AI Creativity Homogenization: Better Stories, More Alike

Picture a fourth-grade class where every student gets AI help brainstorming a short story. Each story comes back stronger. Then the teacher reads them in a row and notices something odd: they’re starting to sound alike. That’s not a hypothetical. It’s the main finding of a 2024 experiment in Science Advances, and researchers now call the pattern AI creativity homogenization. Stories written with AI ideas were rated “more creative, better written, and more enjoyable,” especially from less creative writers. But they were also “more similar to each other than stories by humans alone.”

The authors describe it as a social dilemma. Each writer is better off. The group, collectively, produces a narrower range of ideas. Here’s why that happens and what it means for your kid.

Key Takeaways

• Doshi and Hauser (Science Advances, 2024) found AI-generated story ideas improved individual stories, most for less creative writers, but made stories more alike.
• A 36-person study found ChatGPT users produced less semantically distinct ideas from each other and felt less responsible for their ideas (Anderson et al., 2024).
• In an essay-writing experiment, a feedback-tuned model (InstructGPT) reduced diversity across authors, while the text people wrote themselves stayed as diverse as before (Padmakumar & He, ICLR 2024).
• None of these studies tested children directly. A 2026 review of child studies flagged “template-based” thinking and reliance on AI plot templates as reported risks.
• At home and in class, the protection is simple: get the child’s own idea down first, then let AI help with the rest.

What Is AI Creativity Homogenization?

Homogenization means that when many people use the same AI tool for creative work, their outputs converge, even if each output individually improves. It’s a group-level effect. You can’t see it in one child’s story. You see it in thirty.

The clearest evidence comes from Anil Doshi and Oliver Hauser’s paper “Generative AI enhances individual creativity but reduces the collective diversity of novel content,” published July 12, 2024, in Science Advances. They ran an online experiment on short-story writing. Some writers could get story ideas from a large language model; others wrote alone. Independent evaluators then rated the stories.

The results, from the paper’s abstract:

• Access to AI ideas “causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers.”
• “However, generative AI–enabled stories are more similar to each other than stories by humans alone.”
• The authors conclude: “With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced.”

The participants were adults, not children. And it was one kind of task, short fiction, in one experiment. But the design was causal: random access to AI ideas, not a survey of people who chose to use it.

Why AI Pulls Everyone Toward the Same Ideas

Here’s the mechanism, in engineering terms. A language model generates text by predicting likely next words, based on patterns in its training data. When you ask it for “a story idea about a dragon,” it produces something from the high-probability region of what dragon stories tend to be. Ask again and you’ll get variety, but variety around the center. Two different kids asking the same model get ideas drawn from the same distribution.

Without AI, each child draws from a different distribution: their own memories, books, jokes, pets and fears. That’s why thirty unassisted stories spread out more, even if some are weaker.

There’s a second layer. Chat models are tuned with human feedback to give answers people rate as helpful and polished. That tuning can narrow the range further. Vishakh Padmakumar and He He tested this in a controlled essay-writing experiment published at ICLR 2024. People wrote argumentative essays with a base model (GPT-3), a feedback-tuned model (InstructGPT) or no model. Writing with InstructGPT, but not the base GPT-3, “results in a statistically significant reduction in diversity,” increasing similarity between different authors’ writing. The effect came mainly from the model’s contributions. The text the users wrote themselves “remains unaffected by model collaboration.” (This paper appeared at a peer-reviewed conference but is hosted as an arXiv preprint.)

That last detail matters for parents. The sameness lives in the AI-written parts. Your child’s own sentences stay as distinct as ever.

Ideas, not just sentences

Barrett Anderson, Jash Shah and Max Kreminski ran a 36-participant study for the 2024 Creativity and Cognition conference on brainstorming. Users with ChatGPT generated more ideas with more detail, but “different users tended to produce less semantically distinct ideas” than users of an alternative tool. They also “felt less responsible for the ideas they generated.”

So the convergence can start before writing, at the idea stage. If every kid brainstorms with the same model, their starting points cluster.

What we know about children

Direct evidence with kids is thin. A September 2026 scoping review in Frontiers in Psychology by Niu and colleagues mapped 22 primary studies of generative AI and children’s creativity. Among the barriers reported in individual studies were the “potential for ‘template-based’ thinking” and “potential for over-reliance on AI-generated plot templates.” The reviewers also note there were no longitudinal studies, so nobody knows whether these effects last.

A 2026 preprint with 40 children aged 7 to 12 found that an AI storytelling system narrowed the quality gap between kids by 83.5%, with creativity and richness converging while narrative structure stayed tied to each child’s baseline. We cover it in AI narrowed the storytelling gap for kids 7–12. Convergence in quality isn’t the same thing as convergence in content, but it’s the same family of effect.

Individual Gains vs. Collective Sameness

Measure	With AI ideas	Without AI	Source
Rated creativity of each story	Higher, especially for less creative writers	Lower on average	Doshi & Hauser, 2024
How similar stories are to each other	More similar	More varied	Doshi & Hauser, 2024
Distinctness of brainstormed ideas across users	Less distinct	More distinct	Anderson et al., 2024
Sense of responsibility for ideas	Lower	Higher	Anderson et al., 2024
Diversity across essays (feedback-tuned model)	Reduced	Baseline	Padmakumar & He, 2024
Diversity of the human-written parts	Unchanged	Baseline	Padmakumar & He, 2024

What This Means for a Classroom of AI-Assisted Kids

The “average story” problem

If a teacher assigns a creative piece and most students use the same chatbot for ideas, the class will likely produce better-than-usual work that clusters around similar premises. That makes grading odd (everything’s competent), and it makes the rare unusual story stand out more. In a homogenized class, originality becomes the scarce resource.

Less creative writers gain the most, which is good and complicated

Doshi and Hauser found the biggest boost went to less creative writers. For a child who freezes at a blank page, AI ideas can be a genuine on-ramp. The tradeoff is that the on-ramp leads to the same place for everyone. A teacher might reasonably let struggling writers start from AI ideas while asking them to change at least one big element to something from their own life.

Strong writers may lose distinctiveness

The flip side: a child with a strange, vivid imagination who uses AI ideas may produce something more polished and less theirs. The sameness costs them the most, because their own ideas were the most distinctive to begin with.

How to Protect Your Kid’s Odd Ideas

Idea first, AI second

The Padmakumar and He finding suggests the human-written parts stay diverse. So the order matters: your child writes down their premise, characters or weird twist before opening any AI tool. Then AI can help with structure, spelling or a stuck middle. The core stays theirs.

Make “weird” a compliment

Tell your kid that the strangest idea in the room is the one AI is least likely to have. “A story about a lighthouse that’s afraid of the dark? A chatbot wouldn’t come up with that.” Kids who know originality is valued are more likely to keep it. Our piece on creative writing skills in the AI era has more ways to encourage voice.

Use AI to generate what to avoid

A useful flip: ask the AI for five typical ideas on the topic, then ban them. Your child has to come up with something none of the five covers. It turns the model’s center-of-the-distribution habit into a map of what’s already been done.

Pull from real life

The cheapest source of unique material is your child’s own experience: the time the dog ate the birthday cake, Grandpa’s stories, the bus route. An AI has no access to those. Ask, “What happened to you that could go in this story?”

What not to do

Don’t ban AI from creative work and call it solved. The research shows real benefits for writers who struggle, and blanket bans push use underground. The goal is ordering and ownership, not abstinence.

What to Watch For Over the Next 3 Months

• Week 4: Your child writes their idea down before opening any AI tool, at least some of the time. You can point to a detail in their work that clearly came from their own life.
• Month 2 red flags: Their stories start to feel generic, with tidy endings and familiar plots. They can’t tell you which idea was theirs. They ask the AI for ideas before thinking for even a minute.
• Month 3 self-check: Read three of your child’s recent creative pieces. Ask: would a stranger know these were written by your kid? If yes, their voice is intact.

Frequently Asked Questions

What is AI creativity homogenization?

It’s the tendency for work made with AI help to become more similar across people, even as each piece improves. Doshi and Hauser (2024) found AI story ideas made stories better individually but more alike as a group. It happens because many people draw ideas from the same model, which favors common patterns.

Does AI make kids less creative?

The evidence with children is thin and mixed. Adult experiments such as Doshi and Hauser’s found AI can raise individual creativity ratings while reducing variety across a group. A 2026 review of 22 child studies found reported benefits and risks like “template-based” thinking, but no long-term studies. It depends a lot on whether the child’s own idea comes first.

Should teachers let students use AI for creative writing?

Many approaches can work. The research suggests a sensible middle path: students generate their own core idea first, then use AI for help with structure or stuck points, and change at least one major element to something personal. That keeps the benefits for struggling writers while protecting the class from sameness.

Is my child’s writing still original if AI helped?

The parts they wrote themselves likely are. In one experiment, the human-written text stayed as diverse as before, and the sameness came from the AI’s contributions. If your child’s premise, characters and key details are their own, the work keeps their voice even if AI helped polish it.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Doshi, A. R., & Hauser, O. P. (2024). “Generative AI enhances individual creativity but reduces the collective diversity of novel content.” Science Advances, 10(28). https://doi.org/10.1126/sciadv.adn5290
2. Anderson, B. R., Shah, J. H., & Kreminski, M. (2024). “Homogenization effects of large language models on human creative ideation.” Creativity and Cognition 2024. https://doi.org/10.1145/3635636.3656204
3. Padmakumar, V., & He, H. (2024). “Does writing with language models reduce content diversity?” ICLR 2024 (arXiv version). https://arxiv.org/abs/2309.05196
4. Niu, T., Liu, H., Pang, P., Luo, Y. T., & Liu, T. (2026). “The role of generative AI in facilitating children’s creative thinking and cognitive development: a systematic scoping review.” Frontiers in Psychology, 17. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2026.1880052/full
5. Fan, M., Ma, W., Cui, X., Dai, X., & Huang, S. (2026). “Floor raiser or ceiling limiter? Differential storytelling outcomes with a child-centric GenAI system across individual differences.” arXiv (preprint, not peer-reviewed). https://arxiv.org/abs/2606.27067