Removing Performance Guesswork From Early Hints and Fetch Priority
Cloudflare is announcing Smart Hints, a system designed to automate resource prioritization by combining Early Hints and Fetch Priority with the company's network-wide performance data. Since launching Early Hints roughly a year ago, Cloudflare has delivered around two trillion hints on behalf of more than 150,000 websites. Lab tests at launch showed page load improvements exceeding 30% for sites that properly configured the feature.
The core challenge with Early Hints has always been configuration. To use it effectively, developers must specify which assets to preload or which third-party servers to preconnect, usually via HTTP link headers or HTML link elements. That requires understanding render-blocking resources and avoiding prioritization strategies that saturate network bandwidth on non-critical assets — you cannot simply hint everything and expect good results.
Smart Hints aims to remove that burden. Using Cloudflare's real user measurement data and heuristics, the system will decide which assets need priority for a given response, without requiring developers to configure anything at the origin or in a Worker. The company is now opening a closed beta and is taking signups from the Speed Tab of the Cloudflare dashboard.
Why Resource Ordering Matters
When a browser receives an HTML response, it parses the document sequentially from start to finish. The response is split into two parts: the <head> and the <body>. The <head> contains essential stylesheets and scripts. Stylesheets are critical because browsers wait for them before rendering visible content. Scripts behave differently depending on their attributes: without explicit defer or async instructions, a script becomes a blocking resource that halts page processing until it is fully loaded and executed.
Until the browser reaches the <body>, nothing is visible to the visitor. The faster the <head> is optimized — minimizing the download time for stylesheets and blocking scripts — the sooner rendering can begin. Browsers generally handle the priority ordering themselves, but that default behavior is not always optimal for every page.
How Early Hints and Fetch Priority Work
Early Hints is a server-to-client mechanism that sends headers before the final response is ready. When a client requests a page, the server can respond with a separate early hint containing instructions about important static objects and links to related resources. The browser uses that advance notice to begin preloading and preconnecting while the full response is still being prepared, which can significantly cut page load times.
Fetch Priority — previously called Priority Hints — gives developers explicit control over resource loading order. Browsers natively assign priorities based on resource type, page placement, and position in the HTML document. For instance, images in the visible viewport are typically prioritized above other assets, while critical early scripts in the <head> may receive very high priority. Fetch Priority allows developers to override those heuristics and assign higher or lower priorities to specific resources.
Both tools are effective but demanding. They require access to HTML resources and knowledge of how to prioritize assets against each other. Since web pages evolve, these configurations need ongoing testing and maintenance. Smart Hints is intended to make those strategies self-managing.
Sourcing Bid Data for Smarter Decisions
Cloudflare gathers performance information through two channels: synthetic testing, which loads pages in an automated browser within a controlled environment, and Real User Measurements (RUM), which collects data from actual visitors. RUM works by injecting a small JavaScript beacon into the page. Cloudflare aggregates RUM data across thousands of sites to analyze loading bottlenecks.
When those bottlenecks trace back to render-blocking resources, Smart Hints can generate hints that tell the browser to prioritize those downloads. As the rollout proceeds, Cloudflare plans to explore learning models to identify patterns that can be converted into reusable heuristics — potentially improving performance for sites that don't run RUM or synthetic tests themselves.
For the time being, Cloudflare is building a beta cohort to test the system at scale across different content types. Participants will help verify that the performance decisions hold up before the feature is broadly released.
