In June 2026 we ran an internal four-person panel review on our research methodology. The review identified specific corrections we needed to make to the published numbers, the framing of what we measured, and the model specifications we relied on. The substance of what we found is still useful, but the version that was on this page did not meet the bar we want to hold our own research to.
We pulled the page while we rebuild the analysis on the larger 1.18 million site dataset we have today, with proper popularity and category controls, and re-publish with the corrections.
Updated analysis will appear here when it is ready. We expect this to take about a week.
Published 2026-06-12.