OpenAI ChatGPT 4o created this image. License free. If you’re a writer, then you probably didn’t understand half of that first sentence. I remember when I attended my first writers conference, the Rocky Mountain Fiction Writers Conference, back in September 2016. I’ve been a member ever since.

From talking to the other writers at the conference, I quickly discovered writers were as far from being technical as one could be. They referred to people who could format their manuscript in a format acceptable for publication, like an ebook or PDF, as coders. They thought you needed to know how to program to produce a simple PDF, or any other various formats for publication. I thought that was pretty funny. It’s just formatting. I mean seriously, the typical writer I encountered at that conference was intimidated by Microsoft Word.

I would hope times have changed in 9 years, but I doubt very much that the common writer understands anything about generative AI or large language models (LLMs). And if that’s you, then this blog is for you. And if you found any of my words above to be harsh, understand that absorbing criticism is part of being a writer, but you know that already.

You need this primer because of the benefits genAI can offer you – to be productive. Odds are you self-publish. I don’t know the percentage of self-published authors who pay for copy editing and structural editing services. I know that I paid $800 dollars for structural editing on my first novel. And I’ve yet to make $800 in book sales, so receiving those services for free is very nice. For my second novel, I relied on the beta readers, and I enjoyed our collaboration but that took weeks if not months to play out. GenAI is immediate.

So, are you cheating if you use generative AI? Is it unethical? No. The difference between paying an editor for these services and using genAI is in cost. Either way, you are receiving editing services. Your structural editor might suggest alternative wording, or in my case, prompt me to tighten up my writing by deleting 10,000 words. Claude, my AI assistant, has done the same for me. At the very least, you should agree that you can no longer justify to pay someone to correct your typos. Person or machine, the outcome is the same.

Or is it? Copy editing is not debatable. If you don’t use genAI for that service, your editor or publisher will. I can see though that anyone who has yet to use a genAI assistant might feel like discussing their manuscript with a machine would be and will always be inferior to receiving feedback from a human. You won’t be able to truly know until you try the machine.

I won’t get overly technical, but it’s time to get a bit more into the weeds. Currently, you use a web browser to query the internet. Thirty years ago, you visited the library. Now, you sit on your ass and Google responds with a ranked list of links to yet more web sites for you to peruse. GenAI responds with the answer. In a natural language dialog, I might add, which kind of feels like talking to a human. Maybe not Scarlett Johansson, but it’s extraordinarily personal. That’s a productivity enhancement on par with how the internet has saved you a trip to the library.

To understand why it’s such a personal experience, and this is as technical as I’ll get, I promise, you should understand the word, “context”. Your genAI assistant, which is a chatbot, a user interface slightly different from a web browser but one I expect you’re familiar with, tracks your conversation from earlier queries over time. And it learns about you from those queries. My GPT 4o assistant knows I’m an ultra trail runner from my constant queries on electrolytes and running gear. It often wishes me good luck on my next race. For my writing, it’s learning about my story. Context means the AI assistant is maintaining a window, a history of so many words. But it’s not called history, it’s called context. And it’s not called words. Words are tokens.

Computers don’t know what words are. They know numbers. They convert words into tokens, which are roughly four characters in the English language, but that can vary. With numbers the computer uses vectors, or math, to determine the best likely response. I could explain this in more detail but promised not to get too deep. What’s important is for you to understand the cost model is about one penny to send 1000 tokens to the cloud and maybe two pennies per 1000 tokens for the response. 1000 times 4 characters is 4000 words. Well, 4000 tokens. Your manuscript is perhaps 80,000 words. Uploading your manuscript to the cloud costs roughly 20¢ and maybe another quarter for the ensuing dialog. You need to know this to understand why free genAI assistants like ChatGPT or Claude limit your context. Like me, you’ll quickly become frustrated by your truncated dialog and opt to pay $20 a month for more context. More tokens. But start with free to learn that.

“And the operator said 40¢ more, for the next 3 minutes…”

You’re still reading. You’re curious about genAI. There are a few choices in chatbots and their underlying large language models. The world was introduced to generative AI with OpenAI’s ChatGPT chatbot. The GPT stands for generative pre-trained. A lot of work and billions of dollars went into creating a very large database with natural language capabilities. That’s the LLM that processes your queries. That process is called inference where the LLM responds to your questions. Should you use ChatGPT?

No. It’s perhaps the best at providing you with a deeply personalized experience and it’s very nice. I’ve done some research and some testing and can tell you to use Claude for structural editing. OMG, it was like talking to my old editor. I told it I would upload each scene of my prologue, one scene at a time, and ask it for structural editing comments. Sure, it made some mistakes. They were obvious. Like it told me my prologue was too long at over 5,000 words. My prologue is only 2500 words, Claude’s ideal max length. I corrected Claude, Claude apologized and stated he would put more resources into a more deliberate analysis, which he did. But everything else was gold. Claude gets me. Something he does that is so pleasurable is he first responds with what I did right, then begins his critique. Such polite critical etiquette.

Generative AI is a very dynamic space so Claude might not be the best tool six months from now. If you subscribe to a paid version, eschew the annual discount for a monthly plan so you can cancel at any time.

Now, ending with the elephant in the room, I suppose you could use Claude to write your story from the start. That’s what makes generative AI advanced from simply machine learning, it generates content. I can tell you it’s sort of obvious when a story was written by genAI, although you can then edit all those telltale signs. But you wouldn’t really be a writer anymore, would you? You’d be a structural editor.