category
Nile Blog & Research articles filed under Research.
A secondary analysis of 3,602 ChatGPT ad placements from a Penn/Haverford dataset. Paid advertisers appeared in the actual answer text only 8% of the time. After controlling for the question asked, the average lift from paying was −0.3 percentage points. Paying and getting recommended are separate systems.
Four AI agents said an in-stock product was backordered. Fixing the facts stopped the wrong answers, but it did not make the product recommended. A conversation with ecommerce practitioner Leo Nguyen on contested data, hallucinated policies, and the gap between being readable and being chosen.
We launched Nile with a small cohort of merchants willing to be early. Here are our honest findings from that first cohort — some of them surprised us.
