Half the mobile web is failing a test its owners have never seen. As of the May 2026 Chrome UX data, 49.1% of the mobile web passes all three Core Web Vitals - which means 50.9% does not (CrUX data, reported by PageSpeed Matters, 2026). Your site is statistically more likely to be in the failing half than the passing one. Keep that number in mind while I tell you how we found out what "unmeasured" really costs.
On September 6th this year, we audited our own site and found the homepage promising "20+ tools" when 46 were live. For weeks. Not because anyone was lazy - because nobody owned the number, and nothing was checking it. We fixed the count in an afternoon; the lesson took longer to sink in: what nobody measures, rots silently. Speed is the same rot with a clock on it. Google has been publishing your score for years - field data, real users, your own visitors - and most owners have never once looked at it. This post is the ninety-minute look, and everything in it runs on free tools.
The three numbers Google actually uses
Core Web Vitals are three metrics, each with a "good" threshold, measured at the 75th percentile of real page views (Google's documentation, still current):
- LCP - Largest Contentful Paint: main content loaded within 2.5 seconds
- INP - Interaction to Next Paint: response to a tap or click within 200 milliseconds
- CLS - Cumulative Layout Shift: visual stability score under 0.1
The "75th percentile" part is the detail that changes how you read every score: Google grades your site on its typical visitor, not its worst case. One in four visits is allowed to be slower. That is a mercy for real-world networks - and it means a lab-perfect score on your office wifi proves nothing. Only field data counts.
And the pass rates, because context is everything: the Web Almanac's 2024 performance chapter measured 68% of desktop and 51% of mobile sites passing all three checks (HTTP Archive, 2024). The 2026 CrUX figures above read similar for mobile and better for desktop - the exact values shift with methodology, and we show you both rather than pick the flattering one. Either way the picture holds: a failing site is the norm, not the exception. Being better than average here is cheap points nobody around you is collecting.
The 90-minute reset, in order
1. Read the field data first (15 min)
Open your Core Web Vitals report before touching anything - Google Search Console shows it per URL group, based on real user measurements. Lab tools lie in useful ways: they test one page, one device, one connection, once. Your visitors arrive on a three-year-old Android over a weak LTE signal, and the field data knows it. Write down which of the three vitals fail, and on which templates. Estimate your vitals here if you want a second read without a login.
2. The image pass (20 min)
Images are the usual LCP killer - the hero image nobody compressed is the single most common reason a page loads its main content late. Compress everything (modern formats, visibly lossless quality), set explicit width and height attributes so layout does not jump while images load, lazy-load everything below the fold, and lazy-load nothing above it - the LCP element must load eager. Page size analyzer shows you which pages carry dead weight; you will usually find one or two images doing most of the damage.
3. The interactivity pass (20 min)
INP replaced First Input Delay in March 2024, and it is harsher: it measures the entire interaction - every tap to every paint - instead of grading the first ping alone. The usual culprits are third-party scripts - chat widgets, trackers, embeds - hogging the main thread while your visitor taps. Ship less JavaScript, break long tasks up, and audit what each third-party script actually earns. Time your pages before and after removing each one; the before/after makes the cost visible in seconds.
4. The layout stability pass (10 min)
CLS under 0.1 means: reserve space for everything that arrives late - images, ads, embeds, web fonts. Set dimensions, use font-display swap with a fallback that matches, and never inject content above existing content after load. The reader who taps the wrong button because your layout jumped is not recording a metric. They are recording an exit.
5. The server pass (15 min)
Compression, caching headers, and redirects. Every redirect chain is a round trip your visitor pays for before your HTML even arrives - trace your chains and cut them to zero on landing pages. Check your headers: compression on, sane cache lifetimes, no surprises. This is the fastest win of the six, because it is pure configuration.
6. Re-measure and write it down (10 min)
Re-run the field check, record the numbers, and put a recurring reminder on it. Our "20+ vs 46" incident lasted weeks because nothing re-checked the number. Your vitals report is the same story: it is not a one-time fix, it is a gauge. Gauges get read on a schedule, or they get ignored until something breaks.
Here I'll be blunt
My take: INP is the new kingmaker of this trio, and the one most teams have not heard of. Because: images and layout have a decade of tooling and habit behind them, but INP got its final definition only in 2024, it fails on scripts teams do not consider "their code" - third-party widgets, tag managers, embeds - and it is the vital most commonly failed in current CrUX breakdowns. A site can pass LCP and CLS for years and still bleed INP points on every tap of its mobile menu. What would change my mind: two consecutive CrUX releases showing INP failure rates dropping below the other two vitals - that would mean the third-party script problem solved itself. I do not expect to be changed.
The Devil's Advocate
The steelman: "Core Web Vitals are a tiebreaker, not a ranking engine - content wins, slow sites with great content outrank fast sites with thin content. You are selling anxiety." That is mostly right, and it deserves a straight answer. Yes: relevance beats speed in one-on-one comparisons. But the tiebreaker framing quietly assumes a one-on-one. Real searches serve ten results of similar relevance, and the field data is one of the few signals left that a competitor cannot copy overnight. Nobody optimizes vitals to beat great content. You optimize them so great content stops losing to equal content that loads faster.
The 2am Version
Google grades real visitors, at the 75th percentile, on three numbers: main content in 2.5 seconds, responses under 200 milliseconds, no layout jumps. Half the mobile web fails. Compress your images, cut third-party scripts, reserve layout space, kill redirect chains, then re-measure on a schedule. Ninety minutes, zero dollars, and you are instantly ahead of the half that never looked.
Analyses - the RankForge frame
Data lens. The pass-rate numbers across sources disagree at the edges - the Almanac's 2024 crawl-based methodology and CrUX's field-based origin data are different instruments measuring the same web, which is exactly why this post shows both. What survives the disagreement: mobile trails desktop everywhere, by double digits. And note what CrUX cannot tell you: per-page failures on low-traffic URLs. Your worst template may be invisible in field data simply because too few users see it to fill the report.
Risk lens. The failure mode of acting on this post wrong: score-chasing that breaks function. Teams disable the analytics that measure conversions, lazy-load above-fold images to flatter LCP, delete the chat widget that was closing sales - the score improves while the business gets worse. Every change in the reset should answer one question first: does this remove cost, or remove capability? Only the first kind is a win.
Ethics lens. Speed advice has a dark retail pattern: manufacture panic ("Google penalizes you TODAY"), then sell the audit. We will not do that. This post says plainly: vitals are a tiebreaker, half the web fails, and panic is not a strategy. Every tool linked here is free with no signup - the diagnosis is not the upsell, because we do not sell audits.
RF stance. Measure field data quarterly, fix what serves readers, ignore what serves only the score. Because field data is the only thing Google grades, reader-serving fixes compound with everything else you do, and score-theater fixes decay the moment the next script gets added.
Change-my-mind. A rigorous dataset showing Core Web Vitals failing to break ties between comparable pages - equal relevance, equal authority, different vitals, no ranking difference, sustained across crawls. Show me that, and the ninety minutes becomes optional. The six-in-ten-real-user data we have now says otherwise.
Sources
- Google Search Central - "Understanding Core Web Vitals and Google search results": the three metrics, thresholds, and field-data assessment at the 75th percentile. developers.google.com
- web.dev (Google) - Core Web Vitals explainer: how LCP, INP and CLS are defined; INP replaced FID in March 2024. web.dev
- HTTP Archive - The Web Almanac 2024, Performance chapter: 68% of desktop and 51% of mobile sites passed all Core Web Vitals. almanac.httparchive.org
- PageSpeed Matters (June 2026) - CrUX-based tracker: 49.1% of mobile and 58.0% of desktop origins passed all three vitals as of the May 2026 Chrome UX Report. (Aggregator of Google's public CrUX dataset - primary data at CrUX.) pagespeedmatters.com



