You’ve probably heard over and over again that it’s bad advice to try to write a manuscript with AI – whether you’re working on a novel or a non-fiction book. But why exactly is that? I break down the reasons here – and some of them may surprise you.
Memory issues cause plot holes
One of the first things the untrained reader may spot in an AI manuscript is that the plot doesn’t quite line up. There might be character details that seem to change throughout the book, or things mentioned once and then never resolved. This is because AI conversations, such as the chats you use to generate each new chapter of your manuscript, only have a limited amount of memory available. The AI agent itself usually decides what is important enough to save to memory and what isn’t, and this means it may miss details you would consider to be important.
It’s really easy to test this out for yourself. Set up a very specific location in a prompt: for example, a one-room cabin that has one window, one chair, and one bed. Now ask your AI agent to describe a scene taking place between two people. It might get it right at first, but the longer the chat goes on and the more you ask it to generate, the more likely it is that it will start to forget the basics. You are likely to find both of your characters sitting down on chairs, for example, because the AI starts to assume that two people = two chairs, even though you specified there was only one. Multiply this across a whole novel – tens of thousands, or even hundreds of thousands, of words – and you’ll find eagle-eyed readers catching a side character with blue eyes on page three but green eyes on page two hundred and three.
Hallucinations cause inaccuracies
AI is well-known for hallucinating. Here’s how hallucinations often work: the AI goes to search the internet for the answers to the question you asked, or for the specifics of how something works that you’ve asked it to describe. It hits an error when attempting to read the results, and so it just… makes up what it thought it might find. It can also be swayed by incorrect results in search that aren’t verified by experts – if someone gave an incorrect opinion on a Reddit thread once, that might end up turning into gospel truth by the time the AI gets its hands on it.
This is a problem for novels set in the real world, because the AI might get the details of how things work wrong – and it’s a huge problem for non-fiction manuscripts. You can’t establish yourself as an expert on your subject if your book gets the details of that subject wrong. And if you really are an expert – and just using AI because you’re not confident in your writing skills – then you will have to waste a lot of extra time trying to correct all the hallucinations you spot.
The easy route causes missed steps
As humans, we like it when things are made easy for us. Once something has been automated enough, we take it for granted. When you turn on your smart television, you expect it to show a picture on the screen, connect to the internet, start streaming, and show you the film or show you wanted to watch. But when something goes wrong, the average layperson has no idea how to fix it. In the early days of all of these systems, they were built by experts who knew how to fix any step of the process that went wrong – but now that it’s all automated for us, built in factories and connected by engineers, we don’t have the knowhow necessary to correct things that go wrong.
A manuscript can suffer from the same loss of expert hands. When we automate the process of writing it, trusting it to an AI agent, we might just assume the AI has done everything correctly – because it did fifteen times in a row, so the sixteenth must be correct – and leave it at that. Trusting the AI, or getting lazy due to the ease of automation, can cause the human behind the prompts to skip steps such as proofreading or hiring an editor. That’s why plenty of books are being spotted, already published and out in the wild, with glaring errors such as prompts accidently copied and pasted along with their answers, or the AI agent’s introductory response (“Sure, I can write that for you”) above the intended text. It also allows those plot holes and hallucinations to get through into the live public version of the book. These mistakes can erode public confidence in your book and badly damage your reputation – and they tend to go viral, too.
Homogeneous writing causes disconnect
Another reason why AI manuscripts tend to read so badly is because they’re all written in the same tone and voice. AI has a very recognisable syntax and tends to default to common words and phrases – you’ve probably heard about the overuse of the em dash as an example, but this is far from the only giveaway. One or two things that AI tend to do (like calling a character Marcus Vane or using the “not X, but X” cliché) might not be enough to red flag a manuscript as AI on their own – after all, AI learned from human writers, and human writers must have done those things first. But when they start stacking up page after page, example after example, it becomes glaringly obvious that your book was written by AI.
So, why is it a problem if people can tell? First of all, many readers will instantly put down your book and rate it 1-star to let everyone else know it’s AI slop and no one else should bother buying it. Ratings are incredibly important for book sales: the higher your rating, the more chance you have of algorithms such as Amazon’s search results boosting you. They also affect buyer behaviour. If your average rating score dips too low, your sales will crater.
Secondly, this homogeneous style is everywhere: LinkedIn posts, Substack articles, spam emails, and now your book. When we read the same style over and over again from apparently different authors, those authors and their works fail to stand out. You won’t win an army of dedicated readers if your book sounds the same as everyone else’s. This style, which also tends to be rather emotionless and uncanny, causes a disconnect between your words and reader. They won’t feel them in their soul the same way they might if you put your fingers to the keyboard and actually wrote them yourself.
If you’re worried about writing your manuscript yourself, there are other ways to do it than using AI. Whether it’s a lack of time, a lack of expertise, or just a lack of confidence, there’s a solution that has been around for hundreds of years and still works just fine, avoiding all of the problems we’ve described above. Ghostwriters can write a book for you with the human touch, in your unique voice and style magically translated to the page, with everything correctly researched, verified, checked, and edited. Get in touch today if you want a manuscript that actually works – and the potential for a great book.
No responses yet