First, establish which result is green
Your provider sends a screenshot showing a good score. Yet searching for a product or opening a menu still feels slow. Both observations can be valid: a favourable test does not describe every task on a website.
PageSpeed Insights combines an automated test with data from real visits where available. The automated test uses Lighthouse. Core Web Vitals are three measures covering loading, responsiveness and unexpected movements of page content.
Get the complete report before comparing it with your experience. Check the tested page, date and configuration. Establish whether the green result refers to the Lighthouse performance score, an individual metric or the Core Web Vitals assessment.
The Lighthouse scoring documentation explains that performance scores are calculated from metrics and can vary with testing conditions. They support an investigation rather than guarantee every visitor’s experience.
Lab data and real visits answer different questions
Lab data comes from a controlled test. PageSpeed Insights field data comes from the Chrome User Experience Report (CrUX) and covers the preceding 28 days. After a fix, this part of the report may therefore still reflect visits made before the change.
Check whether the data concerns the page you are reviewing or groups pages from the same website. The report calls that grouping the “origin”. Establish which one is shown before drawing conclusions about a particular page. Insufficient data is not evidence of either a good or a poor experience. The web.dev guide to lab and field differences explains the distinction.
The three Core Web Vitals metrics concern different aspects of the experience:
- LCP: when the largest content element in the viewport appears.
- INP: how long the browser takes to present the next visual update after an interaction.
- CLS: the extent of unexpected layout shifts.
The PageSpeed Insights documentation explains the collection window and Core Web Vitals assessment. The Lighthouse score and that assessment are separate results; their green indicators do not mean the same thing.
A worksheet for looking beyond the score
| Item | Record | Purpose |
|---|---|---|
| Tested page | URL and page type. | Avoid extending a homepage test to the whole catalogue. |
| Configuration | Date and mobile or desktop mode. | Establish which comparisons are meaningful. |
| Data source | Lab or field; URL or origin where specified. | Understand which experience the data represents. |
| Outstanding task | The specific action that feels slow. | Check whether the report addresses that concern. |
| Decision | What is known and what still needs reviewing. | Choose a useful additional check. |
Use the information actually provided rather than filling gaps with assumptions. If the screenshot does not identify its URL or source, recovering those details is the next step.
Worked example: the page loads, but the menu is slow
Fictional situation without measured scores or timings: a page performs well in a loading test, but tapping its mobile menu button produces no visible response for a noticeable interval.
The report describes the tested load. The observation raises a different question: what happens when someone tries to open the menu? Reproduce that interaction under documented conditions and check when its visual response appears.
INP helps assess responsiveness, but does not necessarily measure the full time needed to complete a task. If a loading indicator appears promptly while server results arrive later, that later wait also needs measuring. The INP definition distinguishes the next visual update from an interaction’s subsequent effects.
The technical team may need to inspect requests and browser activity. The outstanding test is the reported interaction, rather than another homepage screenshot. Its cause remains unconfirmed until the evidence has been reviewed.
If you have not located the affected areas, start with the worksheet for slow pages and journeys.
How to compare speed before and after a change
Agree what needs to improve: the main content appearing, a filter responding or a form step working more smoothly. Each objective needs a test suited to the task.
Record the pages, procedure and conditions before and after the work. Keep multiple observations when results vary and explain environmental changes you could not control. A higher score alone does not demonstrate an effect on sales or search rankings.
If you are commissioning the work, include these criteria in the optimisation quote and acceptance checks. They make the delivery easier to assess.
What to do with the apparent contradiction
Keep both the favourable result and the observation of slowness. Each provides context. Identify the unanswered question and choose a test that can address it.
A performance assessment can relate these findings to representative journeys. When several improvements compete for attention, prioritise them using evidence and impact, keeping outstanding checks visible.