Critical AI Evaluation With Color-Coded Analysis

TLDR: Students interview AI personas when real stakeholders are inaccessible, discovering both the tool’s utility and its cultural limits.

What she’s doing: Julie teaches Global Social Enterprise, a course where students design social enterprises (for-profit ventures with impact as a core organizational goal) that address one of the UN Sustainable Development Goals in a global context. The central pedagogical challenge is structural: students propose solutions for communities in distant geographies but cannot easily interview actual stakeholders during a single semester. To bridge this gap, Julie has students create AI stakeholder personas representing three essential perspectives: beneficiaries of the proposed service, potential funders, and subject matter experts. Students populate these personas with contextual information, then conduct extended text-based interviews to stress-test their business models and surface blind spots they wouldn’t otherwise recognize.

What’s working: The assignment surfaced immediate friction points. Students discovered that AI personas default to encouragement rather than critique: praising ideas that needed harder questions. Some groups reported that their initial personas reproduced cultural stereotypes, prompting them to explicitly prompt for more nuanced, diverse perspectives. Student groups experienced the assignment differently based on how deeply they engaged: teams proposing U.S.-based ventures (e.g. e-bike delivery in New York City) found the stakeholder interviews more intuitive than those working on projects in Uzbekistan or other unfamiliar contexts. The distance between student experience and stakeholder reality became evident.

Students who invested seriously in the exercise reported that AI stakeholder interviews were better than guessing and provided scaffolding for thinking beyond their own assumptions. The act of constructing personas required research and specificity; the interviews themselves forced students to articulate their value propositions to skeptical audiences, even simulated ones. 

One unanticipated benefit: the limitations became teaching moments. Students learned to recognize when AI responses felt too accommodating or culturally thin, developing a critical orientation toward generated content rather than passive acceptance.

The constraints are equally clear. AI personas lack the unpredictability and lived knowledge of actual stakeholders. Students working on ventures closest to their own experience engaged more readily, suggesting the assignment may inadvertently reinforce rather than challenge proximity bias. Julie observed uneven engagement. Some teams treated the exercise as a checkbox rather than genuine preparation for stakeholder engagement. 

What’s next: Julie is expanding the assignment from three to five stakeholder perspectives and extending it across the full semester rather than compressing it into half. She plans to explicitly prohibit students from designing ventures for “people like themselves,” forcing confrontation with unfamiliarity. She’s also integrating library instruction on databases and research tools that AI cannot access, ensuring that persona construction is evidence-based rather than speculative. A new element this semester adds complexity: after launching their social enterprises, student groups will switch roles and become marketing agencies pitching go-to-market strategies for other teams’ ventures—a move Julie anticipates will be completely AI-driven, creating opportunities to examine AI-generated marketing alongside AI-simulated stakeholder feedback.