ai _ devils advocate

This project tests how different kinds of feedback shape the quality of strategic business planning under time pressure. Participants complete a short strategy task in the footwear industry: they write an initial business plan in 15 minutes (minimum 150 words) structured around problem identification, solution development, market analysis, and go-to-market strategy. After submitting the plan, they are randomly assigned (block randomization) to one of five conditions that differ in the type and source of feedback they receive. In the control condition, participants receive an AI-generated alternative business plan (“second opinion”) rather than a critique. In the other four conditions, participants receive critical feedback that challenges the plan’s weaknesses and assumptions: either a Devil’s Advocate style critique (adversarial but constructive) or a Ruthless Gatekeeper critique (hyper-critical, blunt, and uncompromising). Each critique style is delivered either as AI-generated feedback or as human-curated feedback (an experimenter reviews and approves AI-generated feedback before it is shown to the participant). Participants then revise their business plan in another 15-minute writing period (minimum 150 words). The study collects the initial plan, the feedback shown, and the revised plan, alongside post-task survey measures and demographics. The primary outcomes focus on changes in plan quality from initial to revised versions, assessed later using both human ratings and AI-based evaluation. Additional outcomes include perceived helpfulness of the feedback, self-reported incorporation of feedback into revisions, and how objective or personally judgmental the feedback felt. The study also measures creative self-efficacy and general attitudes toward AI to explore whether individual differences relate to how feedback is received and used. Expected outcomes are not restricted to a single direction. The study is designed to detect whether critical “devil’s advocate” feedback leads to stronger revisions than receiving an alternative plan, whether a harsher critique style produces additional benefits or diminishing returns, and whether the source of critique (AI vs. human-curated) changes how participants react to and implement harsh feedback.