Six Ways LLMs Actually Make Writing Harder

Over the past year, the majority of editing and ghostwriting projects I’ve accepted from clients were at least partially produced by an LLM (including ChatGPT and the like). And I get it. It’s so fast. You’re imagining all the hours you’re saving. Watching an LLM output text feels dizzying. Depending on your mood, you might even feel dazzled. But, as the saying goes, all that glitters is not gold. From an editorial perspective, LLMs actually create more problems in writing. 

But if you’ve been drawn in, I don’t blame you. There’s no denying tech marketing is powerful. Through the pace of change, market forces, and more, we’ve been conditioned to believe that we need apps and digital tech to do everything. And because LLMs seem magical, it's easy to overlook what, to me, is the elephant in the room: for most  writing, an LLM is actually redundant. 

What, you say? From the bottom of my writer and editor heart, I’m telling you it’s true. By a mile — or several — it’s still humans for the win when it comes to writing. In this post, I will cover six key reasons why LLMs make writing harder, and you’re better off writing on your own.

1. LLMs output words, not substance.

The outputs in manuscripts and drafts that I see over and over again remind me of my school and university years and the papers that some classmates wrote the morning they were due. Lots of words, and lots of glossing over the main ideas and concepts to make it sound like you're saying something. It might even sound pretty or highfalutin. But if you read it — not skim it — there’s nothing there. 

If you go back into the text output, you may see it adds no value. So you write your own points and add meaning. Guess what? You’re doing the writing yourself. Even appropriate or “good” prompts won't solve LLMs’ BSing problems. 

2. LLMs alter (or exclude) your research. 

This is another issue I’ve seen repeatedly, and it leads to discussion when clients are feeling the high of using an LLM to write their books “faster.” In each case when a client wants to include stats or research in their manuscript, the input research or stats don’t match the LLM’s output. That’s a serious problem. 

Before increased consumer use of LLMs, I might have come across a maximum of four fact-based inconsistencies or typos in a manuscript (even an early draft). The number is now considerably higher. And let’s talk about quotes! With the use of LLMs, distortion and misattribution have only worsened.

3. LLMs generate more text that needs to be edited.

To some extent, this is self-explanatory. Think extra headers, extraneous characters, weird spacing, and repetition (meaning outputting multiple sentences that are rephrased, but say the same thing). On top of that, when you use an LLM to write, you run the risk of having to delete text that has absolutely nothing to do with what you’re talking about. Or needing to take time to repair sentences that lack context and important details. 

Sometimes what’s wrong is super obvious, and you can delete whole sentences quickly. But sometimes what’s off or missing is embedded and more subtle. I’ve seen that a lot. You’ll have to worry about more of the text feeling “off” when you let a machine “write” for you. 

4. LLMs don’t sound like you. 

Developing a personal writing style takes time and practice. There’s no getting around it. People struggle to ascertain exactly what voice is, but put simply, it’s some aspect of you on the page. And the perceptive people in your life will be able to recognize your writing right away. All writers have isms. 

ChatGPT and other LLMs don’t have isms, unless you count excessive headers, lists (the lists!), markup symbols, and content hallucinations. (Some of those hallucinations are small and others are very large, to the level of outputting an, um, story — that is, for example, supposed to be about a puppy — but randomly goes on and on about someone getting a beer at a bar, and … an alien? (So you know, I just described an actual, never-ending output that I personally witnessed. And the problem wasn’t the prompt, in case you’re wondering.)

You know how the tech companies advertise that you can produce text messages and writing in the style of Shakespeare, or some other author? That sounds cute and snazzy, but you can do that — with your own personal touch — on your own. Every writer admires specific authors, but do you want to sound too much like them? I hope not. As a writer, editor, and reader, I want my mind and soul fed, I want to see and hear your writing voice on the page.

 

5. An LLM probably won’t help you with flow and organization.

Thinking back on every book project I have taken on over the past 12 months, in around 90 percent of cases, when clients have leaned on an LLM to help them organize their thoughts and ideas, the text output only exacerbated the draft’s original issues with flow and organization, to the point where the client is left more confused.

The reasoning required to ensure proper flow and organization of ideas in a written work is uniquely human. To be sure, plenty of people struggle with both. That’s okay. But asking a friend or acquaintance who has an English degree for help will get you further faster if you’re struggling with organizing a writing project. Plus, there’s an excellent chance they would love to help you.  

6. You’ll probably have to rewrite it anyway (sorry). 

People work really hard to avoid writing multiple drafts to get to the diamonds in their writing. And LLMs are one way to fool yourself into thinking you can jump a row of hurdles in the writing process. Spoiler alert — you can’t. If you try to skip editing and rewriting stages, you’ll likely have to start over. That’s been true as long as humans have written.

In all honesty, the writing process is, more often than not, a slog, and you have to really want to dig for those diamonds. I’ve seen clients lower their personal standards to the point where good enough is, well, bland. LLMs make that worse. 

On the other hand, should you aim for absolute perfection? Of course not. Perfection doesn’t exist. But don’t give into the hype that an LLM can do most (or all) of the work for you. Everything you write should have a tiny bit of your personal artistic glitter in it, and it will, even if it isn’t fantastic. Creatives don’t hear this often enough, but Intention matters in art, too.

So you know, I’m not a Luddite.

I want to be clear that I am not anti-technology — far from it. But I do believe in using technology wisely, where it can create positive change. In certain industries, and for people living with certain physical limitations, specific needs and use cases for LLMs do exist, and LLMs were in use several years before the billions of VC dollars and hype swept in. 

I'm also not suggesting that LLMs can't necessarily be somewhat useful in the writing process. However, I work with people of all backgrounds who write books on all kinds of topics. From what I've seen so far, the use of LLMs in creative writing only leads to disappointment and the need to start over. 

Closing Thoughts 

I want to offer two more aspects to LLMs to consider that aren’t related to the writing process. First, think about our data. Do we know what tech companies are doing with our written thoughts and ideas when we input them into their platforms? Are you comfortable with potentially giving up your intellectual property, even if it’s a one-off note about a specific paragraph you left to yourself? Can we ever really know what they're using it for?

Second, in the current geopolitical environment, thoughtful technology use can be a form of protest. If you want to stick it to the tech billionaires, remind yourself that you can use your own mind to write. We all have moments where we doubt our own skills and abilities. But using an LLM is, to put it bluntly, a crutch. I promise you don't need it. Even on your worst writing days, you can create something far better than software and machine algorithms ever will.