What was each crawler's request for? Pages, render resources, 304 rechecks, redirects

A "Googlebot 3xx 28.8%" figure hides two different things: redirects, which cost a hop each, and 304 Not Modified answers, which are the crawler checking a page that has not changed. On one site the split was 2,270 redirects and 2,258 rechecks, and almost half of verified Googlebot's requests were fetching scripts to render pages, not pages. The new crawl purpose view reads every request for what it was.

What changed

• Crawl purpose per bot on the Bots page: pages, render resources (scripts, stylesheets, JSON, fonts and Next.js build paths), media, other files (PDF, XML, text), 304 rechecks, redirects, refused (401/403/429/451) and errors, with each bucket's share and a render share. A 304 is never counted as a redirect. Googlebot Smartphone, Googlebot and Googlebot-Image stay apart.
• Bytes by bot and bucket, where the feed carries page sizes, and error-page weight: the average and largest body your 404, 410 and 5xx responses send, and the heaviest 404 paths. Where the feed does not measure sizes the figures are off and the page says so, never shown as zero.
• Pages first. Requests, URL patterns and the other page-level SEO lists default to HTML pages; "Assets" (everything that is not a page) and the individual file types are one click away.
• GET /public/v1/websites/{id}/bots/purpose carries the same figures, with verified_only and limit; GET /public/v1/definitions explains crawl_purpose, render_resources, rechecks_304 and error_page_weight. The MCP server adds get_crawl_purpose (1.25.0) and the CLI bots-purpose (2.5.0). The file_type filter accepts assets.

See crawl purpose in the help.