Distinguishing AI from Human Output

People can distinguish AI-generated content from human-created work at statistically significant rates, even when the AI output is objectively “better” on measurable quality dimensions.

In controlled studies, participants correctly identified AI-generated personas (p = .002) and human-crafted personas (p = .003). They did this despite AI outputs scoring higher on informativeness, consistency, and clarity.

Key detection signals participants reported:

• “Robotic” tone and unusual vocabulary choices
• Overly positive framing without acknowledged limitations
• Thematic over-coherence (hobbies matching professions)
• Absence of emotional or personal specificity
• Generic rather than idiosyncratic detail

The implication: humans have developed intuitions for authentic human expression that AI hasn’t yet matched. Whether this gap will close or widen remains unclear.

Related: 05-atom—polish-authenticity-paradox, 05-atom—flaws-as-authenticity-signals