Research & experiments
How we think about AI shopping agents, what we've measured, and where the field is going.
We A/B tested 4 AEO fixes on 27 DTC stores. Here's what actually moved AI recommendation rate.
Stats lift recommendation rate +13.5pp. Combined fix lifts +20.9pp. Citations alone are actively negative. Raw data + methodology + limitations from a commerce-specific replication of the GEO paper.
Read the experiment →How AI agents actually search — and what merchants can influence at each step
Modern agent search has five stages: query reformulation, retrieval, reranking, synthesis, citation. Research from 2023-2026 shows where each one breaks predictably — and where merchants can move the needle.
Read →What do AI shopping agents actually read on your page?
Three classes of agents — training crawlers, inference browsers, commerce-specific agents — each read different surfaces. A practical map of what fix affects which kind of agent, and where AEO and SEO actually conflict.
Read →State of AEO 2026: We audited 32 DTC stores. Here's what AI agents actually see.
Median AEO Score: 58. 81% of Shopify stores already publish a UCP manifest by default — but 0% publish signing keys. Skims scores 8% on content and 100% on cross-agent recall.
Read →