A contractor sends a report: PageSpeed Insights used to show 90, now it shows 60. You changed nothing, installed no new plugins, and the number dropped as if the site had been broken. The first thought is to pay for “optimization” right away. It is calmer to first find out what exactly this number measures, because often it describes not your customer but a single test run.
Below: why this figure jumps, which data to look at instead, what to do if that data does not exist for your site, and which questions to ask the person who reports to you with one number.
Two different pictures on one page
A PageSpeed Insights result combines two kinds of data, and Google’s documentation separates them clearly.
Lab data is a simulation. The service opens the page in a controlled environment: one imaginary device and network conditions set in advance. This kind of test helps find the cause of a problem, but, as the documentation says, it may not capture what happens in real life. The familiar score from 0 to 100 comes from the lab test. The “good” zone starts at 90, the middle one (“needs improvement”) covers 50–89, and everything below 50 Google calls poor.
Field data is measurement of real visitors. The service shows it for the previous 28 days. There is no imaginary phone here: these are people who really opened your page.
For a business owner the difference is simple. The lab score answers “how does the page behave in one test run”, while the field data answers “what was it like for the people who came”.
Why the score jumps when the site was not touched
The formula itself is a weighted average of the scores of several metrics. But plenty outside your site affects the result.
- Conditions of the run. The Lighthouse developers (Lighthouse is the program that calculates the score) list seven sources of variation: the network at home or in the office, the backbone network, your server’s response, the power of the device, other programs using the processor and memory at that moment, the quirks of the browser itself and the variability of the page itself. The last one means, for example, that the page runs a test of two design variants or shows an ad that is different every time.
- A different number of runs. According to them, the median of five runs is twice as stable as a single result. That is why the documentation directly advises running the test several times and not drawing conclusions from one number.
- A program update. The scoring documentation states that the weights of the metrics in the formula have changed over time because the team regularly researches what matters more to people. So the same site can get a different score in a new version. A fresh example: on 27 October 2026 Lighthouse 13.6.0 was released on GitHub, and its notes mention an update to the simulation model. Whether PageSpeed Insights already runs on that version we do not know, and we found no official confirmation; the example is given only to show that the program gets updated and the result may shift along with it.
So a fall from 90 to 60 with no edits on the site is possible without any disaster. That does not mean “the score can be ignored”: if after several runs it stays low, there is something to look into.
What to look at instead of one number
First: field data for 28 days. According to the PageSpeed Insights documentation, the Core Web Vitals assessment (a set of three measures of how comfortable a page is: how fast the main block appears, how fast the page reacts to a tap, and whether it “jumps” while loading) counts as passed if the 75th percentile of all three measures falls into the “good” category. Put simply: at least three visitors out of four had a “good” experience on every measure. If this check is passed, 60 in the lab test is not a disaster, just a reason to check whether the conditions changed.
Second: the Core Web Vitals report in Google Search Console. This is a free Google service where a site owner can add their site. The report takes data from the same field measurements, shows the last 28 days, splits the result into phones and computers, and sets the status of a group of pages by the worst of the three measures.
Third: what search says. In its description of page experience, Google Search states that Core Web Vitals are used by its ranking systems and recommends that site owners achieve good values. Right next to it, it says that chasing a perfect score just for SEO reasons may not be the best use of your time. So the goal is not a “green circle” but a passed check and a page that does not annoy people.
How speed affects advertising we covered separately in the article “Slow website hurts sales even with good ads”: it has the ten-minute check and what you can fix yourself.
What to do if there is no data on real visitors
On small sites this block is often missing from the report entirely, and that is normal. The documentation explains the order: when a specific page has too little data (it was just published or has few visitors), PageSpeed Insights shows data for the whole site. If there is none there either, the service says directly that it cannot show real-user data.
The reason is how Google collects this data. Only pages and sites with a minimum number of visitors get into the Chrome UX Report dataset; Google does not disclose the exact number. Only Chrome users on desktop and Android who agreed to share statistics are counted; visitors on iPhone, in other browsers and in apps’ built-in windows are not in the sample. In Search Console, a URL without enough data does not appear in the report at all.
What to do in that case:
- Run the test not once but five to seven times, write down the results and look at the median (the middle value of an ordered series), not the best or the worst result.
- Compare with yourself: today’s number against the number a week ago, measured the same way: the same page, the same device type, the same time of day.
- Look at the causes in the lab report: large images, extra scripts, a slow server response. These show up even without field data.
- Check not only speed but also the outcome: whether enquiries and calls come in. If they do, the site works well enough.
What to ask a contractor who reports with a score
Something you can check yourself: open pagespeed.web.dev, enter the same address and run it several times. And here are the questions that only the person who took the measurement can answer:
- Which page and which device? Mobile and desktop results differ, and so do the home page and the page the ad leads to.
- How many runs were there, and which number is in the report? One run or the median? Did it drop after one run or consistently?
- What changed between the two measurements? A widget, a variant test, a new ad or a tag may have been added to the site; the measuring program may have been updated.
- What does the real-visitor data for 28 days show, and is Core Web Vitals passed? If it is missing from the report, they should say why.
- What work do you propose and what does it depend on? A good answer names a specific cause from the report (large images, server response), not “I will raise the score to 90”.
A contractor who can answer these five questions usually knows the job. If instead you are offered “package optimization” with no specific cause, be wary.
From our practice
When we review a site before launching ads, we check speed too, but we look at it next to the outcome: how many enquiries come in and from which device. In our experience (an observation, not statistics), what we meet more often is not a slow site but a report with a single number that nobody can explain. That is why we ask to see not only the score but also the conditions under which it was obtained.
Summary
- The PageSpeed Insights score comes from a simulation of one load on an imaginary device, so it jumps between runs and can change after a program update.
- Real visitors are shown by the field data for 28 days; the Core Web Vitals check is passed if the 75th percentile of all three measures is “good”.
- Google Search recommends good Core Web Vitals, but a perfect score just for SEO may not be the best use of time.
- No field data: run the test several times, compare medians, and look at the causes in the report and at enquiries.
- To a contractor who reports with one number, ask five questions: page and device, number of runs, what changed, field data, what exactly to do.
See how we set up advertising in the search advertising and Performance Max sections.
Want a free audit of your Google Ads account? Send us a request: we will look at your account and the page your ad leads to, and tell you what is worth fixing first.