Answers are grounded in this paper and cite their sources.
While generative artificial intelligence (AI) can quickly produce appealing work, human involvement in creation may still matter to creators themselves, audiences and those who assign credit. This review examines research across creative, communicative and service domains, including art, literature, music, science, marketing and advice. We organize the evidence by evaluator, by the outcome measured and its timing, and by what was compared, including differences in content, in information about how a work was produced, and in the human role of creation. Outcomes include quality ratings, relational experiences such as feeling heard, and felt ownership. Our framework links specific human activities, such as setting intentions, selecting, revising and verifying, to specific goals. The literature shows that in AI-assisted work, human verification matters for trust in accuracy of the work, their personal input for sincerity, and their choices for the feeling of ownership. The strongest evidence concerns origin-information effects, where labeling the same work as human- versus AI-made can change immediate evaluator ratings. In blinded studies, AI-generated texts were rated higher than human-written texts, yet labeling the text as AI-generated lowered its ratings. These effects vary in size, and sometimes in direction, across contexts and measures. Less is known about the long-term effects of AI-use on creators’ motivation, creative identity, and skills. A well-rated result does not reveal the value of a particular human contribution or the consequences of delegating it to AI. Nor is more human involvement automatically better. Choosing to delegate work or selected work tasks to AI can itself express agency. As AI agents become capable of taking on more creative processes, deciding what to delegate may become a central form of human involvement. This review offers a framework for identifying which contributions matter to whom and for what purpose.
No takes yet. Share an insight, caveat, or question.
Andersen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: