How to Perform a Website Speed Test and Read the Results

By Dave TanguayPublished Updated Editorial draft

Quick Answer

To test website speed, choose a representative URL, run PageSpeed Insights for mobile and desktop, and read its real-user and laboratory sections separately. Save the URL, date, conditions and findings. Repeat comparable tests before and after a controlled change, then verify the customer tasks affected by that change.

A useful speed test produces a specific investigation. “The score is low” gives a developer less to work with than “the main product image starts downloading late on this mobile page.” Begin with the page and behavior you need to understand.

Choose pages that represent the site

Include more than the homepage. Select an important service or article page and any different template involved in a customer task. For a store, include a collection and a product with representative options. Add URLs where customers report delays.

Record the intended visitor state. A public URL test does not reproduce an authenticated account, an already-filled cart or every step of a booking journey. Those interactions need their own authorized browser tests.

Run PageSpeed Insights and preserve the report

  1. Open PageSpeed Insights and enter the exact public URL.
  2. Check the final URL after redirects so you know which page was tested.
  3. Review mobile and desktop results, preserving each separately.
  4. Save the report or a record of the URL, test time, environment and relevant findings.
  5. Repeat the controlled test to see whether a suspected issue is consistent.

Do not compare a desktop report with a mobile report as if they were the same test. When using other tools, record their locations, connection settings and devices too. Different conditions can produce different measurements without a website change.

Read real-user data in its proper scope

Google’s PageSpeed Insights documentation explains that its field data comes from the Chrome User Experience Report over a trailing 28-day period. Check whether the report represents the individual URL or the broader origin. An origin summary is not a measurement of every page independently.

Field data requires sufficient samples. Its absence does not establish that the page is fast or slow. It also will not instantly reflect a repair made this morning.

Review the Core Web Vitals separately: LCP concerns loading, INP concerns responsiveness to interactions, and CLS concerns unexpected movement. The current Web Vitals guidance describes their field thresholds. Our Core Web Vitals guide explains how to investigate each type of problem.

Use lab diagnostics to form a hypothesis

The laboratory section uses Lighthouse under simulated conditions. Review the elements and requests involved in a delay rather than treating the overall score as a repair instruction.

  • Main content appears late: Inspect when the important resource is discovered and what delays its rendering.
  • The browser is busy: Investigate long-running script work and its owner.
  • Content moves: Look for media without reserved space or content inserted after the layout appears.

Total Blocking Time can help identify lab responsiveness problems. It does not replace field INP, and a load test alone cannot reproduce every later interaction.

Write a test record someone can use

Illustrative record: URL: the product page under review. Device: the report’s mobile configuration. Observation: the main image request starts after a script finishes. Hypothesis: resource discovery is delayed. Next check: inspect the rendered markup and request sequence. This is a sample investigation record, not a measured result from a client website.

Keep observations separate from suspected causes and proposed changes. A third-party request appearing in the report does not by itself establish that removing it is the right solution.

Retest after a controlled change

Preserve a recoverable version and change one relevant component at a time where practical. Repeat comparable tests. Then check forms, navigation, product choices, cart behavior or other functions touched by the work. A faster page with a broken customer task is not an acceptable repair.

Review field data as new visits accumulate, considering traffic and other site changes. Neither a perfect score nor a passing field assessment guarantees rankings or sales. Bring your saved reports and reproduction steps to a website audit or a scoped development review.

Key Takeaways

  • Test representative page types and the relevant visitor state.
  • Keep field history separate from controlled lab results.
  • Turn diagnostics into a specific hypothesis and verification step.
  • Retest customer functions alongside performance.

Frequently Asked Questions

Why does the score change between runs?

Controlled tests still experience variation in resources and conditions. Preserve the settings and repeat comparable runs before interpreting a small change as an improvement or regression.

Why is there no real-user data for my page?

The dataset may not have enough qualifying samples. Check whether the report falls back to the origin. Missing data is a measurement limitation, not a performance diagnosis.

Does a homepage test cover the whole site?

No. Other templates, content and interactions may behave differently. Include representative pages and investigate important customer journeys separately.

Archives