prefetch: fetch a document ahead of time, but not its subresources like CSS or JS.
prerender: fully download, render, and load pages into an invisible tab, including its subresources and running JavaScript.
Prerendering may change analytics numbers and increase server traffic and bandwidth costs as the result of eagerly fetched subresources.
Coordinate with your analytics and hosting teams before deploying.
For example,
Clarity collects data for prerendered pages even when they were never visited.
prerender_until_script: fetch a document ahead of time including its subresources and renders the page until first encountered script. A middle ground between prefetch and prerender.
default: list ("urls") rules default to immediate and document ("where") rules default to conservative when no eagerness was provided.
Chrome limits:
Max 2 kept in memory.
Canceled speculation can be triggered again, pushing out oldest speculation.
eager: A hint to the browser that it should begin prefetching/prerendering at the earliest signs of user interest (e.g. nearby cursor movement, brief hover, or prominent visibility).
Best for aggressively speeding up many potential navigations, even if some won’t be used.
Chrome limits:
Max 2 kept in memory.
Canceled speculation can be triggered again, pushing out oldest speculation.
moderate: A hint to the browser that it should start prefetching/prerendering when there’s a reasonable indication the user may click soon (e.g. in-viewport plus sustained hover/focus).
Best for balancing performance improvements with resource usage.
Chrome limits:
Max 2 kept in memory.
Canceled speculation can be triggered again, pushing out oldest speculation.
conservative: A hint to the browser that it should delay prefetching/prerendering until the user is actively initiating a click (e.g. pointerdown/mousedown).
Best for minimizing wasted bandwidth/CPU while still improving perceived navigation speed.
Chrome limits:
Max 2 kept in memory.
Canceled speculation can be triggered again, pushing out oldest speculation.
