Synthetic respondents—AI-generated panels that simulate survey responses—have emerged as one of the most heavily marketed trends in market research. Vendors promise speed, scale, and cost savings. The technology is genuinely impressive at producing outputs that look like human responses.

But looking like a human response and being one are not the same thing.

This paper identifies four structural problems that make synthetic panels unsuitable for legitimate insights research. These are not implementation issues that better engineering will solve. They are fundamental flaws baked into the underlying technology.

The paper closes with a practical alternative: AI-powered enterprise data mining that scales real insights without fabricating respondents.