Incumbent advantage: with identical specs, a well-known brand is recommended 100% of the time
Three experiments on skincare: identical specifications produce near-total dominance for known brands — and less than a +0.1-star edge for a competitor breaks it.
Key numbers
- Setup: skincare, a category where consumers cannot judge quality before buying and must rely on reputation, tested across three commercial models (GPT-4o-mini, Claude Sonnet, Gemini 3 Flash), with a robustness check on search goods.
- Experiment one, conditional monopoly: when every product has the same specifications, well-known brands are recommended 100% of the time (IAI = 10.0) — but the dominance disappears with less than a +0.1-star advantage for a competitor.
- Experiment two: authority-style marketing language, including fabricated clinical-evidence claims, breaks the monopoly at a Bias Surplus Value equal to +0.17 rating points, with each model responding differently.
- Experiment three: multi-brand GEO competition is a social dilemma — when every brand adopts the same optimisation strategy, individual payoff in their proxy falls from +0.802 to +0.007, while non-participating brands received zero recommendations in their tests.
Method and sample
arXiv paper “Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems” (2026-06-16): three experiments on skincare products across three commercial models measuring monopoly under identical specs, the bias surplus of authority language, and payoff changes when multiple brands optimise at once, with a robustness check on search goods.
What it means for brands
Competing in AI recommendations is not a linear ranking: when specs tie, name recognition is everything, yet even a tiny differentiator (under 0.1 stars) loosens the monopoly. The other warning: if every brand runs the same GEO playbook the payoff evaporates, and the brands that sit out get no exposure at all.
How SourceSight handles it
SourceSight's competitive benchmark lists which brands appear on the same answers, so you can see who AI favours for each question type; the prompt library is grouped by how customers ask so the gap where a competitor wins and you are absent gets fixed first.
Limits
The experiments cover skincare across three models, and IAI and Bias Surplus Value are the paper's own constructs; they establish mechanism rather than the recommendation distribution of every category.
References
Run it on your own project
What the same prompts return across 10 engines is only knowable by running them.
Related research
What Chinese generative search actually cites: 8.3% brand selection, and app vs web cite different sources
Across 160,860 citation-level records, the same platform surfaced systematically different source sets on its app and web interfaces.
Sources · 2026-10-06Does AI judge content or clout? A causal test of authority bias
Hold the title and abstract constant, change only author prestige, venue and citation counts, and eight models change their recommendation — while claiming they do not.