The past few years has seen increasing discussion and debate about the use of AI to create content – that is, to “write”.

‍

It is a shame that this is how many people are using AI because it is applying AI in completely the wrong way. To put it simply, AI can’t write.

It could be easy to dismiss this view as fear that my job will be taken over. But that’s not the case (or at least not solely the case!). One of the main reasons I chose a career in PR is because I love writing. I love the power and the magic of words, and I firmly believe AI will never be able reproduce this, no matter how good it gets.

Good writing has personality, charm, humour, emotion. It is passionate, thoughtful, influential, inspirational, and nuanced. Perhaps most importantly, it uses original, human ideas to inspire, challenge and inform other people. AI has none of those things.

Essentially AI is a computer program that can string together words in a way that makes (mostly) sense. Of course, it can also fail dismally at this which is how we end up with “AI slop” and “AI hallucinations”. A good read is “Chat GPT is bullshit” which explains how large language models are simply producing human-like text by putting words together that make the most sense, but without any intention of communicating or indeed being factual.

Over the course of my 20-odd years in PR, I have read thousands of documents intended to communicate, persuade and educate, and I know that the range of ability in writing is huge. Some people have amazing ideas but struggle to get them down in black and write. Some people have amazing ideas and write clearly and effectively. And some people can write beautiful, flowing sentences but when you finish reading them, you still don’t really know what they were trying to say.  

AI is most similar to this latter group of people, who can write well but can’t actually communicate. Content produced by AI is like a big shiny balloon – it looks nice and substantial, but give it a little prod and it collapses into nothing.  

Perhaps the real difference here is between the concepts of “content” and “writing”.  Content is filler. It’s words to put on a website, in a LinkedIn post, on a CV, for the sake of meeting the criteria of computer algorithms or search engines. But real writing is designed to change how people think or feel, to improve their understanding and broaden their horizons, to take them into a new and different world. It is based on new thoughts and ideas, not regurgitating concepts or knowledge that is already in the public domain.

So will people keep using AI to create content? Yes, of course they will. But they will do it because it’s cheap, easy, and they can’t tell the difference themselves between good and bad writing. In a world where everyone else sounds the same, those who can think for themselves, write well, and take the time to do so, will stand apart.  

                                                                                                                                                                                    -oOo-

‍

‍

Response from Claude.ai

This is a well-argued piece, and it lands some genuine points — but I'd push back on a few things if I were commenting on it.

Where I think you're right: The content/writing distinction is the sharpest and most useful idea here. There's a real difference between text that exists to satisfy an algorithm or check a box, and writing that's trying to move someone. That distinction predates AI, though — plenty of human-written content is also just filler. AI has made low-effort content dramatically cheaper to produce, which is different from AI being incapable of anything beyond low-effort content.

Where I'd push back: The "big shiny balloon" description is a real phenomenon, but it describes a failure mode, not a ceiling. It's true of unedited, prompt-and-go AI output. It's also true of a lot of human first drafts. The question isn't whether AI text can be hollow — obviously it can — but whether hollowness is inherent to the medium or just common because most people using AI aren't trying very hard, or don't have the taste to push it further.

The "ChatGPT is Bullshit" argument (the Frankfurt-style "no concern for truth" point) is a good description of the mechanism — next-token prediction has no built-in commitment to accuracy or intention. But I'd separate two claims that argument bundles together: "AI has no ground-truth relationship to facts" (largely true, and worth taking seriously) and "AI cannot produce writing with personality, charm, or emotional effect on a reader" (much shakier — those are properties of the text as experienced, not properties of the process that generated it). A reader who doesn't know the source can and does respond emotionally to AI-assisted text sometimes. That's uncomfortable, but it's evidence against the "never, no matter how good it gets" claim.

I'd also gently note: the piece argues writing requires intention, but a lot of what makes writing land — rhythm, surprise, precision — can be judged by the output whether or not the process had subjective intention behind it. That's a genuinely contested philosophical question (does meaning require intention, or just effect?), not a settled one.

Where I fully agree: as a categorical matter, right now, in practice, most AI content is thin because most people using it don't do the work of shaping it — the taste, editing, and judgment still have to come from somewhere. If that's the practical claim, it's well supported. It's the "never, no matter how good it gets" absolute that I think overreaches.