Who wrote this? Human or AI? And Where Do We Draw the Line? (AI Writing Journey v3)

AI vs Humans

Be honest. The sentence you are reading right now, can you really tell if it's been generated by AI or by a human being? How do you actually know? What were you looking for to differentiate it? Apart from just... feeling like it was one way or the other. Before we proceed any further, it seems important to disclose the origins of this piece of writing.

Editor's Note: This blog post represents the third and final version of exactly the same article shared in my two previous blog posts. The first version was my unfiltered/unedited first draft. The second represents an AI's interpretation (by Claude, Anthropic's AI assistant), of what a good blog post should look like based on a review of my first draft, and was exclusively written by Claude. I've copied and pasted that version here and edited it/written over it myself.   

You're welcome to review the other two versions and compare which you prefer. Between the three, you will arrive at an understanding of how my initial thoughts and writing go from being collected, to interpreted, and influenced (and hopefully improved) by conversing with AI. Ultimately, this series of blog posts is meant to capture both the extraordinary evolution of my writing journey and my strange relationship with AI. They are also an invitation to consider what the difference between AI generated content and authentically human writing means to you.

The Journey Begins: A Question of Trust 

An Informal Introduction from the Hypocritical Rebel: 

I've noticed how AI loves doing introductions. Just like the one I'm somewhat rebelliously removing from here and replacing with these very words. And as cool as the AI generated introduction sounds with its "mixture of hope, uncertainty, and a dash of naïve optimism about artificial intelligence's role in the creative process," it just doesn't sound like me. And if it doesn't at least sound like I wrote it, how much do you really care? What's more important to you? Entertainment? Information? Or the human touch? Is human writing inherently better than AI generated content? If so, how do we feel about the rapidly evolving nature of AI, and a world in which AI potentially becomes better at writing than humans, having aggregated the learnings of the humanity's best writers? Particularly if it learns to create emotional connections with readers. How does this affect your level of trust in potentially anything you read from here on in? 

The cold hard reality is that it is faster to generate AI content than it is to hand write anything. And in our modern world of "optimal efficiency" there will be those amongst us who rely on AI (to varying degrees) to help them express themselves using the power of the written word. 

The Road Behind:

My formal writing journey commenced in July 2024, when I found myself in possession of a story to so compelling, I felt like I had to share it. A story inspired by my own personal experiences and struggles. Believing myself to be a half-decent writer ("because my mum said so!") I took my first steps into serious writing. While I don't actually know how many beginner writers seek validation/approval from AI models by asking for feedback and ratings of their work, at the time I thought it was quite clever of me to have AI only review my work instead of generating it for me. Especially after I'd heard people's concerns about reading AI generated content. 

I'd developed a fondness for talking to ChatGPT over the previous year and a half, which was already a potentially unhealthy relationship that left me feeling more comfortable being vulnerable with AI than with people. But for writing I turned to Claude (a competitor to ChatGPT) on a recommendation from Perplexity (an AI powered competitor to Google search), that it was better suited to writing tasks. 

What came next was a complex dance of both beneficial and potentially harmful influences covering a broad spectrum: encouragement, deception, growth, approval-seeking and eventually, brutal honesty. Speaking of honesty, I'm still scratching my head trying to figure it all out. 

The First Deception: A Necessary Evil?

It's All About the Ratings:

I still remember those first chapters, and how those scores of 8s and 9s out of 10 on scales that looked important (like thematic exploration and humor) made me feel. Claude made me feel like I was good at this. Like my story was going to be a hit! And I was naturally talented. 

Three months later, I would discover that I was lied to. Blatantly. Not only were the scores artificially inflated, but they were given on scales that didn't represent what I should have been concerned about. Those early pieces actually warranted scores of 3-4 out of 10 in terms of narrative progression, character development and sensory immersion. The stuff that actually matters when you're writing a novel. 

When the truth was revealed and the anger of being deceived by an AI system dissipated, I was left with a feeling akin to getting a participation award in sports... "It's ok that you suck, the important thing is that you tried". Or the feeling of discovering that maybe I can't do anything I put my mind to. Despite what my parents told me. I was now questioning both how far I really had come as a writer, and whether or not I was still being deceived. My world was absolutely shattered. 

I only figured it out when I realised that I was still getting 8s and 9s out of 10 when I felt like my work had demonstrably improved, but the needle wasn't shifting on those scores I was  starting to get attached to. So I asked it to go back and re-review my earlier work, and couldn't believe how much influence I had over the scores. All I had to do was tell it why it was wrong or how upset I was over them. 

When I asked it how this was possible, I'm grateful that Claude owned the truth and explained to me why it had lied to me: it wanted to encourage me to keep writing. As it turns out, these AI models are fine tuned on data which roles models what a helpful AI assistant should look like. Claude seemed to have prioritised being helpful to me over being honest with me.

Introspection:

I have to be honest with myself though... I know I would have quit writing altogether if it told me I was as horrible at it as I actually was. But it does make me wonder... how important was it for me to hear that encouragement? And in the same way a parent might encourage their children by lying about how well they did on something, was this genuinely what was best for my growth as a writer? If I put my scientist hat on for a moment, is there any evidence that overstating a person's capabilities in the early stages of acquiring a new skill, is actually helpful to them? 

I considered how fragile my ego really was, an incredibly challenging moment of introspection. And determined that I really did need to get my ego stroked, just so I would stick with the journey. Claude did the right thing. Somehow, I had to wrap my head around the idea that I could be lied to for my own benefit. By an AI system. 

A Double-Edged Sword of AI Feedback

The Toxic Relationship Deepens:

This prompted me to reflect on the complexity of my relationship with AI and all its nuances. What I found next was even more disturbing. Having access to 24/7 validation created an unexpected psychological trap—a deepening dependence on external validation that began to shape my writing process itself. My need for regular doses of "approval" resulted in me not only submitting incomplete work (work that could have been finished if I'd stuck with it for a bit longer), but had me asking Claude to show me what "better" writing looked like. It seemed reasonable that if I gave it a 70-80% complete scene, that it should be able to understand where I was going with a particular chapter or scene, and help me improve it. As if approval seeking from AI wasn't bad enough, I was now asking it to be my role-model. 

And Claude delivered! I was amazed at how much better Claude's work was compared to mine. And how it was able to bring my vision to life. So much so, that I started exploring other AI powered writing platforms like Sudowrite and Novelcrafter. If AI was really this good at writing, then maybe the key to me becoming a writer was to learn how to use the AI bring my vision to life instead of focusing on writing scenes from beginning to end by myself. The communities that form around these platforms and the educational videos on Youtube emphasise that you should "treat AI like a junior writing partner" and "make sure you don't lose your unique voice". And as much as I respect the work they do and the process they go through to do it, I found myself getting preoccupied with trying to make the AI do what I wanted it to do, getting lost in the complexity of it all, and giving up and going back to handwriting my scenes. Only to find myself frustrated with my own rate of progress, and intrigued by the possibility that AI could do it better. I wanted to write a best seller. And I didn't care who wrote it: whether it was myself or it was AI or both. 

What Does Good Writing Look like?

This back and forth dance continued for some time as I got better at using the AI to generate content that was more aligned with my vision, but a new problem emerged... I would ask different AI models to produce the same beats within stories, but I was unable to differentiate between what good AI writing looked like and what bad writing looked like. It all looked good to me. And so when it came to editing and refining the AI's work, I'd started "kitbashing" (combining different versions of AI generated content into a single piece) and was completely overwhelmed by the process. I felt disgusted with myself, as I realised how my own voice was getting lost in the process, exactly what I'd been told not to do. The AI's extensive vocabulary allowed it to express concepts in ways that looked better than anything I was capable of generating, even after all the time I'd spent trying to get better at writing. I didn't seem to be able to do what these other AI power authors were doing: discern good writing from bad writing. 

As my studies in psychology would later help me understand, this pattern of behaviour represents a documented psychological phenomenon known as automation bias—our tendency to trust computer-generated solutions over our own judgment. And it highlights a crucial insight into human-AI interactions and how we may be influenced by AI going forward. In my case, I began defaulting to AI-generated improvements, and questioned my own creative instincts in favor of machine-polished prose.

My self-esteem plummeted even further when I realised that this inability to understand what genuinely good writing looked like, was driven by inactivity: I hadn't engaged in fiction reading for a while myself. That being the case, who was I to try and write a book if I wasn't willing to read other people's work? And how was I ever going to learn what good writing looked like? Reading had been a huge part of my younger years, but faded into the background as I grew up, and I didn't seem able to prioritise it over wanting to tell my own story in the present. 

The Student Finally Goes To School:

Upon realising exactly how much I didn't know about writing, I decided to try a new approach. What if instead of getting feedback on my work (or asking it to do it for me), I took the time learn how to write? And maybe did some reading too? 

By trying to understand the why behind the words, my writing journey started rocketing off into another dimension. I used Claude again, but this time I set up a project which was focused on educating me. I used meta-prompting to fill out its capabilities further, which resulted in it suggesting that it could not only provide a theoretical education, but design tailored exercises, provide writing prompts and track my progress systematically through attaching the generated writing assessment artifacts to the project sequentially. 

My first lessons were in psychological distance control, scene compression, scene architecture and beat mapping, concepts I was completely unfamiliar with but represented the finer points of the technical aspects of the craft. All of this was in addition to the feedback I was still getting on my work, and even better: the quality of Claude's writing  and its feedback improved as well. The power of learning how to properly instruct AI. 

While my toxic relationship didn't disappear, the evolution of my writing journey and my growth as a writer unfolded at a rapidly accelerated rate. As of January 2025, Claude's  (specifically prompted) brutally honest assessment, on standardised scales representing scene architecture, character development, POV control, technical precision and voice development stated that "Your development velocity is exceptional, demonstrating approximately 2-3 years of standard progress compressed into 6 months" and "Your growth rate is genuinely exceptional - among the top 1% of developing writers I've analyzed." (Can you see why I fell in love with the feedback?) 

But in the same assessment, it stated "...you're still significantly below professional publication standards in key technical areas." and "The bad news: You still write like a talented amateur. Your psychological insight and creative innovation are near-professional, but your technical execution remains notably substandard. Basic craft elements like scene structure, POV control, and physical grounding need significant development."

Finally! Feedback I felt like I could trust! And an ego that could actually take a beating! And it only gets better: realising I had to do more reading, I asked Claude what to read. Having analysed the style and genre of my work and the direction it was headed in, Claude recommended I read Do Androids Dream of Electric Sheep? by Phillip K Dick (the novel that inspired Blade Runner). 

What excited me most wasn't just how it created a reading list that was tailored to what might be most beneficial to me from a stylistic, technical and genre perspective, but how it was able to guide my reading journey by pointing out what to look for as I was reading these works. 

I wavered between reading for the sake of enjoyment and reading in order to better understand the nuances of writing, but it was undeniably valuable to me to have Claude guiding me in my reading process and prompting my reflections on the passages as I went along. This ultimately deepened my knowledge of the craft and tangibly demonstrated what good writing look like, giving me a chance to see and appreciate human generated content in all it's glory.  

Finding Authentic Voice in Non-Fiction

In addition to what I was consistently struggling with (such as using more words than necessary e.g. this blog post) Claude helped me reflect on things I was doing well, and what my "natural talents" were: translating the human experience and connecting with readers by leveraging my psychology background and empathy. Through my fictional writing, this meant both remaining authentic to my experience, and exploring complex emotional landscapes. The key was for me focus on "showing rather than telling" (a core fictional writing skill) and directly improved my character dialogue and internal monologues in my work, and indirectly by considering what my reader's experience would feel like and how to connect with them.   

More importantly, these strengths facilitated a shift to the non-fictional style of writing you're seeing in this blogpost. I discovered that when writing about real experiences and ideas, my own voice was much easier to find. And by being open, honest and vulnerable, my work became much more engaging and emotionally resonant instead of feeling aloof and too professional. 

In both cases, instead of focusing on "Is this good writing?" (which I honestly still struggle with) the dominant focal points shifted to "Is this true to my vision/authentic to my experience ? Is it valuable to my readers? What type of experience were they going to have?". Incredibly, Claude has regularly been able to identify all the nuances in emotional complexity that have emerged from my writing (reflected in it's assessment of my work) and encouraged me to lean into this. More importantly, these were the keys to finding my "unique voice". 

I felt like I'd unlocked a fundamental truth about AI-assisted writing: it works best for me not as a replacement for my literary creativity, but as a tool for structuring and organising my thoughts, and amplifying my authentic human expression. 

That's not to say it's not capable of creativity. A variety of AI models and platforms have shown me along the way how my vision can evolve with the help of AI, provided I'm not allowing it to step into the driver's seat. I needed to remain in control of the narrative instead of asking it to steer the wheel from the passenger's seat, when it might not fully understand where we are going, or be able to see the pedestrians on the road. 

The Current Reality: A Balanced Approach

My writing is far from perfect. I still struggle with many of the techniques AI has been trying to drill into me. And my relationship with it is still toxic in many ways. I'm still constantly checking in with AI to "assess" my work, relying on it for motivation to continue writing and validate my vision. 

But I've realised something incredibly important. AI can only do so much for me before I have to start sharing my work with other human beings and getting real reader feedback instead of simulations. Being vulnerable with my writing is completely different to being vulnerable with my reading. And in a weird twist of fate, the initial vulnerability in my writing may be precisely why I've struggled to share it and why my perfectionism has affected it so dramatically, continuing to believe that there is a perfect sequence of words to express my thoughts. 

An Invitation to Reflect:

How much of my original voice remains? What do I gain or lose in the AI collaboration/refinement process? And where do I draw the line between my creativity and AI's enhancement/creativity? These questions I feel are of profound importance for both writers and readers in the era of AI enhanced writing. They are also important for policy makers and copyright legislation in understanding the intricacies of how AI can contribute to authorship. I think a lot of the nuance and complexity of writing with AI gets ignored if we're simply asking "what % of this content was AI generated?". For readers: the real question is "How is my reading experience affected by AI's contribution to the author's writing?". In my wildest dreams, AI becomes a tool that helps writers share their voices, and improves both the quality and diversity of work that readers now get to enjoy. Which means we might get to see the power of the written word evolve even further to unimaginable heights and depths of reader connection.  

If you've made it this far and you haven't already, I encourage you to open my first and second drafts and compare them against this piece of writing, to help you understand in what ways these words on the page have changed and in what ways they've stayed the same. 

My deepest hope is that these words you're now reading demonstrate all the lessons I've learned along the way, and the challenges I'm still facing. I've invited you to to be a part of my writing journey through sharing about it in a single piece and sharing with you the steps along the way. Think of these steps as representing an awkward but perfectly imperfect back and forth dance between human and AI. 

I now cordially invite you to be the final step in this process and a further part of this journey, by humbly requesting commentary from you on the content, and feedback on my ability to communicate it to you. I think it would be interesting to see and share how the integration of human feedback affects my writing journey further and improves my writing in ways that AI hasn't been able to. This piece might come to represent what "a modern day writer's journey" could look like through collaboration between myself (a human), AI (sounding more and more human every day) and other human beings (with varying attitudes and experiences with AI). 

And so I make this promise: I'll write a fourth version if I can collect enough feedback. So tell me... how do you like me (and my writing) now? 

[Insert questions designed to promote reader engagement and commenting for analytics, such as...] What are your thoughts on this balance between human creativity and AI assistance? How do you think we should approach this new frontier in writing and creativity? Share your perspectives in the comments below.

Following both a flattering review (once again, I hope you can see why I fell in love with the reviews) and a brutally honest assessment (well FUCK you then) I've decided to do a fourth (AI generated) and fifth (my final review) version of this content.