Auditing Public Influencer Marketing Benchmarks: Synthetic Signatures & Evidence Validation
Aditya Harsh (BITS Pilani WILP, Sept 2026). Report Reference: 2024HB21151.
Public Files Displayed Synthetic Signatures
Structural analysis revealed impossible future post dates (up to year 2432), frozen follower counts across posts, near-zero cross-field correlations, and near-uniform view distributions.
Testable Reference Claims Reproduced
Published industry claims (e.g. "reach predicts sales, r = .37") failed empirical reproduction on audited datasets (observed r = 0.003).
Audited Benchmark Datasets
Internal campaign structure showed that log ROI variance is driven by reach (29.1%), CTR (23.8%), inverse spend (19.7%), click-to-conversion (15.1%), and AOV (12.3%). Campaign category, language, and month showed no effect.
Fake follower share correlated positively with reported engagement (\(\rho = 0.44\)) but negatively with video views (\(\rho = -0.29\)). Isolation forest anomaly scoring successfully isolated engagement inflation.
Exhibited near-uniform views, followers, and engagement rates across 5,000 creators, with zero disclosure label or category-fit effect.
Contained invalid timestamps extending into future centuries (up to year 2432) and near-zero Spearman correlations across related interaction variables.
Summary of the 50-Test Battery
Benjamini–Hochberg false-discovery-rate correction was applied across 50 hypotheses. Only 8 tests retained statistical significance, and most carried negligible effect sizes (ε2 < 0.01).