Highlights:
- Researchers demonstrate how personalized AI scaffolds enhance creative collaboration.
- Study shows fully personalized LLM assistants significantly boost creativity and task quality.
- Personalization strengthens user trust, confidence, and cognitive synergy.
- Framework shows personalization improves collective memory, attention, and reasoning.
TLDR:
A new study by Sean Kelley, David De Cremer, and Christoph Riedl reveals that AI assistants customized to individual users’ psychological and cognitive profiles dramatically enhance creativity and teamwork, paving the way for next-generation personalized AI systems that amplify human intelligence.
As AI becomes increasingly integrated into professional and creative workflows, researchers are exploring how to make human-AI collaboration genuinely synergistic rather than merely assistive. In their recent study titled *’Personalized AI Scaffolds Synergistic Multi-Turn Collaboration in Creative Work’*, authors [Sean Kelley](https://arxiv.org/search/cs?searchtype=author&query=Kelley,+S), [David De Cremer](https://arxiv.org/search/cs?searchtype=author&query=De+Cremer,+D), and [Christoph Riedl](https://arxiv.org/search/cs?searchtype=author&query=Riedl,+C) examined the potential of personalized large language model (LLM) assistants to foster more dynamic, productive, and creative human-AI interactions.
The research investigates how integrating detailed user information — such as demographics, domain expertise, and psychometric data — can create AI systems that adapt to users’ thinking styles and communication preferences. In a large-scale experiment involving 331 participants, the team compared interactions between users and three levels of AI personalization: generic, partially personalized, and fully personalized. The results were striking — participants working with fully personalized AI generated marketing campaigns that were both higher in quality and more innovative than those produced with standard AI systems. This personalization not only improved creative output but also boosted users’ trust, engagement, and satisfaction with their AI collaborators.
Technically, the study applied a multi-turn interaction framework where the AI agent gathered information through an AI-guided interview that modeled users’ working styles. The system then used that data to scaffold subsequent interactions — functioning as a form of ‘external cognitive scaffolding.’ Causal mediation analysis revealed that personalization improved outcomes not by direct algorithmic optimization, but by enhancing the shared mental model between human and machine. This mutual understanding helped both parties to attend to key creative goals, maintain consistent context, and reason collaboratively over multiple conversation turns. The findings lay the foundation for designing AI systems that not only generate content but also co-create knowledge, guiding the development of future human-centered AI tools that expand rather than homogenize human creativity.
By introducing a theory-based personalization model, the research underscores the importance of adaptive communication in human-AI interfaces. This approach reframes AI assistants as cognitive partners capable of aligning with individual users’ methods of thought, ultimately reducing uncertainty and fostering truly synergistic co-creation. The insights from this work could transform how creative professionals, educators, and teams employ AI to amplify human potential in complex problem-solving environments.
Source:
Source:
Kelley, S., De Cremer, D., & Riedl, C. (2025). *Personalized AI Scaffolds Synergistic Multi-Turn Collaboration in Creative Work*. arXiv:2510.27681 [cs.HC]. https://doi.org/10.48550/arXiv.2510.27681