In January, Googlebot quietly changed how it allocates render budget. The hints were subtle at first — slightly fewer fetches on large catalogues, a tilt toward pages with cleaner internal linking — but by mid-March the pattern was unmistakable across every log file I touched.
What the logs actually show
I pulled raw access logs from 47 client sites between January 1 and April 30, normalised the user agents, and plotted fetch counts against page templates. Three patterns stood out, and none of them are what the AI-overview chatter on Twitter would have you believe.
- Render-heavy templates lost roughly 18% of their fetch share to lightweight templates with the same internal link weight.
- Pages requiring a second-pass render were re-queued 2.3× more often than in 2025, but the queue itself drained faster.
- Sites with a sitemap + lastmod that actually reflected reality saw fetch parity hold steady. Sites with stale sitemaps lost ground.
The new mental model
Crawl budget in 2026 isn't a single bucket. It's three: HTML-only fetches (cheap, abundant), render fetches (constrained, prioritised by recency signals), and answer-engine fetches (a separate lane entirely). Most of the 'lost traffic' I've audited this year traces back to template choices that pushed pages into the render lane when they didn't need to be there.
If the page can render its primary content from HTML alone, ship it that way. Every dependency on hydration is a tax on your fetch share.
The playbook
Audit your templates by lane, not by URL count. Map every template to one of the three buckets. Anything that doesn't have a hard requirement for client-side rendering should move to HTML-first. Then re-audit your sitemaps — Googlebot now treats lastmod as a strong prioritisation signal, not a hint. Lie to it and you'll be deprioritised within a fortnight.
I'll publish the template-classification spreadsheet I use with clients in next week's essay. If you want a head start, the short version is: server-render anything that earns rankings, and stop apologising for boring HTML.