The State of Web Vitals

State of Largest Contentful Paint

Most sites pass LCP. When they fail, look at the server and late discovery before you touch the image format.

82.5% of sites pass LCP
1.5s Typical site
3.1s Worst 10% of sites

82.5% of sites pass LCP. The typical site sits at 1.5s. The worst 10% are above 3.1s.

Field data PhoneDesktopAll Q2 2026 edition · All devices field outcomes

Field LCP from Chrome, joined to a technique crawl of the same origins. Every rate and ranking below is computed from this edition. Nothing here is a controlled experiment.

Cohorts, not A/B tests. Sites that adopt a technique are not randomly assigned. Read the gaps as associations, then verify on your own traffic.

1
The state

LCP distribution and percentiles

How CrUX LCP is spread across sites, from the fastest to the slowest.

Most sites pass LCP. 82.5% sit at or under 2.5s. The typical site is at 1.5s, and 75% of sites stay under 2.2s. Only 17.5% fail LCP. The slowest 10% are above 3.1s, and the slowest 1% are above 5.6s.

Sites by CrUX LCP

0 6,635 13,269 19,904 26,538
0–250ms35 sites<0.1% of all sites
250–500ms2,066 sites1.2% of all sites
500–750ms10,748 sites6.2% of all sites
750–1000ms22,205 sites12.7% of all sites
1000–1250ms26,538 sites15.2% of all sites
1250–1500ms24,291 sites13.9% of all sites
1500–1750ms20,632 sites11.8% of all sites
1750–2000ms15,790 sites9% of all sites
2000–2250ms12,498 sites7.2% of all sites
2250–2500ms9,230 sites5.3% of all sites
2500–2750ms6,964 sites4% of all sites
2750–3000ms5,317 sites3% of all sites
3000–3250ms4,030 sites2.3% of all sites
3250–3500ms3,162 sites1.8% of all sites
3500–3750ms2,382 sites1.4% of all sites
3750–4000ms1,798 sites1% of all sites
4000ms and up7,003 sites4% of all sites
p50 = 1.5s
p75 = 2.2s
p90 = 3.1s
p99 = 5.6s
0–250 250–500 500–750 750–1000 1000–1250 1250–1500 1500–1750 1750–2000 2000–2250 2250–2500 2500–2750 2750–3000 3000–3250 3250–3500 3500–3750 3750–4000 4000+
Good (≤2.5s) Needs improvement Poor (>4s) Percentile markers Total: 174,689 sites
2
The break

Explore the sub-phases of LCP

LCP is four sub-phases: TTFB, load delay, load time, and render delay.

LCP breaks into four sub-phases. On the typical site they add up to 1.3s, still under the 2.5s good bar.

TTFB is the longest at 647ms. That is almost half the path (48.6%). Load delay is second at 318ms (discovery of the hero). Render delay is 224ms. Download (load time) is the short one at 142ms.

TTFB is also the most interesting. It tracks CrUX LCP hardest (corr +0.72). When the server stretches, pass rate falls. The 50% cliff sits near 1.8s of TTFB. Load delay is next (+0.46). That is late discovery, not a heavy image file. Across sites, the image download is rarely the whole problem.

LCP sub-phases on the typical site

TTFB 647ms
Load delay 318ms
Load time 142ms
Render delay 224ms
2.5s good
TTFBLoad delayLoad timeRender delay

Sub-phase times and share

Lab medians on the LCP path. Corr. vs CrUX LCP. Open a phase for the pass-rate cliff.

Phase Median Share P90 Corr. vs LCP Cliff
TTFB (server) 647ms 48.6% 2.0s +0.72
Load delay (discovery) 318ms 23.9% 928ms +0.46
Load time (download) 142ms 10.7% 358ms +0.26
Render delay 224ms 16.8% 569ms +0.28

CrUX LCP pass rate vs TTFB

50% @ 1.8s

CrUX LCP pass rate vs TTFB

0% 25% 50% 75% 100% 96.0% 95.8% 93.5% 70.5% 1.3% 50% @ 1.8s 100ms 200ms 300ms 500ms 800ms 1s 1.5s 2s 3s 4.5s Field TTFB (all devices, log scale, ms) Field LCP pass rate
Observed bins Logistic fit 50% at 1.8s

LCP rating by TTFB band

94%
94%
100–121 ms
n 347
92%
92%
121–146 ms
n 493
95%
95%
146–177 ms
n 1,195
96%
96%
177–214 ms
n 2,764
97%
97%
214–259 ms
n 5,322
96%
96%
259–313 ms
n 9,614
96%
96%
313–379 ms
n 14,337
96%
96%
379–458 ms
n 17,390
95%
95%
458–555 ms
n 16,915
95%
95%
555–671 ms
n 16,073
94%
94%
671–811 ms
n 14,843
92%
92%
811–982 ms
n 13,656
90%
90%
982ms–1.2s
n 12,538
84% 15%
84%
1.2–1.4 s
n 10,920
71% 27%
71%
1.4–1.7 s
n 9,229
43% 53%
43%
1.7–2.1 s
n 7,668
80%
11%
2.1–2.5 s
n 5,690
77% 20%
3%
2.5–3.1 s
n 3,816
46% 53%
1%
3.1–3.7 s
n 2,206
11% 88%
1%
3.7–4.5 s
n 1,325
Good LCP Needs improvement Poor Field TTFB (log-spaced bands, ms)
3
The mechanism

How the browser finds the hero

Late discovery shows up as load delay. Background images and script-injected heroes are the common paths.

Discovery path

Variant Sites Share Pass LCP p75
Link header 370 0.2% 95% 999ms
HTML 122,164 69.2% 85% 1.5s
Preload 4,020 2.3% 81% 1.6s
CSS 48,623 27.5% 78% 1.7s
JavaScript 1,353 0.8% 70% 2.0s

LCP element type

Variant Sites Share Pass LCP p75
Image 74,505 42.2% 85% 1.5s
Text 46,425 26.3% 83% 1.5s
Video 4,268 2.4% 81% 1.6s
Background image 51,352 29.1% 78% 1.7s

Discovery path: Link header at 95%, JavaScript at 70% (25 points). LCP element type: Image at 85%, Background image at 78% (7 points).

4
The stack

CMS: pass rate and phase anatomy

A high pass rate can hide a different budget. Some platforms pay the server. Others pay discovery after a fast first byte.

CMS field LCP and phase medians

CMS Sites Pass p75 TTFB Load delay Load time Render delay Total
Shopify 7,616 96% 1.7s 390ms 592ms 208ms 298ms 1.5s
Wix 5,212 95% 1.6s 425ms 452ms 135ms 151ms 1.2s
Drupal 3,046 92% 1.7s 568ms 294ms 130ms 211ms 1.2s
Joomla 2,329 90% 2.0s 728ms 313ms 112ms 221ms 1.4s
Squarespace 2,029 89% 2.1s 422ms 438ms 188ms 285ms 1.3s
Adobe Commerce 988 78% 2.4s 1.0s 234ms 84ms 208ms 1.6s
WordPress 58,969 70% 2.7s 1.1s 341ms 166ms 235ms 1.9s

Same stacks as bars

Fleet TTFB Load delay Render delay 1.3s
Wix5,212 TTFB Load delay 1.2s
Drupal3,046 TTFB Load delay 1.2s
Squarespace2,029 TTFB Load delay Render delay 1.3s
Joomla2,329 TTFB Load delay Render delay 1.4s
Shopify7,616 TTFB Load delay Render delay 1.5s
Adobe Commerce/Magento988 TTFB Load delay 1.6s
WordPress58,969 TTFB Load delay Render delay 1.9s
TTFB (server)Load delay (discovery)Load time (download)Render delay

Shopify leads at 96% good LCP. WordPress sits at 70%. Server share of the phase budget runs from 26% on Shopify to 66% on Adobe Commerce/Magento. Same metric, different anatomy.

5
What moves LCP (and what does not)

Which techniques actually separate pass from fail

Every row is the same fleet, split by a technique signal. Gap is best minus worst variant with at least 300 sites. Five points or more is a separator this edition. Under five points, the signal is real in labs and still weak as a fleet-level LCP story.

This edition the separators are Discovery path, Image format, Element type, Responsive markup and Priority hint on the LCP. Ranked by gap: Discovery path (+25 pp), Image format (+11 pp), Element type (+7 pp), Responsive markup (+7 pp) and Priority hint on the LCP (+5 pp).

Under the 5-point line: 1P vs 3P LCP origin and loading attribute on the LCP. Fix them when you see them on a real page. Do not expect them to rewrite the fleet LCP story.

Scoreboard: best vs worst cohort on field LCP

Only variants with at least 300 sites. Gap under 5 points is marked weak for this edition.

Signal Best cohort Worst cohort Gap Role
Discovery path Link header → 95% JavaScript → 70% +25 pp Separator
Image format AVIF → 91% PNG → 79% +11 pp Separator
Element type Image → 85% Background image → 78% +7 pp Separator
Responsive markup picture element → 89% No responsive markup → 82% +7 pp Separator
Priority hint on the LCP preconnect only → 87% Preload → 81% +5 pp Separator
1P vs 3P LCP origin Third party → 85% First party → 82% +3 pp Weak
loading attribute on the LCP eager → 87% auto (default) → 85% +2 pp Weak

Separators this edition

#1

Discovery path +25 pp

Link header passes on 95% of sites (p75 999ms). JavaScript sits at 70% (p75 2.0s). Gap: 25 points.

Discovery path — every solid cohort

Variant Sites Share of signal Pass LCP p75 LCP
Link header 370 0.2% 95% 999ms
HTML (parser) 122,164 69.2% 85% 1.5s
Preload 4,020 2.3% 81% 1.6s
CSS 48,623 27.5% 78% 1.7s
JavaScript 1,353 0.8% 70% 2.0s
#2

Image format +11 pp

AVIF passes on 91% of sites (p75 1.4s). PNG sits at 79% (p75 1.6s). Gap: 11 points.

Image format — every solid cohort

Variant Sites Share of signal Pass LCP p75 LCP
AVIF 6,818 6.1% 91% 1.4s
GIF 651 0.6% 86% 1.4s
WebP 29,710 26.6% 85% 1.5s
JPEG 55,415 49.6% 81% 1.6s
SVG 2,498 2.2% 80% 1.6s
PNG 16,502 14.8% 79% 1.6s
#3

Element type +7 pp

Image passes on 85% of sites (p75 1.5s). Background image sits at 78% (p75 1.7s). Gap: 7 points.

Element type — every solid cohort

Variant Sites Share of signal Pass LCP p75 LCP
Image 74,505 42.2% 85% 1.5s
Text 46,425 26.3% 83% 1.5s
Video 4,268 2.4% 81% 1.6s
Background image 51,352 29.1% 78% 1.7s
#4

Responsive markup +7 pp

picture element passes on 89% of sites (p75 1.4s). No responsive markup sits at 82% (p75 1.5s). Gap: 7 points.

Responsive markup — every solid cohort

Variant Sites Share of signal Pass LCP p75 LCP
picture element 12,249 6.9% 89% 1.4s
srcset only 24,190 13.7% 85% 1.5s
No responsive markup 140,113 79.4% 82% 1.5s
#5

Priority hint on the LCP +5 pp

preconnect only passes on 87% of sites (p75 1.4s). Preload sits at 81% (p75 1.6s). Gap: 5 points.

Priority hint on the LCP — every solid cohort

Variant Sites Share of signal Pass LCP p75 LCP
preconnect only 6,529 3.7% 87% 1.4s
fetchpriority=high 14,632 8.3% 84% 1.5s
No priority hint 152,211 86.2% 82% 1.5s
Preload 3,180 1.8% 81% 1.6s

Looks important, barely separates pass from fail

Weak this edition (under 5 points)

Still worth fixing on a single site. As a fleet LCP separator, they do not clear the bar.

Signal Best Pass Worst Pass Gap
1P vs 3P LCP origin Third party 85% First party 82% +3 pp
loading attribute on the LCP eager 87% auto (default) 85% +2 pp
6
Takeaway

One number, then go slice the data

LCP distribution → LCP phases → Discovery path →
The State of Web Vitals · Q2 2026 · 189,915 sites · all devices field datacorewebvitals.io/state-of-cwv
LCP in the explorer → Full data explorer State of Web Vitals home