Blogs / PageSpeed Says My Site Is Slow but It Feels Fast: Which Is Right?

PageSpeed Says My Site Is Slow but It Feels Fast: Which Is Right?

Nov 21, 20267 min read
Pulkit Khurana

Founder, SproutMe

An upright rain gauge collecting over time, illustrating PageSpeed Insights field data vs lab data for judging real site speed.

You open your site on your phone and it snaps into place. Then PageSpeed Insights gives it a mobile score in the red, and your developer quotes weeks of fixes. Trust the wrong reading and you either pay for work nobody notices or ignore a problem your visitors feel.

The right reading is usually neither your impression nor the headline score. It is the field data at the top of the PageSpeed Insights report: how real Chrome users experienced the page over the last 28 days. If that passes, your visitors find the site fast whatever the lab score says. If it fails or is missing, the lab is closer to the truth than your phone.

What is each half of the report measuring?

PageSpeed Insights puts two different measurements on one page. Most of the confusion comes from reading them as the same thing.

The top section is field data, drawn from the Chrome User Experience Report. It comes from real Chrome users who have opted in to sharing anonymized performance data, collected over a rolling 28-day window. Each metric is reported at the 75th percentile, so the figure shown is one that three in four visits matched or beat. A page passes the Core Web Vitals assessment when those 75th-percentile values sit in the good range.

The large colored number below it is the lab score, and it measures something else. Lighthouse loads the page once, on the spot, in a simulated environment: an emulated mid-range phone on a throttled mobile connection, with nothing cached. That controlled setup makes the test repeatable and useful for diagnosing problems. But it describes one imagined visit, not your audience.

The two answer different questions. Field data tells you how the page performed for the people who actually came. The lab tells you how it would perform for a visitor on a weaker device and network than many of them probably have. That is the core of the PageSpeed Insights field data vs lab data distinction, and it is why a page can fail one and pass the other without either measurement being broken.

Why does the lab say slow when it feels fast?

Your own sense of speed is the least reliable of the three measurements. You are probably testing on a recent phone, on office or home Wi-Fi, on a site you have visited dozens of times. Images, fonts and scripts are already cached. You also know where to tap, so you rarely notice the moment a button fails to respond. That is the best-case visit, and few of your visitors get it.

The lab sits at the other end. Its throttled device and network are deliberately pessimistic. A heavy JavaScript bundle or an oversized hero image that loads instantly on your phone can cost seconds under that simulation. So a poor lab score alongside a fast-feeling site usually means one thing: the page is heavier than it needs to be, and you are viewing it on hardware that hides the weight.

Whether that weight matters depends on who your visitors are, and only field data can tell you that. If the field section passes, your real audience has devices and connections good enough that the weight never reaches them. The lab score is flagging headroom, not a crisis. Field data is also the data Google may use for ranking, per Search Engine Journal's 2021 reporting on the tool. That is the strongest reason to treat it as the deciding reading.

When should you trust the lab score instead?

Field data wins when it exists and covers the page you care about. Two situations break that rule.

The page has no field data

Field data needs enough Chrome traffic within the 28-day window. Newer pages, low-traffic landing pages and campaign variants often show no page-level data at all. In that case the report may show origin-level data for the whole domain instead. An origin-level pass tells you the site is fast on average, not that this page is. With no page-level field data, the lab score is the best evidence you have, so take it seriously.

Field data is poor and the lab is fine

This combination should worry you more than a red lab score. Lighthouse loads one page once and does not click anything. It misses what happens after the first paint:

  • scripts that fire on interaction
  • chat widgets and consent tools that load late
  • heavier pages deeper in the journey

Interaction metrics only exist when a visitor actually interacts, so a single unattended page load cannot fully reproduce them. If real users are struggling and the lab is not, the cause is one of two things. Either it sits in something the simulation never does, or your audience is on slower connections than you assumed. Either way, your fast phone is the outlier, and the field data is telling you what customers live with.

What should you do with the two numbers?

Read the report top down, and let the field section decide.

  • Field data passes at page level. Do not let the lab score drive a rebuild. Treat its diagnostics as a backlog, usually images, unused JavaScript and render-blocking resources, and fix the items that are cheap.
  • Field data fails. Act on it regardless of how the site feels to you, and use the lab diagnostics to find the cause.
  • There is no field data. Treat the lab as your proxy and test the pages that matter most.

"The pages that matter most" is the step teams usually get wrong. Speed work is finite. A landing page carrying paid traffic deserves more attention than one nobody reaches. Whether a slow landing page also raises what you pay per click is a separate question, answered in does a slow website make my Google Ads cost more. For this decision, rank pages by where spend and conversions actually land, not by whichever URL someone happened to paste into the tool.

Then test those pages on the device your customers use, not yours. If most visitors arrive on mobile from paid social, an evening on a mid-range phone over cellular data will teach you more than another run on office Wi-Fi. For the wider picture of how your site holds up on phones beyond this one report, the field-versus-lab question is only the starting point.

Conclusion

Neither the headline score nor your own phone is the authority on how fast your site is. The field data is: real Chrome visits over 28 days, judged at the 75th percentile.

  • When field data passes, a red lab score is a list of improvements, not an emergency.
  • When it fails, your fast-feeling phone is the exception, not the rule.
  • When it is missing, the lab is the best evidence available.

Fix by field data, diagnose with the lab, and start with the pages your budget depends on.

To see which landing pages carry your paid traffic and how they convert, SproutMe Companion answers that from your connected Google Ads and Analytics data.

Frequently Asked Questions

Lab scores vary from run to run even when nothing on the site has changed. Network conditions, server load, cache state, the machine running the simulation and third-party content such as ads or embeds all shift slightly between tests. A swing of a few points is normal. When comparing before and after a change, run the test several times and use the middle result.

Slowly. Each field data update covers a rolling 28-day window, so a fix shipped today is diluted by weeks of older visits recorded before it. Expect field metrics to move gradually, and to reflect the change fully only after about four weeks. In the meantime, use the lab score to confirm the fix behaves as intended.

Page-level data describes the URL you tested; origin-level data aggregates every page on the domain. Use page-level data whenever it is available, because a fast, heavily visited homepage can lift the origin average while a heavy landing page beneath it fails. Origin data is a fallback for pages without enough traffic. It describes the site, not the page in front of you.

Grow smarter with AI marketing tips

Join our newsletter to get practical insights, automation ideas, and performance tips straight to your inbox.

Get a complimentary audit to uncover AI opportunities hidden in your data.

Put these strategies to work