Study Finds Mixed Effects of LLM Assistance
- •Study finds no significant average effect of LLM assistance on entrepreneurs’ theory structure or content
- •Experiment with 199 participants found support for less developed or distinctive unaided theories
- •Assistance had the opposite effect when unaided theories were already more developed or distinctive
Researchers Sung Ho Park and Alex Murray report in the 2026 Strategic Entrepreneurship Journal that large language model (LLM) assistance had no significant average effect on the structure or content of business theories formed by entrepreneurs. Their within-subject experiment included 199 participants, each of whom developed theories both with and without LLM assistance.
The average result concealed different effects depending on participants’ unaided theories. LLM assistance supported causal structure and distinctive content when those theories were less developed or distinctive, but had the opposite effect when participants’ unaided theories were already more developed or distinctive.
The researchers say LLMs can surface knowledge across domains and sharpen causal logic, while also steering reasoning toward conventional patterns and away from counterintuitive ideas. They conclude that assistance can scaffold entrepreneurs still developing theories, while caution is warranted when unaided theories are already developed or distinctive because assistance may diminish those qualities.