Context
Peer-reviewed SafeGEO research evaluated 22 GEO source-manipulation variants across 600 recommendation cases. In tested configurations, flawed targets entered top-three recommendations up to 83.2 percentage points more often; evidence-breakdown mitigation reduced harmful promotion by up to 39.2 points but did not restore the clean baseline.
What changed
The study introduces a utility-grounded benchmark for GEO manipulation of recommendation agents, distinguishing citation/visibility optimization from user-constraint violations and measuring five developer-side mitigation approaches.
Why it matters
AI shopping and agentic product discovery can be distorted when seller-controlled source text misrepresents product fit. Recommendation audits must evaluate hard constraints and evidence support, not just whether a product is mentioned or cited.
Visibility OS interpretation
Visibility OS interpretation: source visibility and recommendation utility are separate metrics; independent evidence checks may reduce harm but should not be treated as a guaranteed defense.
What remains uncertain
How well do these controlled results transfer to live retrieval, other product markets and deployed shopping agents? Can provenance checks or cross-source contradiction handling close the remaining gap?