Humanizer sites deploy mystification through opaque algorithm references, scale claims, and unexplained success-rate claims¶

Claim · All claims
Evidence · 1 study (2 entries) · qualitative r2–r3 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Sites cite advanced algorithms, parameter counts, and success rates (e.g., 99%) without explaining how humanization works, obscuring the underlying mechanism. → Roe 2026

Evidence¶

Roe 2026¶

Roe, J., Perkins, M., Bannister, P., Furze, L., & Wood, J. (2026). Dramaturgies of Deception: AI Humanizers and the Performance of Legitimacy in Higher Education Assessment. A preprint. https://arxiv.org/abs/2605.02649 doi:10.48550/arxiv.2605.02649

q2 · i? · qualitative · r3

MCDA of Website 1, which scored 5 on visibility and reach, referenced 'state-of-the-art natural language processing algorithms' and 'deep learning models', and claimed to humanize over 1 billion words of text a month.

"Mystification is further solidified by the repeated mention of a 99% success rate in evading detection, which serves to enhance confidence yet gives no indication of how or why such a rate of success is possible."

Roe 2026 (2)¶

Roe, J., Perkins, M., Bannister, P., Furze, L., & Wood, J. (2026). Dramaturgies of Deception: AI Humanizers and the Performance of Legitimacy in Higher Education Assessment. A preprint. https://arxiv.org/abs/2605.02649. doi:10.48550/arxiv.2605.02649

q2 · i? · qualitative · r2

MCDA of Website 3, which had the most sophisticated deployment of trust props, including prestige university and corporate logos, satisfaction ratings of 4.9/5, and modes labeled fast, creative, and enhanced.

"The use of numbers to suggest complexity is employed, claiming '1.6 trillion parameters' were used in training the humanizer service."

Discussion¶

Related Claims¶

• Humanizer legitimacy performances show leakage: unendorsed logos, inactive social media, and no registered business address — related
• The review identifies data privacy, algorithmic fairness, academic integrity, and unequal access as interconnected ethical concerns in AI-supported mathematics education — related
• Hybridization of authorship blends human and AI-generated content, often without learners' explicit awareness of the boundary, normalizing formulaic rhetorical structures — related
• Humanizer sites frame AI humanization as a rational, ethically defensible response to flawed surveillance and false positives — related