LLM citation bias research brief

OpenAI’s latest research brief documents systematic citation bias in commercial LLMs: models overweight brands with dense Wikipedia presence, large news volume, and consistent entity naming across languages.

Bias patterns observed

• Recency bias: Press from the last 18 months influences citations more than older authority signals.
• Entity consistency: Brands with aligned names across JSON-LD, Crunchbase, and press releases see 40% higher mention rates.
• Category leaders: Models default to market share leaders unless challengers have superior structured proof.

Brands can counter bias by normalizing entity graphs and publishing machine-readable fact sheets that models can retrieve at answer time.