/01/hollywood-writers-strike-artificial-intelligence

The strange world of high-speed semi-automated genre fiction, by Josh Dzieza
https://www.theverge.com/c/23194235/ai-fiction-writing-amazon-kindle-sudowrite-jasper

. . . . .

This essay was originally published on the blog of the Get Your Words Out writing community on Dreamwidth (here). You can learn more about the community on their website (here), and membership opens every December. The moderators share writing encouragement and advice on the community’s public Tumblr (here) and on Bluesky (here).