Characterizing adversarial GEO via implicit brand injection

Evidence record

Paper evaluating natural-language brand placement in retrieval documents without imperative prompt injections.

Read the original source ↗

Claim-level citations (2)

• supportsOne natural-looking sentence could bias an AI answer toward an inserted brand: The implicit brand injection benchmark reached up to 82.2% retrieval Hit@10 and 62.7% generated brand mentions in selected article-conditioned conditions using fictitious brands.

  Abstract; Results 5.4, article-conditioned endemic setting

• supportsOne natural-looking sentence could bias an AI answer toward an inserted brand: The implicit brand injection study reports that inserting brand mentions into otherwise plausible retrieved articles raised brand visibility in generated answers under its selected benchmark conditions.

  Abstract; IBIBench benchmark; results on brand mention and propagation