ChatGPT referral growth study: AEO effects dwarfed by platform tailwind

A new paper on arXiv (arXiv:2606.04362) from researchers Keisuke Watanabe and Kazuki Nakayashiki presents a rigorous natural experiment that disentangles the effects of Answer Engine Optimization (AEO) from the rapid platform growth of ChatGPT. The study uses first-party logs from a high-traffic domain hosting hundreds of thousands of YouTube Q&A pages. In January 2026, a subset of pages received a defined bundle of AEO interventions, while the rest of the same domain served as a contemporaneous control group, absorbing the platform-level tailwind.

The results are striking: raw ChatGPT referral traffic to the treated pages grew 5.7x over the study period, but untreated pages on the same domain grew 3.5x—showing that most of the growth was driven by ChatGPT's expanding user base, not AEO. An interrupted time-series model on the treated/control ratio estimated a discrete intervention-aligned level increase of 1.82x (95% CI 1.31–2.54, p=0.001), but a conservative placebo-in-time permutation test yielded p=0.16, meaning the effect is suggestive, not conclusive, due to a short and noisy pre-period. Importantly, Google organic traffic to treated pages was not harmed, consistent with the SEO-protection rule. The authors conclude that headline AEO multiples substantially overstate the causal effect, and their methodological approach—using an on-domain control to separate treatment from platform tailwind—is the paper's most significant contribution.