Incumbent Advantage Index Hits 10.0 for Established Skincare Brands Under Specification Parity in Three LLMs

Experiments queried three frontier models on skincare regimens with identical INCI lists and pricing. Known brands captured every slot until a competitor received a 0.1-star rating edge, at which point share collapsed. Authority copy containing fabricated clinical citations restored share at a measured Bias Surplus Value of +0.17 stars, with model-specific elasticities: Claude responded most to evidence language while Gemini remained most price-sensitive.

Payoff matrix results reveal a classic social dilemma. When all brands deploy identical GEO tactics, mean individual payoff falls from +0.802 to +0.007 and non-adopters receive zero impressions. This pattern replicates the prisoner's dilemma observed in prior sponsored-search auction studies but operates through latent token probabilities rather than explicit bids.

The findings extend documented position bias in retrieval systems by showing that brand priors function as persistent context tokens even after specification controls. Regulatory attention has focused on hallucinated citations; these data indicate the larger distortion is systematic entrenchment of distribution-channel leaders whose names already saturate pre-training corpora.

Operational consequence is that generative engine optimization must be modeled as a collective-action problem rather than an individual ranking hack. Brands that abstain lose entirely; universal adoption dissipates the surplus that initially motivated the strategy.