The last original thought: What happens to creativity when AI makes everything?

AI is quietly taking over creative output - and it's hollowing out originality in the process. From blog posts to reports, essays and film scripts, machine-generated content now dominates our world. An analysis of 900,000 newly created web pages in 2025 found that 74.2% contained AI-generated content. But the real problem isn't volume, it's what happens when AI trains AI. Researchers call it "model collapse": A feedback loop that erodes linguistic variety and kills real creativity. Like a photocopy of a photocopy, each regeneration impacts quality and the output loses value.

---

Someone, somewhere, is asking an AI to write their wedding vows right now. Congratulations, your eternal love is statistically average!

But, before you feel too smug about it, consider what you used AI for this week. A work email? A social media post? A summary of that report you didn't want to read? We are, quietly and without much fuss, outsourcing the very thing that makes us human: The ability to make something brilliant from nothing.

The question isn't whether AI is useful. It is - in the right places. The question is what we lose when we give it free rein entirely.

The answer, it seems, is quite a lot.

We handed over the keys and walked away

The scale of AI-generated content is staggering.

• As of late 2024, more than half of all new articles published on the open web were primarily written by AI - up from just 5% before ChatGPT launched in November 2022. This is more than a year ago, can you imagine what the percentage is now?

  
  

• A separate analysis of nearly a million new web pages found that 74.2% contained detectable AI-generated content

  
  

• On social media it gets more surreal: Roughly 71% of images shared across platforms are now AI-generated, and over 80% of the content recommendations you see in your feed are decided by an algorithm

But of course, it’s not just blog posts and stock photos. AI is writing reports, drafting scripts, illustrating books, composing music, generating entire film treatments, and yes, someone is absolutely using it to plan their weekly shop. The machines aren't just taking our jobs, they’re taking our culture, too.

The photocopy of a photocopy

The thing about AI that tends to get glossed over in all the breathless coverage is that it doesn't invent anything. Not one single thing. Every word it produces, every image it generates, is a recombination of the human-made work that was fed into it. It’s a very confident remix artist with no original recordings of its own.

This makes AI, at its core, a second-hand medium. It processes what humans have already created, finds the patterns, and outputs something that resembles the average of all those inputs. Which sounds fine, until you realise the inputs are increasingly AI outputs themselves.

A landmark 2024 study published in Nature confirmed this nightmare scenario in academic terms: Training AI on AI-generated content causes irreversible defects in the resulting models, and the original diversity of ideas simply vanishes over time.

Further research into large language models (LLMs) shows that successive generations trained on machine-produced text suffer a measurable decline in lexical, syntactic, and semantic variety. The effect is especially pronounced in tasks requiring high levels of creativity.  

Researchers call this "model collapse."

Others might call it artistic inbreeding. Because it’s no longer artificial intelligence at this point. It’s artificial mediocrity.

Where have all the Shakespeares gone?

Great art doesn't come from averages. Shakespeare wasn't the statistical mean of Elizabethan theatre. Picasso didn't prompt a machine to "give me something blue and melancholic, no watermark." Warhol's genius was in the provocation - the very human decision, the audacity, to point at a soup tin and call it art. These breakthroughs required lived experience, failure, obsession, joy, grief. None of that goes into a training dataset.

Researchers have already flagged "creative stagnation" as a concrete risk: Models trained on homogenised data become structurally incapable of producing the divergent, surprising output that defines genuine creativity. And the bad news for the AI boosters - audiences are beginning to sense this, even if they can't quite put their finger on what's gone wrong. Whilst consumers initially engage more readily with AI content when they don't know its source, 52% report feeling less engaged once they suspect it's machine-made, and 26% find AI-generated copy impersonal.

The next Warhol is probably out there somewhere. Unfortunately, they're spending their creative hours prompting Midjourney to generate "vintage retro art, photorealistic, cinematic lighting."

The feedback loop from hell

The trajectory is not encouraging. Experts - including analysts at Europol - have projected that up to 90% of online content could be synthetically generated during 2026.

Research also suggests that high-quality, human-generated text data may be effectively exhausted around the same time, meaning LLMs will have read every word we've written since we learned to wield a stub of charcoal. Future AI models will be trained almost entirely on the outputs of current ones. When we get to that, the snake is literally eating its own tail.

Meanwhile, Google's own data shows AI-written pages in its top search results climbed from 11% to nearly 20% between mid-2024 and mid-2025. We’re not just producing AI content, we’re actively surfacing and amplifying it, reinforcing the loop with every search. At this rate, future AI models will be trained on vibes, hallucinations, and the aggregated output of previous AI models. Honestly, not that different from some published authors. But considerably more alarming at scale.

TEN BOOKS THAT AI COULD NEVER WRITE IN A MILLION YEARS

1984

American Psycho

Catch 22

Crime and Punishment

Fear and Loathing in Las Vegas

For Whom the Bell Tolls

One Flew Over the Cuckoo's Nest

The Lord of the Rings

The Grapes of Wrath

To Kill a Mockingbird

How to avoid the beige

There's no need to burn your laptop and write exclusively in quill and ink. AI is a genuinely brilliant tool - for the right jobs. The trick is knowing what jobs those are.

This is how you break AI's iron grip:

• Write the first draft yourself. Always. Even badly. Especially badly. The mess is where the real ideas live

  
  

• Use AI for grunt work, not soul work. Scheduling, data sorting, formatting, expense reports, meeting summaries; This is where AI genuinely earns its keep and frees you up to do the things only you can do

  
  

• Consume human-made art, deliberately. Go to galleries. Read physical books. Watch films made by people who had something to say. Buy music from actual musicians. This isn't nostalgia, it's input

  
  

• Stop outsourcing your voice. Your specific, weird, slightly-too-honest perspective is the thing AI literally cannot replicate, because it has never been you, and it never will be

  
  

• Demand provenance. Initiatives like the Content Credentials scheme are working hard to tag how content was made. Use platforms that label AI content honestly, and reward the ones that do

  
  

• Make something no one asked for. Come on people, you can do it. This has always been where the good stuff comes from

Final word:

AI is a tool, not a muse. A hammer doesn't build a house by itself. The question was never whether a machine could write a sonnet, it can, and it'll do it in 0.3 seconds with five alternative versions. The question is whether a sonnet written by something that has never loved or lost anything actually matters to any of us.

Now go and make something.

Badly.

Personally.

Entirely on your own.

---

Get started with Freelance Words

Strong communication is the bedrock of good business and when there’s a lack of it, problems can arise. Take the ambiguity and doubt out of what your business needs to say. Contact us now.

Contact