Is AI making us all say and create the same things?
Similar content, creative standardisation
The unsettling question
Three students use ChatGPT to write a summary report. They have not coordinated. You read the three texts one after another. You feel as though you have read the same one.
Five communications agencies use the same tool to produce campaigns. Five competitors read the same strategic reports generated by the same models. Every day, millions of assignments, emails, presentations, articles and images come out of the same handful of systems.
When everyone asks the same tool for the same thing, everyone ends up saying the same thing. Distinctiveness, the voice that stands out, the rough edge, the unexpected idea, suddenly becomes rare. Even suspicious.
The starting point
Large language models produce what is statistically probable. By design, they tend towards the average. This is a strength when you quickly need something decent. It is also a risk when hundreds of millions of people draw on them to produce what they then present to the world.
Worse still, these models train on the web, which fills up with more AI-generated text every day. Future models will learn partly from the output of earlier ones. Researchers call this phenomenon model collapse: as AI repeatedly learns from its own copies, variety gradually declines, like a photocopier copying its own photocopies.
Behind the aesthetic question lies a strategic one: if every competitor thinks with the same tools, how can anyone stand out? And a cultural one: what do we lose when the diversity of voices is smoothed away?
Tomorrow
Tomorrow, knowing how to think and write without AI, or with it while keeping your own voice, could become a rare advantage. Organisations, creators and schools will have to decide which aspects of human diversity they want to preserve, and how.
Possible angles
Reading to get started
Wired — The Homogenization Problem in AI
A good introduction to the phenomenon, illustrated with concrete, recent examples.
Shumailov et al. — The Curse of Recursion (Nature, 2024)
The scientific paper that introduced the term model collapse. The abstract is enough to grasp the issue.
Seth Godin — Purple Cow
A classic on differentiation that takes on new meaning in the age of generic content.
Walter Benjamin — The Work of Art in the Age of Mechanical Reproduction
An older essay that feels strikingly contemporary: how reproducibility changes the value of a work.
Articles on divergent thinking (Guilford)
To understand what we lose when we optimise for convergent thinking.
Try it yourself
- Give the same brief to three different AI systems, or to the same one three times. Compare the outputs: what looks similar?
- Ask an AI to write in the style of an author with a very distinctive voice.
- Visit the homepages of ten recent start-ups. Look for repeated phrases, similar visuals and the same structure.
- Find a recent piece of work, such as a campaign, article or presentation, that really stands out. Ask yourself why.
Which context would you anchor it in?
Content standardisation has different implications for an agency, a school and a creative studio.
Higher education
Dissertations, essays, assignments: assessing work when outputs converge
Business
Communications, strategy, brand differentiation in a homogenised market
Creative industries
Design, music, writing, cinema: distinctive voices versus generic models