We're delighted to welcome published author and Marketing for Authors guest blogger Ben Jeapes, who joins us this month to explore a question that's becoming increasingly important for writers: should we be using AI? In this thought-provoking blog essay, Ben examines AI's impact on creativity, copyright, and the wider publishing industry, asking whether AI's convenience might come at the cost of human creativity.
Writing is a lot of hard work, and writers by and large are very receptive towards anything that will make the task easier. We can’t all be Philip Pullmans or (heaven forbid) Jeffrey Archers, toiling lovingly over multiple handwritten drafts. Forty years ago, and to my roommate’s relief, after banging out one and a half novels on my old manual typewriter I fell on my Amstrad PCW8256 with cries of joy. For younger readers, that was Alan Sugar’s Model T Ford of computers: a cheap, easy to use, demystified word processor for every home, with added computing thrown in.
Since then, technology has continued to lay helpful tools in our path. Spell checking, change tracking, the ability to open the same manuscript in different places, storing keystrokes as macros, speech-to-text dictation…
And now we have what is surely the apotheosis of all this, a tool that can effortlessly help us sort our thoughts out, get our drafts straight, come up with plot ideas and even just write the darn thing while we have a cup of tea. Instead of staring at a blank screen for three hours straight, waiting for inspiration to strike, you might think: can’t I just use AI?
Well, of course you can.
Whether or not it’s a good idea is another matter.
Plus or Minus
US agency Gotham Ghostwriters conducted a survey of its members on the impact of AI on their profession, and the report is well worth reading by any writer for both the positive and negative impacts At first sight, the results are counter-intuitive and apparently contradictory. 92% of writers who use AI said it made them more productive, 59% said they produce higher quality results, and the professionals typically earn $47,000 more than others.
At the same time, nine out of ten are worried about AI errors, and 81% were concerned about piracy.
There seems to be a discrepancy here. Let’s look at both those positives and negatives.
Generation Game
AI’s greatest strength and greatest weakness are the same thing: it does not follow a list of either/or instructions like a conventional computer program. AI is trained. The developer shows it existing data, and says, ‘Be like that.’ The AI then studies that data, observes the statistical links between the various items, and works out its own rules to create what is called a Large Language Model, or LLM. The result is that the LLM is far more flexible than a conventional, rules-based program. Think how limited any conversation would be if you could only answer ‘yes’ or ‘no’, with not even a ‘perhaps’ or ‘maybe’.
However, as statistics are just another branch of probability, the best an LLM can say is that its output is probably right. It has no way of discerning whether its statements are correct or any good; they’re just sums, the product of an equation. It strings data together based on what it thinks is probably the right order. This is why AI products like ChatGPT or OpenAI or Claude are properly known as generative AI. A conventional program, run the same way under the same circumstances, will always give you the same answer. AI generates each response individually for you, based on the data it currently holds.
“EASY READING IS DAMN HARD WRITING” – NATHANIEL HAWTHORNE
If you want an AI to be running a smoothly working machine, say, or developing new drugs from a mass of biomedical data, finding order amidst masses of confusing and contradictory inputs, this is extremely helpful and has proven benefits. Put another way, it works, and has been shown to work.
But it is the first and best reason for why AI should not be used to write creatively. Because it doesn’t create, it copies.
For some purposes, this is fine. Say you want to write a job application letter or CV. Despite what HR departments would have you believe, there is no such thing as a truly excellent job application letter or CV. There are merely plenty of instances of bad ones. Good ones are all good in different ways: one good example is just as good as another and the stylistic differences are immaterial. So, on that basis, getting AI to write your job application may be absolutely fine. (Unless you’re applying for a job that depends on your creativity, but then, you’ll soon be found out once you have your foot in the door, so probably best not to in the first place.)
But that is not creative writing. Let’s talk stories, plays, novels, where there is such a thing as excellence, and where it matters. At this stage of your career, you may struggle to write prose that shines in the literary heavens like an exploding star. But there is still nothing to say you won’t one day stumble upon exactly the right form of words to say what you want to say. And when you do then – believe me – the adrenaline rush, the glow of satisfaction, the sheer joy in doing what you do will be unbeatable.
Getting AI to write for you will never achieve that. It will at best produce a workable, workmanlike approximation of what millions of other writers have done before you. (That number is not an exaggeration, as we will see later.) It is presenting you with the output of an equation. A statistically generated response that is the kind of thing you will probably like. It does not deliver excellence, because it cannot. It can only deliver averages drawn from its data set, and by definition an average is not excellent.
But supposing I train AI on my own writing, so that it gets an exact ear for how I write? I can then give it my notes, my ideas, and tell it to write it for me. And it will, exactly like I usually do.
You know who else writes exactly like you? You do!
Yes, AI can duplicate exactly what you normally write. But in that case you will never excel. You will never develop as a writer. Every book you write – rather, every book that comes out under your name – will be frozen at exactly the same stage of your career, a variation on whatever you trained the AI on. You will only ever write the kind of thing that you usually do.
That might not bother you, but why should any reader be bothered to read something you couldn’t be bothered to write?
Okay, I’ll do the writing and get AI to edit for me …
And the same applies. By definition, it will just give you an average job, probably missing every point you’re trying to say, with a total tin ear for any clever turn of phrase you might use, every subtle joke you make.
But I heard about this guy …
A neurodivergent man called Jonathan Kemp wrote an article titled ‘Without AI I wouldn’t be a writer’. He has found the technology essential in getting the words down.
Fair enough. I am fully able-bodied; I do not need a crutch to assist me; therefore if I did use a crutch to help me walk, it would be pointless and lazy and possibly counterproductive as my leg muscles weakened through lack of use. If however I did have a disability and used a crutch, then I would be walking, and no one would have the right to call me not a real walker.
At the same time, note that even in Kemp’s case, he does not use AI to create. Just to assist. The original ideas, the concept, the vision come from his Mark 1 human brain, and that is the minimum definition of creativity. Your crutch should not replace your limbs, merely assist them to do the best they can.
If you don’t actually need AI then research recently reported in The Independent suggests that ‘Using AI for even 10 minutes significantly reduces your ability to stick with difficult tasks’. Naomi Alderman said it best in The Author, Spring 2026: “If you’re a writer, you think by writing. And thinking is the job. If you let your brain get used to the idea that there’s a shortcut, it will get progressively harder to come up with new ideas, thoughts and observations. The muscles just atrophy.”
Point taken. Now, what was that about 81% being worried about piracy?
I’m very glad you asked me that, and frankly I’m amazed the number is so low …
It’s a Steal
I’ve said an LLM is trained. Where does that training data come from? Despite the second ‘L’ standing for Language, an LLM’s data can be anything. For an LLM to write music, you train it on music. To draw pictures, you train it on art. To talk, chat, be just like a human being, you train it on the most widely available data that contains human interaction, emotions, speech patterns, thought processes and behaviours: in other words, books.
There is absolutely no reason why, in principle, said music, art and writing can’t be acquired from their content creators fairly and legally. To date this has very rarely been the case. AI developers have historically displayed very sticky fingers when it comes to other people’s intellectual property. Put more bluntly, they have stolen it, sending their bots out into the web to locate and download copies of, well, just about anything they can find – a process known as scraping – without due care and attention to who actually owns it.
I can say this without fear of libel because it has been established as fact in a court of law: specifically, the Californian court that heard the landmark case of Bartz v. Anthropic. This was a class action by authors Andrew Bartz, Charles Graeber and Kirk Wallace Johnson against AI developer Anthropic, for using their illegally downloaded books to train its AI product. Other authors piled in and Anthropic settled in advance, coughing up $1.5bn to pay to the authors of 440,490 pirated books.
Even that was not the huge win people hoped for, because those books had to have been published in the US within a certain period, and the number barely scratches the surface of the number of stolen books. The court in California found that Anthropic had acquired at least seven million copies of books which it knew had been pirated from so-called ‘shadow libraries’, the main offender being one called LibGen. Imagine that each of those books had an average cover price of $10, do a bit of mental arithmetic, and work out how much money simply vanished from the publishing industry as a result of this piracy.
The Atlantic magazine bought a copy of the pirated data used by Anthropic, and kindly turned it into a searchable web database so that authors can look themselves up and see if they’ve been pirated. Have fun searching it for your own name.
As of June 2026, there were 92 similar cases ongoing in federal US courts, and the message for the techbros is sinking in: you really need to do this the right way.
Why This is a Problem
But Ben, how is this different to you buying a book from a secondhand bookseller? The author doesn’t get anything from that sale either.
Actually, not necessarily. The UK has AuthorSHARE, a scheme run by ALCS – the Authors’ Licensing and Collecting Society – that pays ALCS members royalties on the sale of their used books bought from online secondhand booksellers.
Even without that, when I buy a secondhand book, I know that somewhere in the process of that book’s creation, the author and publisher were paid. Pirates make copies that are not paid for, which is otherwise known as stealing. And even if the book is secondhand, even if a book is out of print, the author still owns the copyright, unless they have died over seventy years ago.
Copyright is intellectual property, the key word being ‘property’. It gives you certain rights that cannot just be ignored: for example, the right to recompense when your property is used, and a say in what that use might be. In both the UK and US, copyright applies the moment you write something down. Even if it’s just a shopping list: you wrote it, it’s your copyright. The only difference between the two countries is that in the US, the work must also be registered for copyright protection to kick in, which is one reason the Anthropic settlement only applies to a limited number of titles. But still, just taking something without payment for a purpose the creator has not approved breaks at least a couple of laws.
This Side of the Pond
The Americans are tackling this with law cases, because due to differences in our legal structures, getting together in class actions against monolithic empires is easier over there. Over here it’s more about getting preventative with the law in the first place.
At first the UK government couldn’t see the problem. They want the AI developers to be investing billions in our economy, so they want to be as welcoming as possible and not make them start paying. Eyes were opened when the Society of Authors and other interested parties sat down with the government and had a quiet word in their ears. They reminded the government that, in 2024, the creative industries contributed £146 billion in gross value added to the economy, a sum that depends on: created works actually being bought and paid for; taxes being paid on royalties and rights sales; the general circulation of money. In other words, predicated on the copyright system operating as it’s meant to. And it’s the kind of figure you notice if it suddenly vanishes from your economy.
The government’s first response was to try to find a middle way, suggesting that copyright material could be scraped for LLMs, but authors could opt out of this if they don’t want it to happen. This failed because why would and should anyone want to opt out of an automatic legal protection? Having to opt out of someone taking your intellectual property would be like having to opt out of someone taking your actual property. It would be like burglary remaining illegal, but unless you put up a sign on your lawn saying ‘No burglars’, you can be still burgled with impunity.
The UK government has now conceded that opting out isn’t the answer. On 28th August it said: “The UK is home to world-leading creative industries and a growing AI sector. The Government does not believe we need to choose between them. We want an approach to copyright and AI that supports innovation, rewards human creativity, builds public trust in AI, and provides the certainty needed for both sectors to grow.”
But it’s not saying what the answer is, so I can’t yet report a happy ending.
Dangerous Visions
Where does this leave those Gotham responders who spoke favourably of AI?
It would suggest AI might have a place. Those responders who found it was doing them good weren’t using it for actual writing but for everything that goes with it. All the tedious grind of being an author that does not involve the actual creativity: admin, research, marketing.
All of which are viable options for using AI … If you’re careful.
Remember, when you ask AI a question, it doesn’t delve into a deep database of facts to produce an answer. It merely looks like that, because it presents its answer with a brash confidence that brooks no argument. In reality, when asked a question it makes a calculation, ‘Based on my training data, what would an answer to this question probably look like?’ and provides that instead. Which means it can probably get it right, first thing – but equally it might just make something up and blithely present it as fact. Which is, to say the least, unhelpful.
Recent cases of AI hallucination to have made the headlines include making up fictitious case law that lost a barrister his case; advising the West Midlands police of a non-existent football match that led to them banning fans of an Israeli football team; and assuring disgruntled airline passengers that they were entitled to refunds when in fact they were not.
AI might produce a whole raft of facts and figures for you, in the blink of an eye.
Assume it’s lying, and check them. Ruthlessly. It’s only a machine so it won’t mind. If it starts to sing ‘Daisy’ then just ignore it.
Rage Against the Machine
So, I see it might have its uses but I’m also part of the surprisingly low 81% worried about piracy. What can I do?
For a start, you or your publisher can insert text in your copyright statement that clearly forbids LLM training unless you say so. A bot might just plain ignore this, depending on what it has been told to do, but it makes your case stronger when it comes to court. And one thing bots are good at is following instructions, so a bit of text might (might) be enough to deter it.
Penguin Random House has recently made a blanket declaration that AI companies are prohibited from using authors’ copyrighted work to train AI models without permission. It’s a good start and hopefully the rest will follow.
PRH also makes a general statement on its attitude to AI on the Penguin website. It’s interesting to note that this includes the fact that PRH has been using AI of one kind or another for a decade or more: for example, in developing machine learning tools for sales forecasting and stock management. Another example of a valid use for it, but I bet a human takes responsibility for the implementation.
If you have a traditional publishing deal then you can ask for an AI clause to go in your contract, saying similar. And you could have a clause regarding translation rights, forbidding the publisher to use AI to translate your book into a foreign language and require a human to be used instead. (I haven’t yet touched on the use of AI for translation but it is another field of human expertise that is under attack.)
Meanwhile the Human Authored scheme is free to enter and already has thousands of titles registered. If you can show your book is 100% human-authored, then you get a logo you can put on the cover or the copyright page. Some would say this is still a sticking plaster and we would much rather have AI content clearly labelled in the first place, but it is a good start.
There is also a petition to ask the Government “to introduce a statutory duty requiring AI developers to disclose the copyrighted works used to train generative AI in enough detail for creators to identify use of their work, enforce their rights, negotiate licences and be fairly paid.” The deadline for signing is 29 October 2026.
So, What is the Answer?
In due course, I strongly suspect the AI and creative industries will find a happy medium, with the former paying licensing fees to the latter for the use of its content. This has happened before, at the turn of the century, when the Wild West of music file sharing eventually settled down to the tamer, more regulated and somewhat fairer scenario we have today, where it is just as easy and a lot safer to download music for a small fee.
ALCS was set up to tackle an even earlier copyright problem caused by the introduction of photocopiers. Every year it now distributes thousands of pounds to authors based on licensed copying of their work. And in September 2026 ALCS for the first time paid out to authors whose work has been used for licensed LLM training.
But until then, whatever your feelings about using a product predicated on mass theft – and I would like to think you’re against it – until the licensing issue is sorted out, any AI-related product will have a massive legal question mark hanging over it. Many companies – Amazon, for one – simply won’t touch anything associated with AI because they have better things to do than be embroiled in a copyright dispute. At the moment, copyright issues are a sword hanging over every AI developer. The onus is, or will be, on them to prove that their AI’s training data is ethically sourced.
And Speaking of Copyright…
So, I’ve found an AI product that openly declares its training sources and can show it’s paid them. Is that OK?
Probably, but take care. If you do use an AI to produce something, who owns the copyright? It won’t necessarily be you. Remember, under UK law, anything you write is automatically copyright to you the moment you commit it to a permanent medium: a sheet of paper, the back of an envelope, a computer’s hard drive, the cloud … And if you are writing on someone else’s behalf, then unless there is a contractual arrangement in place, that too is your copyright.
By the same logic, anything that ChatGPT or its ilk coughs up at your behest is copyright to ChatGPT, unless you already have an agreement in place.
That’s the output. Related to this, what does the AI do with its input – in other words, your questions and instructions to it? LLMs learn from human-generated content. Your query to the AI comes into that category. Does it use that text to learn from?
This is not such an issue if your query is simply “write me a shopping list”. It does matter if say you are inputting a chapter of text and want AI to tidy it up for you. Free, human-generated, automatically copyrighted text handed to it on a plate; it won’t look that gift horse in the mouth, and may result in your output being contaminated by copyright ambiguity.
So, if you are able to find such a rare bird, make sure the deal includes a cast iron contract about copyright and ownership.
It’s Not Easy Being Green
A further issue to consider, which again may or may not bother you as an individual (though I hope it will) is AI’s environmental impact. Which to be blunt is horrendous. Modern AI requires massive amounts of computing that takes place in massive data centres running 24/7. The components alone require vast amounts of rare earth elements, precious metals and base minerals, all of which must be mined. And then you switch them on. In Ireland, data centres now eat up 22% of the nation’s electricity. That electricity generates heat and needs cooling: in the Netherlands, Microsoft’s data centre guzzled 84 million litres of drinking water, at a time when the country was facing water shortages.
It could fairly be argued that all those resources will be consumed anyway, so not using AI would be like not flying because of the environmental impact of the air travel industry, even though those airliners will take off and land with or without you. And these points are not insuperable. As we learn to generate cleaner energy and reuse waste energy more efficiently, the problem may recede. But for the time being, this is an issue.
Unite and Unite and Let Us All Unite
Phew, you had me worried, but I now see that if I can find an AI product that is trained ethically on legally acquired, licenced data, running in data centres that are powered by green energy sources, and if I take care to double check it at every stage, then that’s fine, right?
Actually, there is more. You’re a hard-working professional and you expect to be compensated for your efforts. Just how delighted would you be to be dropped in favour of unpaid, and hence cheaper, AI?
The Society of Authors recently surveyed its members on the effects of AI. Findings include:
86% of authors say generative AI has already reduced their earnings.
72% of authors say job opportunities have already been cut.
57% of authors say their career is no longer sustainable.
43% of literary translators and 37% of illustrators saw earnings fall due to generative AI
26% of illustrators and 36% of literary translators report cancelled or redirected commissions.
So even if your attitude is, ‘I wouldn’t use AI to write but might to design my cover or typeset, or do my marketing,’ then you may just be kicking the can down the road. That use of AI is replacing a human agent, and one day that human could be you.
Not Fade Away
If you take a photocopy of a photocopy of a photocopy … Quality degrades over time and you just get a mess.
To repeat, AI does nothing more than copy and imitate, and a copy is always less than the original. Don’t blame me, blame the Second Law of Thermodynamics. So, while AI continues to learn from human-created examples, it will at best produce perfectly competent work with nothing to distinguish it from any other perfectly competent work. If AI content increases and human content decreases, then the AI content will learn more and more off other AI content, and the quality will spiral ever lower into sludge. AI training requires a steady flow of quality data – in other words, data generated by humans.
This is a marketing blog so let’s use that as an example. Marketing yourself as an author is a daunting task. You need words written, publicity material designed, and a strategy to pull it all together. It looks like hard work and quite possibly expensive too. So yes, an AI might be able to do all that for you. But everything it then does – the material, the campaign – will then be itself used as training data. You will get the average of all the other campaigns the data has trained on, and every campaign after that will get the average of all those plus yours. For the campaign to be any good, a human needs to be involved, to apply their own insight to its average-level suggestions and make them excellent.
I Couldn’t Be Bothered to Read Any of That And Just Jumped to the End. So, Can’t I Just Use Ai?
It’s a free world.
If you’d read all the way through to this point then hopefully by now you would have concluded that an ideal AI:
Is trained on licensed, paid-for data.
Lists these training sources.
Does not reuse your input to train itself.
Has a user agreement that protects your copyright at all stages of the process.
Runs on carbon neutral clean energy data servers.
If you can find an AI that ticks those boxes – and I think the day is approaching when there will be such an animal – then AI might have its uses. But you might still be using it to replace genuine human effort, and one day it might come for you too. Use your own discretion.
Don't use it for anything creative, because
It doesn’t actually create.
You could land yourself in copyright trouble.
You will be complicit in the problem already facing other creatives.
If you use it for research, check the answers yourself before using them.
If you want imitation, by all means use AI. It might be good enough for what you’re after.
If you want excellence, go with the human.
Acknowledgement
Thank you very much to Gary Dalkin, editor and writer at To The Last Word for helpful comments on an earlier draft.
Ben Jeapes
Ben Jeapes is the author of five professionally published novels and three self-published. Quite by accident and to his considerable surprise, he became a full-time ghostwriter in 2015, and has now written over a million words of mostly fiction for various clients (which is his answer to anyone who says he isn’t a real writer). His non-fiction book Ghostwriting Novels: A Guide For Ghostwriters And The Ghostwritten was published in 2024.
He lives in Abingdon and his web page is at www.benjeapes.com.