Study ties agent-ready websites to more AI recommendations

A preprint posted to arXiv September 28 argues that “agent experience,” or AX — whether an AI agent can successfully fetch and read a business’s own site — now matters more than answer-engine optimization’s focus on off-site citations. Running 37,927 simulated buyer-question journeys across four agent harnesses and 1,056 real businesses matched on fame and prior model knowledge, the authors found agent-ready businesses had their AI answers built from their own pages 78% of the time versus 56% for the rest, were clearly recommended 1.9 times more often, and got answers that were 41% more accurate; answers about sites that weren’t agent-ready were 3.7 times more likely to omit facts the buyer had asked for. Only 7-10% of the finished answer came from the model’s training knowledge either way. Co-author Assaf Elovic co-founded Tavily, an AI-agent web-search company Nebius acquired in 2026; the paper lists no institutional affiliation or funding source.