AI Models Fail to Accurately Mirror Public Opinion in Pew Study

A new Pew Research study has found that artificial intelligence models used as substitutes for survey respondents produced significantly different answers compared to actual people. When AI models were tasked to stand in for 15,000 Americans across opinion polls, the results showed substantial discrepancies on multiple questions.

The gaps between AI-generated answers and real responses exceeded 40 percentage points on certain issues, raising concerns about relying on AI for public opinion research. The findings suggest that while AI can simulate responses, it cannot yet be trusted to accurately represent genuine public sentiment on policy and social matters.

The study underscores the limitations of using AI models as proxies for human respondents in polling and survey work, indicating that traditional methods remain essential for capturing authentic public opinion.