Database reads dropped significantly across the board this release. The site export API, dashboard next-step card, public scan previews, and the scan directory all reuse recent results instead of hitting the database on every request. Site export edits still appear immediately for brand sites and within a minute elsewhere — nothing about the content changed, just how it's fetched.
Site Analytics was the biggest bottleneck. Missing pages, whole-site AI traffic, and the hourly AI-visit rollup were scanning the entire visit log on every load. New indexes and a rewritten Missing pages query cut that from 7 seconds to under half a second for the busiest accounts.
Brand pages, shared reports, prompt details, and scan status checks now load only the data they actually display. If pages felt sluggish before, they should feel meaningfully faster now.
Token usage in AI monitoring now counts every step of a multi-step generation, not just the last one. Usage numbers will be more accurate going forward, particularly for prompts that involve streaming or chained tool calls.