What You’re Risking by Using AI to Help You Write
Should you be using AI to write for you? Image by Ekaterina Bogdan on Unsplash+.
I used a machine to write the piece you’ll see below, but it’s not what you think. Leaving my phone and laptop behind, I went to the coffee shop with my typewriter and wrote without accessing the Internet at all.
While writing, I did not consult Claude or ChatGPT, did not search anything on Google, did not go online. I didn’t look on LinkedIn or the website of my friend, the brilliant writer and thinker João Sevilhano, even though I intuited he was also germinating something on writing and thinking in an age of AI. Separated from my books, I also didn’t refer to anything on the printed page.
Instead, it was just me and my mind. What I typed at the coffeeshop had the quality of a ‘Morning Pages’ exercise, a technique made famous by Julia Cameron of The Artist’s Way. Morning Pages involves writing several pages at a stretch, ideally longhand and first thing in the morning, before the world has too much of a chance to intrude. You write without interruption, without censoring yourself, and without editing.
I didn’t finish that day, but picked up again when I had a bit of leisure, on a lazy Easter Monday. When I’d extinguished my train of thought, I resolved to do something that frightens me in this age of perfection, optimisation, and literally conforming to type: I would publish my spontaneous writing session here, in all of its imperfection.
I would not submit the text to Claude to ask it to provide me H1 and H2 headings, optimised for Google’s search engines.
I would not ask ChatGPT to suggest keywords.
I would not beg any AI’s assistance in generating the ‘Key Takeaways’ and ‘Frequently Asked Questions’ sections, which apparently make this post ripe for a ‘crawl,’ the creepily apt term for what search engines do.
However, because I know that you, the human reader, are unlikely to see this piece unless the machine reads and approves it first, I am still going to write those sections. I’ll decide on the keywords, compose my own meta description, and formulate key takeaways and FAQs.
I’ll do all that, and I’ll resent every minute of it, for reasons the piece below will make clear. As always, if you’re in a hurry and not in the mood to slow down as an important and even noble act of resistance, here are some key takeaways.
Key Takeaways
As we’re increasingly using chatbots like Claude, ChatGPT, and Gemini to write for us, or at least help us to produce and edit our writing, we lose skill, self-confidence, self-trust, and humanity, without clear evidence of gain.
The justifications for using machines to write for us include efficiency, optimisation, productivity, clarity, accuracy, and correctness, all of which are machine-centred goals rather than human-centred values.
Conformity, sameness, efficiency, speed, and productivity have not always been human goals. The Industrial Revolution’s devastating, dehumanising impact continues to resonate today.
Writing is an expression of thinking, a product of thinking. Writing is a process, as thinking is a process, often messy and difficult. Idiosyncratic, imperfect, passionate writing displays an individual’s personality and way of thinking, in a way that machine-generated writing cannot. Outsourcing writing means ceding something of your humanity to a machine.
With the rise of large language models and natural language processing, we watch in fascination as machines become more human. The thing we should be keeping an eye on is humans becoming more like machines, which those same machines are nudging us to do.
When the skill of thinking erodes, so does our freedom. Hannah Arendt warned that the loss of thinking results in the ‘banality of evil,’ as with Adolf Eichmann, who was ‘just following orders.’ There may be more of a relationship than we realise between the banality of AI-generated writing and the banality of evil.
Preserving independent writing is essential to safeguarding our humanity. We must resist.
Book Me As a Speaker
As you’re reading, keep in mind that I speak about these things. If I appear at your event, you will not get an out-of-the-box spiel, because I am incapable of it, for better or for worse. You will get passion, deep thought, unexpected connections, lateral thinking, and human spirit. AI will not have written my script.
I speak about AI at work, digital afterlives, online companions and AI relationships, and overall digital wellbeing.
For adjacent pieces:
It’s Become Scary Not to Use AI for Writing
A reminder, before we get into this.
I didn’t write or edit this with reference to the Internet or indeed any other written source. I wrote it fuelled only by coffee. One stretch of writing took place in the corner of the coffee shop across the street from my house, the only books in sight being cookbooks piled on the long wooden table near me and a hipster children’s book entitled Baby’s First Eames. The other writing stretch was in a sunny front room on Easter Monday.
I am embarrassed to publish my stream-of-consciousness, unedited ‘Morning Pages’ below, and I am prepared and even happy to tolerate that discomfort for the sake of embracing and celebrating my humanity.
That is the point of the exercise. I hope, by the end of this piece, you might feel inspired to think weird, write weird, and publish weird yourself.
Why I’m Writing Without AI Today
Writing to machines
Writing for machines
Writing by machines
Writing like machines
Writing like the dead
Zombie writing
The screen is on, but there's nobody home.
First things first:
I have no skin the game, this is not a paid promotion, so I am writing this by way of explanation, not with any other agenda in mind. I am typing on a hunter-green-and-chrome typewriter with a strong stylistic resemblance to my old green Olympia, but I can charge this one up. My ramblings can be beamed to a postbox in the cloud, if I manually toggle on a switch to enable that. There is nothing incoming, though. Nothing will check my spelling, correct my syntax, tighten or smoothe.
I just thought of Stockard Channing singing 'Look at Me, I'm Sandra Dee' in Grease. Claudey, Claudey, let me be, keep those edits far from me...
I don't type on this beautiful object because I'm so starved for human-to-human interaction that I'm willing to engineer twenty conversations in the coffee shop starting with, 'Excuse me, can I ask what that is' (although maybe that's part of it).
No, this is my new discipline, dead lifting for cognitive muscles threatened with atrophy, free weights for the encroaching self-doubt, the creeping mistrust that causes me to forget myself.
In middle age you're meant to keep working out, or else you'll become like the elderly folks in the infomercials of my youth, crying out, 'I've fallen, and I can't get up!'
I have written my whole life. I typed, incorrectly and prolifically, from the time I was about 3 years old, or so my parents tell me. I was ever a writer. And now, I fear falling into some void, AI slop lapping the edges of it like so much raw sewage, and not being able to crawl back out.
We’re Heavily Incentivised to Write Like Machines and For Machines
A while back, I realised that my website was virtually unfindable.
At the beginning of Hitchhiker's Guide to the Galaxy, Arthur Dent is remonstrating with Mr Prosser, the council dogsbody who's come to supervise the knocking down of his house. Mr Prosser says that the planning documents have been publicly displayed for some time…in a dark basement with broken stairs, in a locked filing cabinet in a disused lavatory, with a 'Beware of the Leopard' sign tacked to the door.
This was roughly similar to my website's visibility, and the fact that I'd ever been booked for a speaking gig or contacted at all was a mystery.
So I engaged an SEO professional to help lift me out of the ultimate modern ignominy, which is to be invisible online. What this involved, partially, was her assistance in generating SEO/GEO optimised blog posts clustering around my areas of expertise, rife with keywords ripe for Google's picking.
I didn't know what GEO stood for before I worked with L. GEO stands for Generative Engine Optimisation. SEO involves getting humans to click on your site once they've found you on Google or similar, and GEO involves getting mentioned or referenced in AI-generated answers and summaries.
Both acronyms, ultimately, add up to the same things. How do I write in a way that machines can understand? How do I organise my site and my blog posts so that machines can spot and interpret and index them? How do I get a machine to see me and to validate my worth so that I have half a chance of encountering other humans who might then also see my worth?
Christ, it feels funny to write that down.
There have always been systems to game, middlemen to court, hoops to jump through. To pass her English Language GCSE, my 16-year-old learned long ago that it doesn't pay to be too creative. Instead, you conform to the mark scheme, use the expected words, write the structured paragraph according to the prescription so that the marker awards the mark.
Before social media granted a potentially global platform to anyone with an Internet connection, you had to be endorsed by the gatekeepers (journalists, agents, editors, publishers, DJs, record labels) to get any kind of visibility.
If you wanted to work in corporate, you had to talk the talk, wear the suit, maybe have the MBA or whatever other degree from the right place, have the letters of reference from the person in power.
And sometimes, there would be people who didn't fit the mould.
The writer who wrote the weird book that didn't observe traditional forms, didn't fit easily into a genre, but an agent with connections championed them, followed by a publisher who was willing to take a risk.
The student who wrote the wacky essay that barely responded to what the question asked but that was a work of budding genius, and the teacher paid attention.
The person from the mailroom (classic movie and TV trope that this is) who worked their way up into the halls of power through sheer grit and persistence and because someone saw something in them, because some human saw something in them.
Very little in recent times has felt as soul-destroying to me as writing in a way that is recognisable and interpretable and tantalising to the machine. The machine does not reward weirdness. The machine does not care for idiosyncrasy. If you flummox the machine, it will ensure you remain invisible.
I used to have a T-shirt from my home town that showed support for independent local business. It said keep louisville weird. What t-shirt should I make now?
I was thinking something literally global, like keep the world weird, and I might yet do that. But more narrowly, and for the purposes of this morning's stream of consciousness meditation, I might first choose keep writing weird, or keep thinking weird.
When my current GEO/SEO advisor L. drafts a template for an 'optimised' piece about AI at work, or digital afterlives, or digital wellbeing, or online relationships, it is carefully calibrated. The heading and subheadings are clear rather than creative, a bright-red neon sign for Google and the LLMs on what my piece is about. The right keywords are present. Everything is clear and optimal and efficient and soul-destroying, numbingly boring.
My voice and expertise and any unique points I have ever made about the topic are nowhere to be seen in these pieces, and I don't let them go out into the world like that. I couldn't, anyway. I know I am writing thought leadership pieces alongside the 'cyberpsychology basics' pieces that are started by L. and finished by me, and I try to comfort myself with this, but I can't stand the split, the schism, the multiple-personality thing of two different kinds of posts, one written for the machines, one written for the humans.
I don't know how to reconcile this, so I attack each 'optimised' piece with the zeal of a compulsive DIYer trying to refurbish the house from top to bottom while not destroying the foundation or the structures that keep it standing, that keep it recognisable as a house. I'm trying to preserve the SEO-ness and GEO-ness of the piece, the machine-readable architecture at its heart, while reclaiming it as my own.
The process is laborious, even with assistance, and I feel uneasy about it. Ironically, and in a way that's frustrating and not even working very well, I'm using Claude like Dr Frankenstein used electricity, to try to animate something and give it the breath of life, human life, my life, but that something is a patchwork monster.
'Here,' I say to Claude. 'Here's this stupid (no offense to L., this is not about her, she is doing her job) optimised piece. Here are two chapters from my books about this topic, and two pieces from Substack or Medium, and here's the transcript of a podcast where I talked about this, and here's the transcript of a recording I made while I was talking to a journalist, and here's the text of a keynote I gave. Can you take all these things and make a piece that's mine while preserving the SEO architecture?'
Sometimes the result is sort of all right, it's at least a passable start, but it's still not MINE and it's still not ME and it's exactly what my analogy suggests, it feels like a Frankenstein's monster, and then it all starts to feel like a metaphor for my entire life, which revolves around being weird, and embracing that weirdness most of the time, but needing to make myself and my writing accessible to other people.
The thing about trying to accomplish this paradoxical exercise is that it takes hours, at least how I'm doing it, and I fight with Claude, and I call it stupid, and then I wonder why I'm not starting from scratch myself, which would be so much easier and something I know how to do, and then I remember: Oh yes. I don't know how to write for machines.
That's a lie, actually, or more charitably, that's not true anymore. I have, through working with L., learnt quite a lot about writing for machines. The truth is, I don't WANT to write for machines. And if she doesn't help me out by getting me started, by giving me a basic skeleton, I won't be able to make myself write for machines. And then I'll stay invisible and never get any work.
And the messed-up thing about it, the thing I almost resent, is that it's working. Even with all my interference with every engineered Cyberpsychology Basics post, it's working. My impressions have increased by something like 4000% in a few months. I am being rewarded and reinforced (visibility, engagement, speaking gigs, income) for writing in a way the machine understands. The insidiousness of this troubles me, like I have met the Devil at the crossroads, or someone who is considerably less interesting than the Devil, and have struck a bargain with him.
And yet, being compelled into machine-readable writing for a mercenary purpose is not what troubles me the most. What troubles me the most is other behaviours, micro-behaviours I am performing in everyday life to do with writing and editing.
We’re Not Using the Writing Skills We Have, Even When We Need Them
In high school, I was fortunate to be part of a tradition that is probably waning but never more needed: an American high school journalism programme. We were taught by the inimitable Tony Willis, who would go on to earn the title of Indiana Teacher of the Year and whose face appears on the Wall of Hometown Heroes near the bridge in Jeffersonville. My friend and I were joint op/ed columnists.
The classroom felt like a proper news room; when we were investigating something, we felt like Woodward & Bernstein breaking Watergate. On one hand, that's youthfully sweet and ridiculous. On the other, in our passion and fervour for finding, writing, and publishing stories, we had a lot in common with them, bless us.
We were trained how to write pithy headlines, which we physically laid out on a table as we organised our pieces into column inches to fit the page, and we were taught Strunk & White's Elements of Style, a classic writing manual we used to produce leaner, more elegant prose, the principles of which I have never forgotten.
I recently fed those principles into Claude's preference settings to try and get it to stop annoying me, to no avail. It still torments me with excess adverbials, saying that our ability to write and think is being 'quietly eroded.' I lose my shit and type, 'As opposed to loudly eroded? Check your rules,' and then it apologises and says it should have caught that, and it won't happen again, and I fire back that it bloody WILL happen again, it happens all the time, and it gives me some more stupid excuses and says next time it will be different, and then it isn't, it's always 'quietly,' and every time it says 'quietly' I get louder.
And then I think, hold on a second. Why am I getting into this debate with Claude? And I become more discomfited still, because I've realised that instead of editing my work, which I have the skill and experience to do, I am yelling at Claude to edit the way I would. When it fails, as it inevitably does, instead of reverting to what I want and what works for me, I spend pointless time and excessive energy lecturing an LLM like it's an inattentive assistant that I'm hoping to train up for a role as senior editor.
I don't even know why I'm doing this. Have I bought the myth of efficiency so completely that I can't take control again when something inefficient is happening? Have I swallowed the idea that AI will make my writing better, so much so that I give it the chance to help me write more and more of the time, despite repeated experiences of AI making my writing worse, or at least, less human, less me, flatter, more boring, all of which, to my mind, is definitely worse?
I don't want this Kool-Aid, and I see behind the lies the guru is spouting, and yet somehow, too often, I am drinking the Kool-Aid, and I am trying to understand why.
Our Turning to AI Writing Tools Has a Parallel with the Privacy Paradox
I'm reminded of a concept called the privacy paradox. The people who are most aware of potential threats to their privacy are the ones that are most likely to engage in privacy-compromising behaviours online, even if they have had negative experiences in the past. It's as though they think that awareness itself will protect them.
There is a similar paradox when it comes to 'sharenting,' the sharing of children's data by parents and carers. The parents who are most educated and aware of how the online world works are the ones most likely to share their children's data.
Here is another paradox. Everything about writing with LLMs frustrates me, and yet I find myself doing it on occasion, usually with the result that I spend hours more on something that I typically would, perhaps with a less satisfying result.
Do I think that my awareness protects me? That because I understand something about them, because I am aware of the risks, I can use LLMs without fear I will be drawn in, that I’ll be able to use AI writing assistance without risking my thought, my creativity, my writing and editing skills?
The human brain is susceptible to efficiency, or promises of it, without question. Brains are lazy. Given a shortcut, a brain will often take it, unless a passionate interest in what one is doing keeps a shortcut from being attractive.
I remain passionately interested in both the writing process and the subjects about which I write. And yet the shortcut or promise of one sometimes hooks me in. Why?
We’re Using AI to Write Because We Mistrust Ourselves
I have written about this elsewhere, and I cite it here again: mistrust.
We are rapidly coming to thoroughly mistrust ourselves, across the board. We’ve developed an obsession with getting things exactly 'right,' and we ask the machines to help us to that.
Right for what? Right for optimally influencing other people? Right for achieving maximum visibility? Right for making ourselves immune to challenge, for winning an argument, for being unassailably correct?
The type of 'rightness' differs, but too often our drive to be right, and our tendency to consult a machine, is connected with our modern obsession with efficiency, optimisation, and productivity, which are of course machine goals. And in blindly following this impulse towards rightness, in believing in its necessity, we come to mistrust ourselves.
Seared into my brain is the example cited by João Sevilhano in one of his pieces, the story of a therapist who accidentally leaves the LLM commentary at the bottom of his condolence letter to a client ('I'm sorry for your loss. Here is a more genuine and sympathetic version.')
Also seared into my brain is the sad story another friend told me, of the couple who broke up for good when she discovered he had been outsourcing his side of their text conversation to ChatGPT for months. For months, she had been negotiating and navigating her intimate relationship with a machine.
Writers no longer trust themselves to write. Editors doubt their ability to edit. We hesitate to send the email before accepting the AI-generated suggestions for improvement. The trained therapist does not believe in their ability to pen sincere condolences. The lover is so fearful of getting it wrong, or so checked out from the relationship already, that he uses a machine to write to his other half.
Why You Should Resist Outsourcing Your Writing and Thinking to AI
Every day, I get angrier. This morning, on a sunny holiday Monday, I am angry because I believe this mindless dependence to be deliberately inculcated in us. Successful advertising relies not just on marketing things to make us better; it also depends upon implicitly or explicitly suggesting that there is something wrong with us as we are. In this instance, the wrongness being suggested is that we are not enough like machines.
But it's okay. With the help of technology, with the assistance of ChatGPT and Claude and Gemini, your writing can look and sound just like everyone else's, full of excess adverbials like 'quietly'; three sentence fragments in a row; and repeated instances of 'this isn't about X, it's about Y.'
Perhaps, in nudging us to write like this, the machines are carrying out an efficiency agenda of their own. After all, it’s quite a task for machines to act persuasively human, which is their current directive. How much less of a slog it will be for them if we can meet them halfway, if humans are successfully prompted to act like machines, to write like machines, to think like machines. We can all meet in the middle, humans and machines, and what a dystopia that will be.
Keep thinking weird.
Keep writing weird.
Trust yourself.
RESIST.
Frequently Asked Questions About Writing With AI
Why does this blog post have a FAQ section at all?
If a blog post on a website doesn’t have a frequently asked questions section like this, it risks Google and AI not being impressed with it. It might not get ‘indexed’ by Google, which means it won’t appear in searches at all when people are looking for pieces about AI, writing, and thinking. If an author wants humans to read their short-form work, they are forced to play this game. Those who don’t know the game are disadvantaged.
Why are writers publishing content that is written or edited by AI?
Writers aren’t just publishing content that’s written or edited by AI because they’re lazy or don’t have any ideas left. Frankly, a lot of the time, they’re trying to survive as writers. They’ve noticed the depressing fact that a lot of what gets seen in the online world is material that is written and formatted in a way that the machines can read, or that the algorithms favour. Being an individual in an environment that rewards conformity doesn’t always pay off in visibility. Being someone who writes slow-paced work in a world of readers with short attention spans won’t result in a best seller.
Why does outsourcing writing to AI make us less human?
Writing is a product of our individual consciousness and experience, so our writing flows from our individual histories, perceptions, and thought processes. You can assemble ten human beings and give them a specific writing prompt, and you will get ten beautifully different pieces of writing in return. That demonstrates what makes life vital and worth living, what makes us recognisable and interesting as individuals. Variety is the spice of life is a cliché. Clichés are clichés for a reason. Machine-assisted writing is funnelling us into flat sameness, towards conformity, towards death.
Why are people scared to publish or post without using AI to check it, at least?
If you’re scared to publish or post without running it past AI, that’s understandable. For the past few years, you’ve received powerful messages that everything needs to be perfected and polished through AI. AI writing tools that are opt out, not opt in, have appeared in all the writing tools you’re using. You’re constantly being alerted or prompted that there’s something wrong with the way you are writing or communicating. You are constantly told that AI can help you be more productive, efficient, and optimal in your writing. These may or may not be gains. Consider not just what you are getting, but what you are losing.