Who Holds the Pen? New Research Rethinks Writing With AI

Picture two writers facing the same blank page.

One asks AI to invent the characters, develop the plot, and write the story. The other imagines the story first, then invites AI to help bring it to life.

Both have access to the same technology.

But will their stories—and their experiences creating them—be equally satisfying?

New research suggests that the difference may depend on who holds the creative reins.

The Discovery

In November 2026, researchers Kartik Hosanagar and Daehwan Ahn published a study in PNAS Nexus exploring how different forms of human–AI collaboration influence creative writing.

Their question wasn’t simply whether artificial intelligence makes writing faster.

They wanted to understand what happens to creativity, satisfaction, and originality when responsibility for different parts of the creative process shifts between people and AI.

The researchers recruited 285 university students for an experiment involving two 1,000-word creative writing assignments.

First, participants wrote without AI assistance. During a second session, they used ChatGPT 3.5 under one of three randomly assigned collaboration arrangements.

In one arrangement, AI handled nearly everything while humans approved its work.

In another, humans developed the ideas and outlines while AI handled drafting and editing.

In the third, humans and AI collaborated throughout the process.

The results revealed an important distinction.

Completion times were similar across the three AI-assisted approaches, but story quality, participant satisfaction, and creative diversity differed.

When humans were primarily responsible for approving AI-generated material, the researchers observed weaker outcomes in several areas.

The findings suggest that the way people collaborate with AI can matter as much as the technology itself.

The Human–AI Partnership

What makes this research particularly interesting is that the investigators examined creativity as a shared process rather than simply judging a finished product.

Writing involves more than arranging words.

It requires imagination, emotional understanding, choices about meaning, and a sense of what deserves to be communicated.

The researchers divided that process into activities such as generating ideas, developing outlines, drafting, and editing.

In the human-led planning approach, participants created their own story concepts before AI assisted with producing the text.

In the continuous collaboration approach, people could exchange ideas with AI throughout the creative process.

Both arrangements preserved meaningful opportunities for human contribution.

The research found that these approaches generally avoided the clearest quality and satisfaction losses observed when participants were limited largely to approving AI output.

Preserving human involvement in early creative decisions was also associated with greater diversity across the resulting stories.

This finding challenges the assumption that giving AI more responsibility necessarily improves creative results.

AI can suggest possibilities, organize information, and assist with expression.

But humans contribute experiences, intentions, cultural understanding, and personal perspectives that give creative work its identity.

The most effective partnership may depend not on how much AI accomplishes, but on whether people remain meaningfully involved in shaping what they create.

The Possibilities Ahead

These findings invite us to reconsider how AI might support creativity in classrooms, workplaces, and everyday life.

Imagine a student developing an original story concept, then asking AI to explore alternative ways of structuring it.

Or an aspiring writer using AI to experiment with dialogue while deciding which words authentically express a character’s personality.

A creative team might use AI to organize ideas and explore alternatives while retaining responsibility for the vision and meaning behind the work.

These are possible applications, not outcomes established across those settings by this study.

The researchers examined university students completing relatively short creative writing assignments. Their experiment used ChatGPT 3.5, and all participants completed the unaided writing task before the AI-assisted assignment.

Additional research with experienced writers, longer projects, and newer systems will be needed to understand how broadly the findings apply.

There is also no reason to conclude that AI-assisted writing must lack originality.

Instead, the research highlights a design consideration: reducing human participation to accepting or rejecting generated material may diminish aspects of the creative experience that people value.

Future AI tools could become better at supporting exploration without taking control of the process.

The opportunity is not simply to produce more content.

It is to make creative expression more accessible while protecting the individuality that makes it meaningful.

The blank page has always belonged to someone with something to express.

AI can help fill it with possibilities, but the choice of which story is worth telling remains profoundly human.

Perhaps the most compelling collaboration begins there.

Sources & Further Reading

1. Original Peer-Reviewed Research

Hosanagar, K., & Ahn, D. (November 10, 2026). Designing human and generative AI collaboration: Trade-offs in productivity, satisfaction, and creative diversity. PNAS Nexus, 5(11), pgag353.

DOI: 10.1093/pnasnexus/pgag353

Read the original PNAS Nexus study

2. Original Research Manuscript

Hosanagar, K., & Ahn, D. (December 14, 2024; revised January 29, 2025). Designing Human and Generative AI Collaboration.

This earlier manuscript documents the research preceding its November 2026 peer-reviewed publication.

Read the researchers’ original manuscript

3. Research Summary — PNAS Nexus

(November 10, 2026). Preserving human involvement at the start of a creative task can improve AI-assisted outcomes.

PNAS Nexus research summary, distributed through Tech Xplore.

Read the research summary