I am going to begin with a small piece of writing:
Writing a book is a journey. But it isn’t just about putting words on a page. It’s about finding your voice, embracing your truth and creating something that truly resonates. Every word matters. Every choice matters. And that matters.
Nothing is technically wrong with it.
It is grammatical. It is encouraging. It has excellent posture.
It is also the sort of paragraph that now makes a growing number of readers quietly suspect that nobody human was terribly inconvenienced during its creation.
Spend enough time editing AI-assisted manuscripts, LinkedIn posts, blogs, newsletters, and marketing copy and you begin to recognise certain habits. Not proof, I hasten to add. There is no single word, sentence structure or punctuation mark that establishes machine authorship. Humans invented all of these things first, which is rather inconvenient for the internet detectives currently attempting to arrest one particular punctuation mark on sight.
AI learned from human writing. Of course it can sound human.
The problem is that it has also developed some very recognisable defaults.
And when those defaults pile up, the writing begins to sound less like a person thinking and more like Language Product No. 7 emerging smoothly from the content factory.
But what if what you are sharing is not merely information, but you. Your perspective. Your expertise. Your story. Your peculiar combination of experience, contradiction, humour, prejudice, mistakes, discoveries, and conclusions.
For that, I would quite like your brain to remain involved.
I have a son, who indeed works with me as he is now 28, and has several years in publishing, editing, ghostwriting, and book design under his belt, independently of me. He adds massive value to my small work team.
He also holds very strong views on AI, especially for the illustration work he enjoys (he is Fine Art qualified). He believes there is also a bigger ethical problem here that has nothing to do with whether AI can produce a tidy paragraph. The real questions are who owns it, who profits from it, what it was trained on, who is accountable when it gets things badly wrong, and how much human judgement we are quietly handing over because the machine sounds confident.
I still see AI as a tool, but tools are only useful if you know which end to hold; a knife will also cut you if you get lazy.
The danger starts when people stop using AI to assist thought and begin using it instead of thought.
Add machine hallucinations, flattery, bias, borrowed expertise and a few giant corporations packaging all of this neatly for profit, and caution starts to look rather sensible. AI can very easily become yet another tool used by mysterious beings in power to keep the rest of us thinking the way they would like us to—or, at minimum, to keep tabs on that thinking. We cannot know for sure, but we can kinda see some of the writing on the wall. Or the screen. Or whatever.
With spyware everywhere, and AI regularly inviting you to hand over your brain and quite a lot of its contents, anything is possible. I don’t have all the answers, and I still happily let AI tidy up a rushed email or flesh out a concept now and then. If the bots do eventually decide to come for me, however, I remain fairly confident they will battle to locate me hidden away in my rural, misty hills. I mean, not even Temu can find me here.
But let’s move on from the bigger debate to the main point of this article: how to fix AI writing when a bot has quietly moved in, stripping out the worst tell signs so your finished draft sounds unmistakably like you. If the illuminati are going to use your fantasy novel to achieve world domination and eugenics control, well, eish. I just don’t know.
First, AI is Not One Thing
We have actually been living with artificial intelligence for years without holding emergency meetings about our novels.
There is a ton of debate in writers’ circles about what AI even is, and many tremulous baby writers asking if they used Grammarly, do they need to declare it on Amazon. Umm, no, dear readers, you do not need to declare ALL your tools, same as you don’t need to tell anyone what computer you wrote it on. The concern is about generative AI, not that lil spelling check you did in 2007.
Generative AI is different because it produces new material: text, images, audio, video or code based on patterns learned from enormous quantities of existing data. This is where artists and writers are up in arms, to a large degree, as hard-won, human work has been used without consent to create these data libraries which everyone then randomly samples from for their “creative” work. So intellectual rape with a dash of nothing new thrown in.
Generative language models operate probabilistically. Put very simply, they have learned immense numbers of relationships between words and ideas and generate sequences that are statistically plausible given the prompt and context. They are extremely sophisticated pattern machines, but they do not have your childhood, your private memories, your professional scars or that AHA moment you reached after twenty-eight years of actually doing the work.
And probability has a natural tendency towards the familiar. Which is where the fingerprints begin appearing.
The Great “Not This, But That” Epidemic
You will recognise it immediately:
This isn’t about working harder. It’s about working smarter.
This isn’t about perfection. It’s about progress.
Leadership isn’t about having all the answers. It’s about asking the right questions.
You aren’t simply writing a book. You’re building a legacy.
Used occasionally, rhetorical contrast is perfectly effective. Humans have been contrasting one idea against another since somebody first announced that they wanted that mammoth, not this mammoth. (This mammoth breathes one massive, mammothean-sized sigh of relief).
But generative AI adores the construction because it is neat, clear, persuasive and beautifully symmetrical. It establishes an assumption and overturns it within seconds. No dangerous loose ends.
After seventeen of them in a manuscript, however, I start expecting the book to inform me that this isn’t about reading anymore; it is about embarking on a transformative experience.
There is nothing inherently artificial about the device. It is the frequency that gives it away.
Everything is Now a Journey
Careers, healing, writing, leadership and entrepreneurship are apparently all journeys now.
Choosing the correct breakfast cereal may, at this point, also be a journey of profound personal transformation. I mean, these are journeys. But does everything have to be described as one? I will buy more descriptors, adjectives, and verbs, and donate them to starving writers if I must.
Again, humans use the word. I use the word. You probably use the word.
The problem begins when abstract nouns replace actual thought. Journey, transformation, growth, impact, connection, authenticity, empowerment, resilience, meaningful, powerful, profound, pivotal and transformative can create the sensation that something important has been said without requiring the inconvenience of saying precisely what it is.
This is not merely editors becoming cantankerous. Researchers studying millions of academic texts have identified abrupt rises in particular words associated with LLM-assisted writing. One large-scale study of biomedical abstracts found unusual increases in style words after the arrival of ChatGPT, while subsequent research identified terms including delve, intricate, and underscore as being substantially overrepresented.
Another analysis across millions of academic papers found that between 2022 and 2024 the use of delve rose by roughly 1,500%, underscore by around 1,000% and intricate by approximately 700% in the databases studied.
The machines learned the words from us. We are now apparently learning them back. There is something beautifully circular about this, although possibly not in a good way.
And then we discover that everything “matters”
There is another little tail I now see dangling from paragraphs with extraordinary regularity:
And that matters.
Or:
And here’s why that matters.
Or:
Why does this matter?
Or:
And that matters more than you might think.
Sometimes it genuinely does matter, and sometimes the author has already spent the preceding 180 words explaining why.
AI tends to favour explicitness. It likes tying a small bow around the point so nobody misses it. Human writers are more likely, at least sometimes, to trust the reader to make the final half-step themselves.
That slight gap matters to good writing because reading is not merely receiving information. The reader participates. We infer. We connect. We notice irony. We remember the thing the author told us forty pages earlier. We get the joke without somebody arriving afterwards to explain that it was, indeed, humorous.
Not every thought needs a small employee standing beneath it holding up a sign saying IMPORTANT THOUGHT.
Then Come the Brick-Wall Paragraphs
This is probably one of the easiest AI-shaped habits to spot because it has spread far beyond AI into modern internet writing generally.
You get this:
You don’t need more content.
You need clarity.
You need confidence.
You need connection.
Because readers know.
And readers care.
And that matters.
Stack enough of these together and the page begins to resemble somebody constructing a wall out of inspirational fridge magnets.
Short paragraphs are useful online. White space helps screen readability. A single isolated sentence can land beautifully when the writer has earned it. But when almost every sentence has been promoted to its own paragraph, nothing is being emphasised anymore.
AI did not invent this style. It inherited a world full of SEO articles, advertising copy, email funnels, social media posts, mobile-first journalism, self-help content, and copy designed for readers who may disappear the moment their phone vibrates.
Generative AI has simply become exceptionally good at reproducing it.
The result can be clear and highly scannable. It can also make a 70,000-word book feel as though somebody has accidentally printed LinkedIn.
The Suspiciously Perfect Three
AI also has a fondness for triplets.
Clarity, confidence, and connection.
Purpose, passion, and progress.
Think clearly, act boldly, and lead authentically.
Again, rhetoric loves threes. Humans love threes. Fairy tales have known this for centuries.
But machine-generated prose can become strangely symmetrical. Three benefits. Three problems. Three solutions. Three adjectives. A sentence explaining them, followed by a sentence summarising them, followed by a conclusion reassuring us that all three matter.
Real thought tends to be slightly less house-trained. Sometimes there are two reasons. Or seven. Sometimes the fourth one is far more interesting than the first three and ruins the tidy structure entirely.
That is often where I start becoming interested.
AI Loves Telling you Where you Are
Here’s the thing.
At its core…
The truth is…
The key takeaway is…
What this really means is…
Ultimately…
In today’s fast-paced world…
In conclusion…
There is a lot of traffic management.
The prose is constantly placing little orange cones around itself and directing the reader towards the next idea. Human writers use signposting too, of course, and complex non-fiction often needs it. But strong prose does not usually announce every turn before taking it.
Then There is the Relentlessly Reasonable Voice
AI is remarkably good at producing something that sounds balanced, polite, emotionally literate, and professionally reassuring.
It produces prose in which every difficult subject has been carefully padded for transport. Betrayal becomes a “powerful journey of healing and transformation”, a furious argument is reframed as “an opportunity for meaningful dialogue”, a business failure becomes “a valuable learning experience”, and a dead marriage is repackaged as “a season of profound personal growth”. The problem is that real life is not always so obliging. Sometimes a disaster was simply a disaster, somebody behaved appallingly, the author is still furious, and grief does not arrive at the end of Chapter Twelve carrying a useful lesson and a neat little bow.
Human beings are allowed jagged edges. In fact, those edges are very often where the voice lives.
The Biggest Tell May Be What is Missing
Machine-assisted writing can contain thousands of words and still feel oddly short of anything specific.
It gives us “a challenging childhood” instead of the afternoon someone’s mother threw his school shoes into the swimming pool; “leadership challenges” instead of the meeting where twelve executives spent ninety minutes discussing a printer while the company was losing R2 million a month; and “the importance of resilience” instead of what the author actually did at 3.17 a.m. when payroll was due in six hours.
Specificity costs more because it requires memory, observation, embarrassment, judgement and, occasionally, the willingness to say something another person may dislike. AI can help organise those details, but it cannot retrospectively have lived them.
That is why I care far less about whether a sentence has technically been touched by AI than whether the author’s mind is still visible inside it.
AI Tells Are Not Proof
This deserves saying clearly because AI detection is rapidly becoming its own slightly hysterical sport. A human author can use delve, write three short paragraphs, say something “matters”, or deploy whatever punctuation the internet has currently decided is suspicious without secretly hiring a robot. Equally, a skilled AI user can simply prompt the model to avoid all of those obvious tells, which is why none of them proves very much on its own.
Research already shows significant stylistic differences between AI and human text in some controlled settings, including greater use of formal vocabulary and highly descriptive adjectives in AI-generated material. But those are statistical patterns, not individual verdicts.
The interesting thing is not whether one word is “an AI word”.
It is whether a dozen machine-favoured habits have arrived together and flattened the voice. Think less fingerprint. More smell. One whiff proves nothing. By Chapter Six, however, the dog may reasonably wish to investigate.
So Why Are Readers Becoming So Irritated?
Because books are unusual products.
If I ask a machine to tell me how many millilitres are in two cups, I do not require emotional intimacy with the person who performed the calculation. If I ask how to stop Word restarting my page numbers because Microsoft has apparently developed an active personal grievance against me, I am delighted to receive machine assistance.
But when I buy a memoir, essay collection, thought-leadership book or novel, part of what I am buying is access to another mind: someone else’s observations, arguments, peculiar connections and the things they noticed that I did not. I want sentences that feel as though they could only have come from the person whose name is on the cover. I already have bots whenever I need them; I do not particularly want to buy 80,000 words only to meet another one.
And there is evidence that readers feel some version of this distinction. A 2025 experiment involving about 1,500 US adults found a consistent preference against AI-authored books across genres when authorship was disclosed. Perceived effort, emotional attachment and a sense of proximity to the author helped explain the effect. Even among people more welcoming towards AI, the preference weakened but did not disappear.
At the same time, this is not a simple “AI writing is rubbish” story. Other controlled research has found AI-generated short fiction can be rated as highly as, and sometimes more highly than, human stories when readers judge the text itself. Yet perceptions change when people believe the work was created by a human.
That tells us something important: the debate is not simply about whether the writing is good enough, but about who or what actually created it.
Why the Publishing Industry is Nervous
There are several different arguments currently being bundled into the great AI shouting match.
One is economic. Writers, illustrators, editors, translators, narrators and other creatives quite reasonably wonder what happens when clients can generate passable material in seconds.
Another concerns training data, copyright, and consent. In a 2025–26 Book Industry Study Group survey involving 559 publishing professionals in the US and Canada, 86% identified inadequate controls around copyrighted material as a concern.
Then there is volume. Generative AI makes mediocre content extraordinarily cheap to produce, meaning marketplaces, search results and freelance platforms can be flooded with material that nobody cared enough about to think through properly.
We have reached the rather magnificent stage where human freelancers are now being hired specifically to clean up machine-generated “slop” after machines were introduced partly to avoid hiring human freelancers.
The circle of life remains undefeated.
Most importantly for authors, there is trust.
The Authors Guild launched its Human Authored certification in 2025 and expanded it in 2026. By March 2026, more than 3,000 authors had certified around 5,000 titles, explicitly positioning human authorship as something readers may want to identify.
A separate industry analysis reported that more than 70% of surveyed consumers were less interested in, or unwilling to read, books generated partly or wholly by AI. That does not mean 70% of readers object to an author using AI to research headline options or organise notes.
There is an enormous ethical and psychological difference between using a tool during creation and outsourcing the creation itself.
Which brings us to the more uncomfortable question.
What Happens to Your Own Brain When the Machine Does All the Thinking?
There is a phrase I keep hearing: AI will make us stupid.
That is far too broad, and the current science does not prove that AI causes literal brain atrophy. And if you don’t understand how confirmation bias and distorted thinking can be exacerbated by the bots, please pause and consider that too.
But there is enough emerging evidence around cognitive offloading to make writers pay attention.
In a 2025 MIT Media Lab experiment, 54 participants completed essay-writing tasks either using an LLM, using search engines or using only their own brains. EEG measurements showed the strongest and most widely distributed neural connectivity in the brain-only group, intermediate engagement among search users and the weakest connectivity in the LLM-assisted group. The researchers summarised the pattern rather neatly: “cognitive activity scaled down in relation to external tool use.”
Then came the particularly interesting part.
In a fourth session, some participants who had previously used the LLM had to write without it. They still showed weaker neural connectivity than the people who had been working unaided all along. Those who began brain-only and were then given AI showed stronger recall and broader activation. The researchers also found that LLM users had poorer recall of what they had just written and reported a weaker sense of ownership over the work.
There are important cautions here. The study was small, the fourth session involved only 18 participants, it was an essay-writing experiment rather than every imaginable creative activity and the original paper was a preprint rather than definitive proof of long-term neurological harm. The MIT team themselves explicitly warns against turning the findings into claims that AI makes people “dumb”.
But as a writer, I find the question considerably more interesting than whether AI can produce a nice paragraph.
What happens if you repeatedly stop doing the cognitive work that writing used to force you to do?
Microsoft and Carnegie Mellon researchers approached the issue from another direction. In a 2025 study of 319 knowledge workers reporting 936 real-world uses of generative AI, greater confidence in AI was associated with less critical-thinking effort, while greater confidence in one’s own ability was associated with more. The researchers did not conclude that critical thinking simply disappears; rather, some of it shifts towards verifying, integrating and supervising AI output.
Using a calculator does not destroy your mathematical ability. Never learning the arithmetic because the calculator is always there is another matter.
My Preferred AI rule: Brain, Machine, Brain
When learning how to fix AI writing in your own manuscript, the safest and most effective approach comes down to a simple rule: Brain, Machine, Brain. You think first.
Work out what you believe. Decide what the chapter is doing. Find your examples. Remember your stories. Make the argument. Develop the premise. Wrestle with the uncomfortable bit until you know why it is uncomfortable.
Then bring in the machine where it is genuinely useful.
Ask it to challenge the argument. Find gaps. Generate alternative headlines. Help organise notes you have already created. Compare two versions. Explain a concept you do not understand. Suggest questions for further research. Help you see another structural possibility.
Then take the work back.
Check the facts. Reject the obvious. Rewrite the bland language. Restore your rhythm. Put the peculiar example back in. Add the story you almost removed because it was too specific. Decide what you think.
The machine should leave you with more thinking to do, not relieve you of thinking altogether.
If your thinking is the product, do not outsource the thinking.
That is the distinction I wish more authors understood.
And Yes, Popular Culture Warned Us
Those of us of a certain age may remember WarGames from 1983, in which a teenager accidentally begins interacting with a military computer called Joshua and the situation becomes somewhat more serious than his original plans for the weekend.
Interestingly, the humans do not finally solve the problem by being cleverer at computation. They make the machine confront the logic of the game it is playing.
Then Harry Potter and the Chamber of Secrets gave us Tom Riddle’s diary: a charming little object that writes back, appears to understand you and becomes increasingly involved in your affairs.
I am not suggesting ChatGPT is secretly Voldemort. I personally think that he is one of my exes, but more about that on my personal medium.com feeds 😉 Anyway, that would make technical support far more exciting than it currently is.
Nor are current generative models sentient beings harbouring secret ambitions. They do not “think for themselves” in the human sense simply because their language encourages us to experience them that way.
The useful lesson from the fiction is much less dramatic: something that speaks fluently can very quickly acquire more authority in our minds than it deserves.
Humans anthropomorphise extraordinarily easily.
So use the thing.
Just perhaps do not hand it your frontal lobe and go out for lunch.
Before and After: How to Fix AI Writing Examples
To understand how to fix AI writing in practice, look at how a generic, bot-heavy paragraph transforms when reinfused with human thought. Consider this:
Writing a book isn’t just about putting words on a page. It’s about finding your authentic voice and creating meaningful connections with your readers. Every chapter is part of your journey, helping you grow not only as a writer but as a person. And that matters.
It is perfectly serviceable.
It is also almost entirely made of abstractions.
Now give the same thought to a human author with something particular to say:
Writing a book forces you to discover which of your ideas survive contact with an actual reader. That can be inconvenient. The argument that sounded brilliant over coffee may collapse by Chapter Four, while the throwaway story you nearly deleted turns out to contain the entire point of the book.
Same broad subject.
But now somebody appears to have been present.
Or take the brick version:
You don’t need more words. You need clarity. You need direction. You need courage. And that matters.
I would probably edit that into:
Most struggling manuscripts do not need another 10,000 words. They need the author to decide what the existing 60,000 are actually trying to accomplish.
One contains motivational shapes.
The other contains a thought.
How to Fix AI Writing Without Losing Your Author Voice
If you want to master how to fix AI writing without turning your prose into a sterile academic exercise, do not begin by hunting banned vocabulary. Begin by looking for genericity.
Where have you used an abstract phrase when you could give a specific example? Where have you explained the same point three times? Where is the writing suspiciously balanced? Where have you inserted a neat conclusion because the paragraph felt unfinished rather than because the reader needed one?
Look at rhythm. Five consecutive one-line paragraphs are rarely all equally deserving of dramatic lighting.
Look at certainty. Humans contradict themselves, change their minds, admit exceptions and occasionally say, “I still don’t know.”
Look at vocabulary. If everything is pivotal, transformative and profound, absolutely nothing is.
Look at examples. A real person usually remembers dates, places, objects, conversations, embarrassments, mistakes and oddly irrelevant details. Generic AI examples often possess all the physical texture of a corporate training manual.
And look for your opinion.
Not the position everybody can safely agree with.
The thing you actually think.
That is often the first thing AI smooths away because bland consensus is statistically much safer than a peculiar, defensible human position.
Where I Think This is Going
I do not think AI is going away.
Nor do I think readers are going to spend the next twenty years brandishing detectors at every suspicious adjective.
But for the love of all holy, can we ask that you don’t feed us AI-gen bumpf while pretending it’s yours.
Those of us in the know, well, we know. Turns out my wonderfully weird ADHD brain is better than AI-detection software. I see the typical patterns and stop reading, feeling vaguely nauseated, and irritated at the deception of it all. Would I feel less outraged if you put “AI-created” at the top of your post, article, or manuscript? For sure, but the I might not read it anyway. I can read what the bots think at any time, from the comfort of my dressing-gown.
The tells will change. Models will learn to avoid whichever words become notorious. Writers will copy machine habits. Machines will train on writing influenced by machines. Language itself will shift, which research into academic vocabulary already appears to be documenting.
Detection will therefore become increasingly difficult from style alone.
I suspect provenance and trust will matter more.
In mass content, product descriptions, commodity information, internal documentation and other areas where people mainly want an answer quickly and cheaply, readers may care very little whether a human composed every word.
Books are different.
Particularly memoir, literary fiction, essays, thought leadership, psychology, self-help and expert non-fiction, where the reader is paying partly for access to another person’s experience or mind.
My loose prediction is that we will increasingly see two markets.
One will be enormous, fast, inexpensive, and heavily machine-generated.
The other will place a growing premium on identifiable human authorship, personal provenance, real experience, unusual voice, and credible expertise.
The current anti-AI movement will probably soften in some places as the technology becomes ordinary. But I doubt the underlying demand for human connection will disappear. The Authors Guild’s certification programme, current consumer research and the continued importance readers place on human recommendation all point towards human-made becoming a meaningful signal rather than an obsolete distinction.
The argument may eventually stop being:
“Did you use AI?”
And become:
“Where is the human?”
That is a much better question.
At Harvard Ink, our editing approach isn’t about running witch trials over whether a manuscript used ChatGPT. We teach authors how to fix AI writing so their personal voice, lived experience, and sharp arguments remain at the centre of the book. I am interested in whether it still sounds like you.
If AI has helped you brainstorm, research, organise, or get unstuck, that is your choice as an author. The more important question is what happened before and afterwards. Did you decide what you wanted to say? Did you interrogate the output? Did you verify it? Did you rewrite it? Did you add your evidence, experience, humour, discomfort and judgement?
Or did you type a prompt, copy the answer and move directly to Chapter Two?
Because readers may not always identify the machine. But they can often feel when the person has gone missing.
At Harvard Ink, developmental editing is not simply about fixing structure, and line editing is not simply about improving grammar. Much of the work is about protecting the author’s intellectual and emotional fingerprint: stripping away generic language, recovering natural rhythm, finding the real argument under the respectable-sounding fog and making sure a book contains enough of the person whose name is going on the cover.
Use AI if it helps, but use it intelligently and sceptically. Keep your own brain involved before, during and after the process.
The machine can produce the words. Your job is to make sure the book still contains a mind.
If your manuscript has begun sounding suspiciously as though it has been on a transformative journey through a rapidly evolving landscape, and that matters, we can probably help.
Book a free chat: Harvard Ink’s consultation calendar Find out more: Harvard Ink
Thank you for reading Part 6 of our series – The Indie Author’s Publishing Survival Guide. To read our other posts on self-publishing for authors and writers, they’re right here:
Read Part V of the series here.
Originally published on LinkedIn September 5 2026. https://www.linkedin.com/pulse/indie-authors-publishing-survival-guide-did-you-write-bot-emese/
