Generative AI floods and dilutes the market for books
arxiv.org
arxiv.org
Perhaps in time it will be better but you will need a proper pipeline to write, assess, track changes etc... perhaps all that's needed is a GitHub of sorts for writing.
However... what is the chance of any book finding an audience if there are so many to compete with. It might make more sense in having something created just for you though.
This is a workflow problem.
What the LLM normally needs:
- very basic outline: what the story is about
- basic cast of characters: like fiction/plays of the 18-1900s
Then you go chapter-by-chapter. But the most critical thing is that you MUST provide the chapter beats. From start to end. Location, environment, events, any critical conversations. For a chapter that is 2-3K words, this chapter outline often becomes 2-400 words or more. Because you have the story in your head, you can handle the stuff that the LLM is likely to screw up.
There are ways to build a system where the LLM can keep track of many things, but then that becomes programming, which is just as boring.
To add to this, in my 9-5 I work as a designer. Colleagues in other parts of the business sometimes come to us with stuff they have created using AI tools. To a layman they look really good, but if you know what to look for you can spot the issues.
At this point in time my opinion is that any AI generated thing is great at making fools of the unaware, or those who are not looking at something critically. It can create something that looks like a duck, walks like a duck, quacks like a duck, but on closer inspection it's not a duck.
I think and write the outline for a chapter. Then I have the LLM produce it. I read the draft. Tweak things. Reroll. Take the best from multiple rolls and merge them to produce a organic whole. Manually patch the minor differences in pacing and tone.
Rinse and repeat.
It takes 100-250 hours across 3-4 months to produce a story. But I drop 99.9% of the stuff within the first day or two because the plot or characters do not work.
I use the web front end directly. Let's me switch between any model including Gemma 4 running locally.
That is where expectations differ I guess.
Perhaps for some uses (say a summary of some business document, or generating code which is then discarded or edited) that's acceptable. For literature, do you really feel they are useful tools? So many changes would be required IMO that you're better off just writing the story you want to tell.
They don't. I don't think people are taking full advantage of their capabilities. They produce perfectly functional prose. Sometimes, they produce something genuinely amazing to read. But that is quite rare.
People should stop expecting brilliant prose from LLMs without doing the preparatory work themselves first. Very few human authors can manage that in any case
Variations on a Theme of Saki
https://gist.github.com/s-i-e-v-e/b4d696bfb08488aeb893cce3a4...
Stuffed full of adjectives, a common problem when LLMs attempt to make the text literary.
Stuffed full of incidental detail (window inexplicably open), which doesn’t quite make sense - why would you leave a window open for people to return?
Things which suddenly loom large in the narrative without any previous mention (roaring fire).
Overuse of certain mechanisms (for example here ellipsis).
Meanders without any discernable goal from one scene to the next.
So in short, it’s a tale told by an idiot savant, full of sound and fury, signifying nothing.
Take this sentence as an example:
A biting incident, if I'm to be perfectly honest. Involving… several toes.
The ellipsis just gets in the way, the ‘perfectly honest’ likewise and this should be one better formed sentence. The several toes bit is overly explicit without being clear (he lost several toes, or toes were bitten?). Most of the text is like this - infuriatingly vague, clumsy in construction and full of non-sequiturs.
Compare this with real writing:
https://archive.org/stream/GrahamGreeneShorts/21%20Stories_d...
https://www.gutenberg.org/files/3077/3077-h/3077-h.htm#link2...
Ideally, you should read all 14 exhibits. Each has been produced by a different LLM, some local, others frontier models of the time.
E4 = Claude
E6 = ChatGPT
E8 = Mistral
E11 = DeepSeek V3
E12 = Original
The work you posted was distinctly unimpressive, particularly as it had a seed text to impersonate and improve. The 'window' is I assume a french door, which I suspect is a mistranslation. If you want to seed LLMs I'd start with texts that are not translated!
Honestly compare it to the human writing examples I showed you and consider the differences in intent and style, which are vast.
I see why you're interested in this, I myself am interested in what these models can do and have tried similar experiments, but you should always be aware when you're asking others to waste their time on word generator output.
It is a french door/window. The original story is by Saki (HH Munro), one of the greatest short story writers of all time. He was English. The story is literally called "The Open Window."
> Sorry, I'm not doing all that work for you
Well, you asked. And I gave you stuff that I generated for comparative analysis. I am quite satisfied with the output. Most human writing, including genre writing, is subpar. LLMs are far superior to that. If you want to read only Faulkner and Eliot, then stick to them I guess?
> human writing examples
You must provide a solid prompt. And you must use a strong model. Deciding LLMs suck at writing based on output from Gemini 2 Flash, and expecting writing like Graham Greene is a bit much. The day LLMs start writing like him based on one-line prompts, writers will have no option but to look for other jobs.
I would not be satisfied with this output, I mean if your goal is mildly satisfying potboiler fiction I guess it could be ok?
If a great writer used LLMs, then the output would be outstanding. Of that I have no doubt.
By default, yes. That's exactly the same for code: by default, even with planning and patient nudging towards best practices, you get passable code at best. Not elegant, not performant, and not particularly readable, either. Basically, an uninspired salaryman type of code.
In both domains, you can get much better results in some specific circumstances. Prompt, skills, memory, and the task must align, but when they do, you can get good quality building blocks that you can then work with. It's crucial to recognize the instances where there's a chance of getting better-than-average results and ones where you could dump your whole week of tokens into and still end up with a mess. This is a skill of the LLM operator.
Once get that skill, the only issue is capitalizing on it: basically, how you fit handling of the generated building blocks into your workflow. If it's seamless, it can be a big win. If it's not - yeah, writing by hand from scratch is often better.
Given the examples of writing I’ve seen so far from them I’m skeptical that LLMs are useful for any writing where quality and truth are important.
That depends on the problem you're trying to solve.
I found it to be of great benefit when trying to overcome "writer block". Mediocre prose - but at least some prose to work with, tailored to what you want to write - can be a much better starting point than a blank sheet of paper.
Also, editing can be fast: the output is mediocre in predictable ways; it's not hard to quickly identify everything of value in a paragraph and rewrite the rest. It probably takes longer than writing the paragraph from scratch - but only if you already know precisely what you want to write, which is far from "always" when writing prose, IME.
What you did was you started off with the high level vision of the story, you'd figure out the characters in detail, supply their motivation, think about how those things might conflict with each other etc. Then you'd think about how to add various plot points into it so different characters with motivation might find productive tension with each other. You map it out chapter by chapter, what beats go where, what things happen to push interesting conflict.
But the amazing thing about this AI system was it let you go deeper, you could go into individual conversations and actually place the words people said to each other, add color to the words by appending it with stuff like "she said, hesitatingly". Entire descriptions of scenes could be vibe generated by you imagining in your head exactly what it would look like and finding perfect words such that the image in your head would be transmitted over to the reader.
Finally, the AI system had a way to inspect the psychological state of people who might want to read your work by recruiting people who are friends of yours to read each work in progress and you can interrogate them just like a chatbot over what worked well, what didn't, did you understand what was going on over here? Did you feel invested in the characters? You can use this feedback to spin up a brand new version.
Anyway, all this stuff is a lot of work but don't believe the people who say you can just press a button and get a good story out. If you put in the effort, this AI program called notepad.exe can have you generating genuinely human sounding prose!
The overwhelming majority of content on Youtube is also plagued by AI.
You're 6 minutes in a documentary about Russo-Ottoman wars and only then you start noticing something's wrong, paintings are made up, the voice sounds similar to voices of other videos and the script has an odd non-human character to it.
Then you dig into it, and indeed it's a bunch of crap prompts glued together on top of some lame ai-generated internet research.
Can you please provide sources for this? My understanding is that he used the expression "I appreciate the pushback" in a context in which it made sense (responding to criticism), and people flagged that it sounded LLM-like.
There’s some poetry seeing people complaining about how LLM provide unreliable information in a confident tone, who then go on to share incorrect information in a confident tone.
Now it’s easier than ever. So the market explodes. But it makes legit writers harder to find. Its bleak.
Then they would be selling at a higher price due to this, increasing the profits for digital authors.
And even at that point, others will continue to write using AI to simulate being one of those people.
The thing is ... ideas matter just as much as the writing. You cannot simply say: "Hey Claude, write me a Sherlock Holmes pastiche set in the bronze age."
I have found that a 1:4 to 1:8 ratio gives nice results. That is, for every 4-8 paras of expected output, you must provide 1 para worth of character detail, plot beats etc. So you would need 100-200 pages of your own writing to produce 800 pages worth.
But most people are lazy. So the output is too.
Just give me the prompt at this point.
People are too precious about man-made art. I don't see the point. Every book ever written might have some novelty value of its own. But it mostly has the same plots, character archetypes and other tropes found in other fiction. We think there is something special about human writing. But LLMs have read the same stuff we have. And produce similar inspired writing.
It is the same with books. Reading a story based on a plot of your design, and characters you built, dealing with situations you conceived of, and having conversations you directed works for me ... even if an LLM assisted with the writing duties.
Nah. What I have is the skeleton. You want it made flesh. That is what the prose does. It produces a fully realized world. I do not care to read stories in the form of bullet points. Which is what my character sheets and outlines are.
> There is _some_ stuff you read for the prose.
By this I mean you read it for the way the writer writes: prose qua prose. Most times, perfunctory prose that serves as the glue is sufficient
Of course you can use it to quickly try out stuff, for some editorial tasks and such. But unless you learn the craft of writing I don't think others will want to read any of this stuff.
Can the writer use AI to improve their output ... is the only outstanding question.
> It's literally an art.
So was making clothes, or shoes, or cabinets, or knives, or cooking, or so many other things. Now we mass produce all of them.
But as a person who reads (and I assume you have some care for what you're reading) why would that have anything to do with you as an audience? Like, I get why folks would want a curated set of Readers Digests, but why would you care to make -more- of that output?
Speak for yourself. I would, if the plot appeals to me and the writing quality is good (which it increasingly is).
Producing a bunch of filler content to pad things out is the opposite direction from most of the reading that I have found helpful. In fact, shortening works seems to be the main part of good writing. I don't generally want to watch a 4-hour film that could have been a tight 50 minutes. I don't want to read 800 pages that could have been a solid 200.
I'm not reading for distraction, and if I were Zane Gray and Robotech and Harliquin romance novels already have that filled- heck, I could just re-read Cujo (which probably wasn't edited very much at all, admittedly).
Generating a ton of crap, as you note, has never been a problem.
Getting rid of the 4-8 paras of thoughtlessness has always been the real work in my experience.
Padding, though it is probably the wrong word for what the LLM does when it expands the outline, lets you experience the world. The modern 8-episode tv season, for instance, might be technically well done but does not leave you with sufficient time in the world as compared to the 22-episode season with a bunch of filler episodes
Doesn’t really matter what the original motivation was (money, reputation, fame), people will use LLMs as a shortcut to generate text, and we’re all poorer because of it.
Now you can do it in minutes, get 50 books out, 20 of them sell 10 copies, you've earned $200.
(i might be exaggerating the times and earnings, but the point stands)
Some HR systems even support that and have AI writing/rewording integrated.
Corporate hellscapes intensifying, I guess.
It's both a new technology that maps a geometry to existing language - which is very interesting from several literary theory POVs - and represents a new kind of typewriter. If you're interested in innovative techniques in fiction, I think LLMs are arguably deeply interesting. Mass-producing slop, however, is probably quite limited in terms of its theoretical interest, and in how it changes the craft.
Personally, an example of how I use gen AI when writing is that I sometimes use a little text editor I wrote. It polls a local model, and grabs the 50 most-likely next token(s), rounded to one word, based on where the cursor is.
"Voice" in fiction was already about how the author chooses to deviate from the expected-next-word. So - while you could just use the tokens returned - I find myself using them much more often to choose what words to avoid using, or how to deviate from the expected outputs, and where it might be most interesting to do that.
I think there's plenty of interesting things that can be done with gen AI in terms of fiction writing, but I think we're quite a long way from working all of them out yet. Wholesale slop generation is one of the most obvious - and most boring - things writers can do with an LLM.
Both the internet and books have been flooded with ‘content’ which appears like human content but has no coherence or intention to it and tends to fall apart completely the longer form the story.
Even our autocomplete is being infected with LLM suggestions, distorting and guiding how people write.
This is a VERY significant problem for our culture long term.
We need authenticated online spaces where your account is linked to a real, government issued ID. This way, moderating content becomes feasible again. Bans will hold value, since you can’t just create new accounts all the time. I predict people will gravitate towards authenticated markets (and social media).
I for one am sick and tired of second guessing every text I‘m reading, at least give me someone to hold accountable.
People will just flood to places that are not official and where they'll be able to express their views freely. That's how these things have always turned out to be.
In other words, let's say that I'll get banned from the likes of FB or IG as a result of me trespassing my government's ideological wishes, well, tough luck for FB and IG (and fuck them while I'm at it), I'll just make an account on a platform that won't abide by EU/Romanian (I'm Romanian myself) propaganda, where I'll still be able to share views with people thinking similar thoughts to what I think.
If FB or IG don’t want to use it, fair. I‘m hopeful that there’s gonna be competition if there’s a system with little friction.
It’s a losing battle in my view.
Problem is short of filtering out anything published after 2023 it’s almost impossible to find something that isn’t slop now.
Is it even feasible to filter this? Such metadata can be easily faked too in a lot of contexts.
So yes, allowing AI slop in a market where anyone can publish anything and quality is not a very strict criteria, of course it will dillute the market. But we are unlikely to see this shift in traditional publishing houses.
https://www.theguardian.com/books/2026/jul/31/crime-novel-de...
Personally I prefer stunningly good over exclusively human created.
Clearly the editor didn't think it was stunningly good, or they wouldn't have flagged it up.
To the extent that their reviewers get lazy and are not reading entire texts, as evidenced by this slipping through and then being caught by an editor (who did read the whole thing).
So no, this does not prove your point.
The book retail in my city is thriving, and consists mainly of shops where the owners curate the books.
But it doesn't feel good. At the moment it feels scary. Crossing my fingers we get through this somehow.
If I can write a query into Amazon for a book on topic X, I can write that query into ChatGPT or whatever and cut out the middleman.
Some of this may work in the short term while some people still equate the format “book” with “went through a rigorous process”.
That erosion was underway before AI (e.g. on Amazon I instinctually ignore any self-published Kindle books), but I can’t imagine this will last.
I think/hope the market will somehow make human-written books easy to recognize.
[0] https://higgsfield.ai/original-series/hell-grind/episode-1
You need a mechanism to filter out the crud. There are many, some better than others.
Oh and btw amount of human generated crap is also so great that I has long became a problem for me to find good thing in that sea.
That's very interesting, and not as dramatic a dilution as I'd expected over the period (2023 - 26), or even close to as dramatic an increase in overall number of titles (achieving a sale) as I expected.
This sample is limited to what the study's authors have defined as "self-published genre fiction", which isn't representative of the market as a whole, and has very different economics compared to, for example, fiction titles published by "Big 5" publishers.
Whether the sales of non-self-published books are potentially being diluted is an interesting question, because it will tell us slightly more about where those AI sales might be coming from. I'd like to know unit and revenue numbers for non-self-published titles for the same period, in the same Amazon categories - on the off-chance anyone here has BookScan access and that's something you can pull.
Overall, though, this seems like a relatively optimistic result. If the economics don't work in the longer-term - the dilution effect ends up not being worth the token cost & labour, unless you can "hit" relatively reliably - we'll end up with less slop, put out by slop-auteurs running book farms, which I think is a fine outcome.
This is not without precedent by any stretch - the only meaningfully new part of that would be the writing mechanism. It strikes me as a possible eventual equilibrium that's both realistic and makes logical sense.
Putting aside the AI-fiction panic, and "the novel is dead" hyperbolics (which is something that's been said about as long as novels have existed), it probably will settle around a % of low-budget genre fiction being slop - as it always has been.
It feels a little patronising to suggest readers looking for anything more "serious" will be seduced and have their sales diluted by mass-produced LLM outputs. Readers deserve more credit than that, I hope.
But it looks, to me, really hard to read anything directionally on that at all, especially without equivalent numbers on titles from actual publishers.
This is what is happening with the low-hanging fruit of digital software, art, music and writing.
"Abundance" can be a great thing depending on who you ask but a nightmare for those who are selling quality.