A Stimarium Dossier
AI can improve your idea and flatten the room at the same time
People who brainstormed with ChatGPT came up with ideas rated more creative. Their ideas were also more alike: in one test, nine people separately named their invented toy “Build-a-Breeze Castle.”
What’s known, what’s not: Known, Our read, Open
Your idea can get better while the room gets narrower.
In 2024, researchers reported that people using ChatGPT produced ideas rated as more creative than people working unaided or with conventional web search.
A 2025 reanalysis of the same experiments looked at the whole group. Assisted brainstorming helped individuals and still left the room with a narrower range of ideas. [S01, S02]
A larger comparison of people and language models working separately, published online in December 2025 and included in a 2026 journal issue, found that humans retained an edge among the highest creativity scores. [S03]
Assisted brainstorming helped individuals and still left the room with a narrower range of ideas.
Think with the idea.
Take me sideways
This looks a lot like recommendation systems: optimizing individual relevance can make the whole culture more homogeneous.
Analogy, not a finding from the creativity studies.Challenge it
Similar ideas can sometimes be useful. Shared tools may establish workable conventions, reduce differences that add little, and help beginners meet basic quality standards.
Counterpoint for interpretation.Trace it
The underlying tension appears in cultural evolution, recommender systems, design systems and organizational best practices: coordination creates efficiency and sameness at once.
Cross-domain synthesis.Take it with you
Generate several independent directions before using AI. Ask it to widen the search space, attack assumptions and produce counterexamples before polishing.
Practical design principle, not a universal rule.What’s known, what’s not
↑ Back to where you wereA 2024 experimental study reported gains in rated creativity for assisted individuals. A 2025 reanalysis of its data found reduced idea diversity across groups.
We think AI collaboration may need to aim for ideas that differ from one another, as well as for fluent, well-rated ones.
Whether this homogenization persists across richer real-world creative work is still uncertain.
What the three labels meanClose explanations
Known marks what the cited evidence directly supports, including what named researchers argue.
Our read is Stimarium’s own interpretation of that evidence.
Open marks a question the cited evidence can’t answer yet, or a possibility we are raising ourselves.