Why Your Mobile PageSpeed Score Dropped (And How to Fix LCP in 2027)

Learn why mobile PageSpeed scores drop, how CrUX field data differs from Lighthouse lab tests, and how to fix the four LCP subparts on mobile.

As seen in

  • Fox News
  • ABC News
  • CBS
  • NBC
  • Associated Press
  • USA Today
  • Bloomberg
  • HuffPost
Otto the beaver next to a browser window icon
On this page5

Your mobile PageSpeed Insights score drops when Lighthouse lab metrics regress—in Lighthouse 10 the Performance score weights Total Blocking Time at 30%, Largest Contentful Paint at 25%, and Cumulative Layout Shift at 25%, with First Contentful Paint at 10% and Speed Index at 10% on a simulated Moto G4 mobile network—or when trailing 28-day CrUX field data shifts at the 75th percentile. To fix LCP in 2027 and reach 2.5 seconds or less at the 75th percentile, optimize all four LCP subparts: Time to First Byte, resource load delay (fetchpriority="high" and never loading="lazy"), resource load duration, and element render delay.

Field data vs. lab data: why your mobile PageSpeed score dropped

PageSpeed Insights reports two evaluations: real-user field data and simulated lab diagnostics. Per Google's PageSpeed Insights documentation, the top section shows Chrome User Experience Report (CrUX) data collected over a trailing 28-day period at the 75th percentile, falling back to origin-level data when a URL lacks sufficient traffic.

The bottom section is Lighthouse lab data: a controlled load on a simulated Moto G4 mobile network, while desktop uses a wired emulated desktop—so mobile lab scores usually sit below desktop. Lighthouse performance scoring counts metric values only (not opportunity warnings) toward the 0 to 100 score. Field thresholds match Web Vitals—for example LCP Good at 0 to 2500ms and Poor over 4000ms, INP Good at 0 to 200ms, CLS Good at 0 to 0.1, FCP Good at 0 to 1800ms, and experimental TTFB Good at 0 to 800ms per PageSpeed Insights. Lighthouse 10 weights are 30% TBT, 25% LCP, 25% CLS, 10% FCP, and 10% Speed Index.

MetricGood (CrUX / PSI field)Needs improvementPoorLighthouse 10 weight
LCP0 to 2500ms[(2500ms, 4000ms]](https://developers.google.com/speed/docs/insights/v5/about)over 4000ms25%
INP0 to 200ms[(200ms, 500ms]](https://developers.google.com/speed/docs/insights/v5/about)over 500msCrUX field only
CLS0 to 0.1[(0.1, 0.25]](https://developers.google.com/speed/docs/insights/v5/about)over 0.2525%
FCP0 to 1800ms[(1800ms, 3000ms]](https://developers.google.com/speed/docs/insights/v5/about)over 3000ms10%
TTFB (experimental)0 to 800ms[(800ms, 1800ms]](https://developers.google.com/speed/docs/insights/v5/about)over 1800msDiagnostic
Total Blocking Time (lab)———30%
Speed Index (lab)———10%

According to Google Search Central documentation, Google's search ranking systems look at real-user Core Web Vitals—LCP, INP, and CLS at the 75th percentile—rather than your Lighthouse lab score. A laboratory score of 95 does not replace real field performance. To monitor these signals without manual testing, you can connect PageSpeed Insights to Otto with no Google API key required.

Why mobile scores drop even when you didn't touch your website

Site owners often notice a sudden drop on mobile PageSpeed Insights without pushing code changes. This happens because field metrics and lab runs reflect external factors.

First, third-party marketing scripts change independently of your site code. Tag managers, Meta conversion pixels, live chat widgets, and review badges can update without a deploy on your side. Lighthouse performance scoring lists A/B tests, changes in ads served, internet traffic routing, device differences, browser extensions, and antivirus software as common reasons Performance scores fluctuate between runs. Heavy new JavaScript on the main thread inflates Total Blocking Time, which carries 30% of the Lighthouse 10 Performance score in that weighting table.

Second, your 28-day CrUX window shifts continuously. Real-user field data aggregates trailing performance across 28 days at the 75th percentile. Paid campaigns on Meta Ads or Google Ads can increase cold mobile visits that arrive on uncached device connections. Ad clicks often route through tracking redirects (fbclid, gclid), which add redirect latency before the browser receives HTML. Those cold visits may pull your trailing CrUX percentiles even when your homepage code did not change. Reviewing how Core Web Vitals drift impacts landing page conversion shows how paid traffic shifts alter measured Core Web Vitals.

Third, when a URL lacks enough CrUX samples, PageSpeed Insights falls back to origin-level 28-day percentiles that cover all pages on the site. Fourth, Lighthouse runs in Google datacenters with variable routing, which Lighthouse performance scoring lists among run-to-run score fluctuations. The Otto AI SEO Agent for weekly automated SEO reports highlights improvement areas and tracks keyword rankings over time.

The four subparts of LCP (and why compressing images alone often fails)

According to web.dev's specification guide for Largest Contentful Paint, LCP measures the render time of the largest visual element in the viewport relative to navigation start. Eligible elements include <img> elements, <image> elements inside SVG, video poster frames, elements with CSS background images loaded via url(), and block-level text nodes.

Many site owners attempt to fix a dropped score by converting JPG files to WebP or AVIF. When the mobile score does not improve, the culprit is usually element render delay. Web.dev's guide to optimizing Largest Contentful Paint explains that LCP consists of four non-overlapping subparts that total 100% of LCP time.

LCP subpartGuideline share of LCPWhat the browser doesTypical bottleneck
Time to First Byte (TTFB)~40%Waits for server response and receives initial HTMLServer response, ad redirects, cold cache
Resource load delay<10%Discovers the LCP asset URL in markupLazy-loading, CSS background rules, client-side JS
Resource load duration~40%Downloads the image, video, or font file bytesOversized assets, slow CDN paths
Element render delay<10%Paints the downloaded asset to the mobile screenRender-blocking CSS, synchronous JS, main-thread tasks

Say your mobile LCP is in the Poor range and the image file downloads quickly. Web.dev describes how shrinking an image into WebP or AVIF can reduce resource load duration yet fail to improve total LCP when JavaScript or CSS delays rendering and shifts saved time into element render delay. You must address all four subparts to pass Core Web Vitals.

Step-by-step: how to fix each LCP subpart on mobile in 2027

Follow these four steps to eliminate bottlenecks across every subpart of mobile LCP.

1. Eliminate resource load delay

The browser must discover your LCP element immediately within the initial server-rendered HTML.

Never add loading="lazy" to your hero image or primary above-the-fold visual. Web.dev explicitly warns that lazy-loading an LCP image creates unnecessary resource load delay, directly degrading LCP. Remove loading="lazy" from any image likely to appear in the mobile viewport.

Add fetchpriority="high" to your primary LCP image. This tells the browser scanner to download the asset ahead of lower-priority scripts and stylesheet images. Web.dev cautions against setting high priority on more than one or two images, as saturating network queues destroys priority benefits.


<img 
  src="/images/hero-mobile.webp" 
  srcset="/images/hero-mobile.webp 600w, /images/hero-desktop.webp 1200w"
  sizes="(max-width: 600px) 100vw, 1200px"
  alt="Mobile hero image for a pilates studio homepage" 
  width="600" 
  height="400" 
  fetchpriority="high"
  decoding="async">

If your hero image loads through a CSS background-image: url() property, the browser cannot discover the file until it downloads and parses the stylesheet. Preload the asset inside your document <head>:

<link rel="preload" fetchpriority="high" as="image" href="/images/hero-mobile.webp" type="image/webp">

When the LCP candidate is block-level text that depends on a custom web font, web.dev recommends preloading that font in <head> so text does not wait on a late font download:

<link rel="preload" as="font" type="font/woff2" crossorigin href="/fonts/your-font.woff2">

2. Eliminate element render delay

Element render delay happens when synchronous stylesheets or JavaScript block painting after the LCP asset downloads. Move tracking and non-critical scripts to deferred tags:


<script src="/js/analytics-setup.js" defer></script>

Extract above-the-fold mobile CSS into an inline <style> block and load the rest asynchronously:

<link rel="preload" href="/css/non-critical.css" as="style" onload="this.onload=null;this.rel='stylesheet'">
<noscript><link rel="stylesheet" href="/css/non-critical.css"></noscript>

3. Reduce resource load duration

Resource load duration measures the transfer time of the image or font file.

Match image display dimensions to mobile screen widths. Serving a full desktop-width image to a phone viewport wastes cellular bandwidth. Use responsive srcset and sizes attributes so mobile browsers request appropriately sized files. Using automatic image compression and scheduled maintenance passes in Otto v1.1.5 handles image re-encoding across your media library without manual exporting.

Host critical media assets on the same origin or an edge CDN proxy. Connecting to third-party storage buckets requires separate DNS lookups, TLS handshakes, and TCP negotiations, adding extra connection setup time on cellular networks. Ensure static assets return strong caching headers:

Cache-Control: public, max-age=31536000, immutable

4. Reduce Time to First Byte (TTFB)

TTFB sets the baseline for LCP. If your server takes 1.5 seconds to return the first byte of HTML, your page cannot achieve a 2.5-second LCP.

Enable edge page caching via your CDN or host so static pages and cached CMS outputs can return from edge points of presence when possible. On GitHub-based sites, fast Vercel deployment workflows for landing pages walk through how Otto links repositories and supports preview and deployment orchestration.

Audit paid marketing URLs. Ad campaigns often accumulate multiple tracking redirects before resolving to the destination page. Each redirect adds a cellular round trip. Update ad destination links to point directly to the canonical HTTPS URL with no intermediate HTTP hops.

How to audit PageSpeed Insights and ship the fix with Otto

Diagnosing a bottleneck is straightforward; staging HTML and theme edits on a live CMS is not. Otto Webmaster is an AI digital webmaster built by IslaIntel in San Juan, Puerto Rico, with 227 tools across 15 integrations and works on your existing platform without forcing a migration.

Connect Claude to Otto MCP (or ChatGPT or Cursor) with an organization-scoped bearer token so OAuth credentials stay on Otto's servers. The built-in psi_analyze_url tool queries PageSpeed Insights with Otto's platform key and requires no individual Google Cloud API setup. PageSpeed Insights, Google Search Console, Google Analytics 4, and Google Ads are read-only in Otto MCP.

To audit your dropped score, paste this instruction into Claude connected to Otto MCP:

Run psi_analyze_url on https://example.com with the mobile strategy. 
Break down the Largest Contentful Paint into its four subparts: TTFB, 
resource load delay, resource load duration, and element render delay. 
Identify the exact LCP candidate element, check if loading="lazy" is 
improperly set, and list the top render-blocking CSS and JS files.

Once Claude identifies the issue, instruct Otto to stage the remediation via WordPress (29 tools ) or GitHub and Vercel deployment orchestration:

Based on the PSI audit for https://example.com, stage the speed fixes. 
For our GitHub repository, open a pull request that adds fetchpriority="high" 
to the mobile hero image, removes lazy loading from that tag, and inlines 
critical hero CSS. If working on WordPress, draft the page changes so we 
can review them before publishing.

GitHub fixes ship as pull requests with preview URLs; WordPress edits stay in draft. Production changes require approval, including one-tap WhatsApp approvals from June 2, 2026; checkout, contact, and legal pages always need explicit review.

Otto does not guarantee search rankings, Map Pack placement, or generative AI citations. Paid ad and social writes require Master; core webmaster and SEO fixes run on Pro. The Otto AI Website Manager for approval-gated WordPress drafts and GitHub pull request previews starts with a $1 trial for 3 days that includes a homepage redesign on your brand assets. Pro is $199/mo (or $99/mo billed annually); Master is $3,500/mo (or $3,000/mo billed annually).

Start $1 trial to run the audit and ship approval-gated fixes.

Written by

Otto Team

Editorial & SEO Team at Otto

Practical website, SEO, paid ads, and analytics playbooks for small businesses from the Otto team.