Key Takeaways
Why a 94 in PageSpeed Insights changes nothing, what the score is actually made of, and which number decides whether you pass.
- Half the Lighthouse performance score comes from metrics that are not Core Web Vitals, and INP contributes nothing to it at all.
- Google assesses Core Web Vitals at the 75th percentile of page loads, so your worst quarter of sessions decides the outcome.
- Your field data excludes every Chrome on iOS user, every Microsoft Edge user and every Android WebView session.
- HTTP Archive measured 43% of mobile pages passing all Core Web Vitals in 2024, against a Lighthouse score of 90 being routine.
- Google has never called Core Web Vitals a ranking factor. Its own documentation uses noticeably more careful wording than that.
Your score and your assessment measure different things
The number you optimized and the number Google uses are produced by two different systems, on two different populations, under two different sets of conditions.
Quick Verdict: Stop working the Lighthouse score. Open the field data section of PageSpeed Insights, find which of LCP, INP or CLS is failing at the 75th percentile, and fix only that. If your field data is missing entirely, the page has too little traffic to be assessed and the score is the only thing you have, which is a traffic problem rather than a speed one. A site rebuild is worth it when the architecture is the cause, and a staged redesign is enough when it is not.
Google’s own PageSpeed Insights documentation draws the line plainly. It states that PSI provides both lab and field data about a page, that lab data is useful for debugging issues as it is collected in a controlled environment, and that field data is useful for capturing true real-world user experience but has a more limited set of metrics.
| Lab | Field | |
| Tool | Lighthouse | Chrome User Experience Report |
| Population | One synthetic visit | Real Chrome sessions |
| Device | Simulated mid-tier phone | Whatever people actually hold |
| Network | One fixed set of conditions | Everything from fiber to a train tunnel |
| Output | A score out of 100 | A distribution, assessed at the 75th percentile |
| Can measure INP | No | Yes |
PSI is explicit about the lab side: the lab data is based on a simulated load of a page on a single device and fixed set of network conditions, and currently Lighthouse simulates the page load conditions of a mid-tier device, naming the Moto G4.
One visit, one device, one network. That is the 94.
Do this: Open PageSpeed Insights for your top landing page and read only the field data panel. Ignore the score entirely for the next ten minutes.
Read what the Lighthouse score is actually made of
The score is a weighted average of five metrics, and the weights are published.
| Metric | Weight | Is it a Core Web Vital |
| Total Blocking Time | 30% | No |
| Largest Contentful Paint | 25% | Yes |
| Cumulative Layout Shift | 25% | Yes |
| First Contentful Paint | 10% | No |
| Speed Index | 10% | No |
Add it up. Half the score comes from Total Blocking Time, First Contentful Paint and Speed Index, none of which Google uses to assess Core Web Vitals.
Now notice what is absent. Interaction to Next Paint, the only responsiveness Core Web Vital, contributes zero percent to the Lighthouse performance score. It cannot, because a lab run never clicks anything. INP needs a real human doing real things, which only happens in the field.
So a page can score 94 while its interactivity is unmeasured and its real-world responsiveness is poor. That is not a flaw in Lighthouse. It is what a lab is.
Google’s own Lighthouse documentation also warns about treating the number as fixed, noting that a lot of the variability in your overall performance score and metric values is not due to Lighthouse, that fluctuation is usually because of changes in underlying conditions, and suggesting it is more useful to think of site performance as a distribution of scores rather than a single number.
This is the same error as reporting a single average for anything that varies, which is why choosing the metric carefully matters more than improving it.
Do this: Write down your score’s five component values, not the score. If Total Blocking Time is dragging it, note that fixing it may not move a single Core Web Vital.
Check the thresholds against the binding source
There is a whole genre of article claiming Core Web Vitals thresholds changed in 2026. Google’s documentation says otherwise, and the documentation is the thing that decides.
| Metric | Good threshold | Measures |
| Largest Contentful Paint | 2.5 seconds or less | Loading |
| Interaction to Next Paint | 200 milliseconds or less | Responsiveness |
| Cumulative Layout Shift | 0.1 or less | Visual stability |
Google Search Central’s Core Web Vitals page, last updated 10 December 2025, names exactly those three metrics at exactly those thresholds. The web.dev reference page states that LCP should occur within 2.5 seconds of when the page first starts loading, that pages should have an INP of 200 milliseconds or less, and that pages should maintain a CLS of 0.1 or less.
The set does change over time, just not this year. web.dev notes that a stable metric can be retired and replaced by another metric that addresses the problem area more effectively, and First Input Delay is the worked example: INP replaced FID as a stable Core Web Vital in 2024.
I searched for the claimed 2026 threshold changes and found them only on low-authority marketing blogs, with no corroboration in any Google document. My read: when a whole page of search results agrees on a number that the owning organization’s own documentation does not contain, the search results are the thing that is wrong. This is the cheapest check in technical SEO and almost nobody runs it.
Do this: Before accepting any threshold from an article, open the vendor’s own documentation and check its last-updated date. Two minutes.
Your field data does not include everyone
The field number is real, but it is not everyone, and the exclusions are specific.
CrUX is the dataset behind the field panel. Chrome’s documentation describes it as a dataset that reflects how real-world Chrome users experience popular destinations on the web, and says plainly that not all origins or pages are represented in the dataset.
To contribute a session, a user has to meet four conditions: enable usage statistic reporting, sync browser history, have no sync passphrase set, and be on a supported platform. Supported means desktop Chrome on Windows, macOS, ChromeOS or Linux, or Android Chrome including Custom Tabs and WebAPKs.
Here is the part that changes how you read your own dashboard. Chrome on iOS does not contribute. Android WebView apps do not contribute. Other Chromium browsers, Microsoft Edge among them, do not contribute.
So every iPhone visitor using Chrome, every visitor arriving through an in-app browser, and every Edge user on a corporate desktop is invisible in your field data. If your audience skews toward any of those, your Core Web Vitals assessment is measuring a subset of your traffic and your analytics are measuring a different one.
Eligibility also has a floor. A page or origin must be publicly discoverable and have enough visitors for a statistically significant dataset, and Chrome’s documentation says an exact number is not disclosed. A page returning a non-200 status, carrying a noindex header, or carrying a noindex meta tag fails the discoverability test.
Do this: Check your analytics for the share of sessions on iOS and Edge. That percentage is the blind spot in your field data.
The 75th percentile decides, not the average
This single rule explains most of the confusion between a good score and a failing assessment.
web.dev states that assessment uses the 75th percentile of page loads, segmented across mobile and desktop devices. In plain terms: sort every session by how slow it was, walk three quarters of the way up, and that session is your result.
An average hides your worst sessions. A 75th percentile is designed not to. It means your result is set by the slower quarter of real visits, which is exactly the group a lab test on one good connection never produces.
That is why median-looking sites fail. Your typical visitor has a fine experience, your slowest quarter does not, and the slowest quarter is what gets graded.
It also changes what counts as a fix. Shaving 200 milliseconds off an already-fast path moves nothing. Removing the thing that makes a slow connection catastrophic, a blocking third-party script or an unoptimized hero image, moves the percentile. Those two pieces of work look similar on a ticket and are not remotely equivalent.
The external script problem deserves naming, because it is the most common cause and the least owned. Tag managers, chat widgets, consent banners and analytics all execute on the client, and every one is a performance decision made by someone outside the engineering team. Treat them like any other third party integration with a budget and an owner.
Do this: List every third-party script on your highest-traffic page with the name of the person who asked for it. Review anything with no owner.
Google has not said Core Web Vitals are a ranking factor
The documentation is more careful than the industry quoting it, and the difference is worth reading closely.
Google Search Central’s Core Web Vitals page, as of 10 December 2025, says it highly recommends that site owners achieve good Core Web Vitals for success with Search, and that this, along with other page experience aspects, aligns with what core ranking systems seek to reward.
Notice what is absent. No weight. No claim of a direct ranking signal. No promised position change. The page also does not state which percentile Search uses, which is why the 75th percentile figure above is cited to web.dev rather than to Search Central.
So the honest version is this. Core Web Vitals describe real user experience, real user experience affects whether people stay and convert, and Google says good vitals align with what its ranking systems reward. That is a sound reason to fix them. It is not a reason to expect a rank jump from a score change.
HTTP Archive’s 2024 Web Almanac, published 11 November 2024 and written by Jevgenija Zigisova and Ines Akrap, gives the scale of the real problem. Drawing on CrUX field data, it found 43% of mobile pages passing all Core Web Vitals against 54% on desktop, with good mobile LCP at 59%, good mobile INP at 74% and good mobile CLS at 79%.
Fewer than half of mobile pages pass, in a web where a Lighthouse score above 90 is unremarkable. That gap is the entire subject of this article. We publish what our builds measured afterward rather than the scores they reached on launch day.
Do this: Replace “we scored 94” in your reporting with the three field values and the pass or fail on each.
Work the fix in field-first order
Sequence matters, because the cheapest work and the highest-value work are rarely the same ticket.
| Order | Action | Why here |
| 1 | Read the field panel and name the one failing metric | Everything downstream depends on which of three it is |
| 2 | Check whether field data exists at all | No data means too little traffic to be assessed |
| 3 | Fix the failing metric only | The other two are already passing at the 75th percentile |
| 4 | Audit third-party scripts for the metric you are fixing | Usually the largest single contributor you control |
| 5 | Re-measure after 28 days of real sessions | Field data aggregates, so same-day checks tell you nothing |
| 6 | Only then look at the Lighthouse score | Useful for debugging, never as the target |
If LCP is failing, the work is almost always the hero image and the critical path. If CLS is failing, it is ads, embeds, late-loading fonts or images with no declared dimensions. If INP is failing, it is JavaScript on the main thread during interaction, and Lighthouse will not have told you, because it never clicked.
A platform choice can cap how far this goes. Hosted builders decide what JavaScript ships on your behalf, which is a real constraint rather than an excuse, and worth understanding before you pick one. The Bubble and Webflow comparison and the wider no-code against custom build decision both come down partly to how much of the output you control.
Do this: Pick the one failing field metric and close every other performance ticket until it passes.
What a faster site will not fix
Speed removes a reason to leave. It does not create a reason to stay.
If your conversion rate did not move after a real field improvement, the bottleneck was never speed. It was the offer, the pricing clarity, the form length or the fact that visitors arrived with the wrong intent. Those are not performance problems and no amount of millisecond work touches them.
Some things are genuinely outside your control. You cannot make Chrome on iOS report into CrUX. You cannot get field data for a page nobody visits. You cannot stop Lighthouse scores varying between runs, which Google’s own documentation attributes mainly to changing underlying conditions rather than to the tool.
I also cannot tell you how much revenue a passing assessment is worth to you. Nobody can from outside your analytics, and any agency quoting a conversion lift from a Core Web Vitals fix is quoting a number it cannot produce.
Where we fit: if your field data names a failing metric and you want the architecture fixed rather than the score gamed, that is a contained piece of work and we will tell you up front if the cause is a platform limit we cannot engineer around. We are the wrong call if your field data is already passing on all three, because then the problem is elsewhere and a performance engagement would waste your money. Send us the field panel screenshot and we will say which of those two you are. Our own build approach starts from the failing metric rather than from a target score.
Do this: Read the field panel before commissioning any performance work. Not the score.
Frequently Asked Questions
Why did my traffic not improve after fixing my PageSpeed score?
Because the score is lab data from one simulated load, and Google assesses Core Web Vitals from real Chrome sessions at the 75th percentile. Half the Lighthouse score comes from metrics that are not Core Web Vitals, and INP contributes nothing to it. A better score can leave all three field metrics unchanged.
What is a good Lighthouse performance score?
Google’s PageSpeed Insights documentation puts 90 and above as good, 50 to 89 as needing improvement, and below 50 as poor. Treat it as a debugging aid rather than a goal. The number that determines whether you pass Core Web Vitals is the field assessment, which uses different metrics and a different population.
How long does field data take to update after a fix?
Field data aggregates real sessions over time, so a same-day recheck shows nothing. Plan to re-measure roughly 28 days after your change has been live, and compare the same page rather than the site total. If the page has low traffic, it may never accumulate enough sessions to be assessed at all.
Did Core Web Vitals thresholds change in 2026?
Not according to Google’s own documentation. Search Central’s Core Web Vitals page, last updated 10 December 2025, still lists LCP at 2.5 seconds, INP at 200 milliseconds and CLS at 0.1. Articles claiming new 2026 thresholds appear on low-authority sites with no corroborating Google source. Always check the vendor documentation’s last-updated date.
Are Core Web Vitals a Google ranking factor?
Google has not said so in those words. Its documentation says it highly recommends achieving good Core Web Vitals for success with Search, and that this aligns with what its core ranking systems seek to reward. That is a weaker and more careful claim than the industry usually reports, and worth reading in the original.
Resources
- Web Vitals, Google web.dev, page last updated 31 October 2024. The LCP 2.5 second, INP 200 millisecond and CLS 0.1 thresholds, the 75th percentile assessment rule, the FID retirement example, and the statement that lab measurement is not a substitute for field measurement.
- Core Web Vitals and Google Search results, Google Search Central, page last updated 10 December 2025. The current three-metric set and the carefully worded position that good Core Web Vitals align with what core ranking systems seek to reward.
- About PageSpeed Insights, Google. The lab against field distinction, that field data is powered by the Chrome User Experience Report, that Lighthouse analyzes in a simulated environment on a mid-tier Moto G4, and the 90, 50 to 89 and below 50 score bands.
- Lighthouse performance scoring, Chrome for Developers. The exact score weightings of Total Blocking Time 30%, LCP 25%, CLS 25%, First Contentful Paint 10% and Speed Index 10%, and the guidance to treat performance as a distribution rather than one number.
- CrUX methodology, Chrome for Developers. The origin and page eligibility criteria, the four user conditions, and the exclusion of Chrome on iOS, Android WebView apps and other Chromium browsers including Microsoft Edge.
- Performance, The 2024 Web Almanac, HTTP Archive, published 11 November 2024 by Jevgenija Zigisova and Ines Akrap. CrUX-derived pass rates of 43% of mobile pages and 54% of desktop pages passing all Core Web Vitals, with good mobile LCP at 59%, INP at 74% and CLS at 79%.
M. Sabir Junaid
Founder • CEO
M. Sabir Junaid is founder and CEO of Velcod, the senior product studio. He leads product and delivery across MVPs, SaaS platforms, and mobile apps — system architecture, no-code and low-code engineering, and AI integration. An official Bubble partner, Sabir has shipped products across fintech, marketplaces, edtech, and creator platforms. He works with founders raising pre-seed to Series A who need a product built to launch and scale in weeks, not a six-month agency build.
Share
- Previous
- Next
Ready to build?
Your idea, live in 3 weeks.
Book a free strategy call and get a fixed-price roadmap for your MVP, SaaS, or app.