The State of Web Vitals

A/B testing impact

Performance impact by ab_testing provider.

Field data PhoneDesktopAll Scope All sites Q2 2026 edition · Phone field outcomes
Metric LCP INP CLS
1

At a glance the headline numbers for A/B testing impact

Performance impact by ab_testing provider.

8
Providers ranked
After min-sites filter
21,139
Sites in sample
Combined across all items
125ms
Best INP (p75)
Google Optimize
The State of Web Vitals · Q2 2026 · 189,915 sites · phone field datacorewebvitals.io/state-of-cwv
2

The ranking sorted by INP at p75, fastest first

# Provider INP p75 Passing Sites
1 Google Optimize 125ms 93.5%
19,960
2 AB Tasty 194ms 75.6%
109
3 Convert 196ms 77.1%
192
4 VWO 204ms 73.5%
433
5 Optimizely 239ms 61.0%
137
6 Kameleoon 241ms 62.0%
112
7 LaunchDarkly 254ms 66.7%
131
8 Dynamic Yield 320ms 39.6%
65
The State of Web Vitals · Q2 2026 · 189,915 sites · phone field datacorewebvitals.io/state-of-cwv
3

Passing INP per provider which group passes the INP most often

ProviderSitesPassing INPp75
Google Optimize 10.5% 19,960 94% 125ms
Convert 0.1% 192 77% 196ms
AB Tasty 0.1% 109 76% 194ms
VWO 0.2% 433 74% 204ms
LaunchDarkly 0.1% 131 67% 254ms
Kameleoon 0.1% 112 62% 241ms
Optimizely 0.1% 137 61% 239ms
Statsig 0% 2 50% 402ms
Split.io 0% 9 50% 295ms
Monetate 0% 26 45% 230ms
Dynamic Yield 0% 65 40% 320ms
Evolv ai 0% 1 494ms
Good Needs Improvement Poor Sorted best-passing first · median colored by its own rating · pass = good INP (200ms at p75) · one value per site

94% of Google Optimize sites pass INP. Optimizely trails 33 points behind, leaving 39% of its sites failing. computed

The State of Web Vitals · Q2 2026 · 189,915 sites · phone field datacorewebvitals.io/state-of-cwv
4

All five vitals at once the whole category without toggling - cell is the pass rate, small number the p75

LCP
INP
CLS
FCP
TTFB
VWO 0.2%
762.4s
74204ms
830.05
692.0s
531.3s
AB Tasty 0.1%
782.3s
76194ms
720.12
731.8s
56933ms
Statsig 0%
1002.2s
50402ms
1000.02
502.0s
100540ms
Evolv ai 0%
02.8s
0494ms
00.26
02.4s
01.7s
Google Optimize 10.5%
812.3s
94125ms
870.03
731.9s
541.3s
Convert 0.1%
662.7s
77196ms
800.07
692.1s
511.5s
Kameleoon 0.1%
772.5s
62241ms
770.07
701.9s
531.1s
Dynamic Yield 0%
652.7s
40320ms
640.22
712.0s
631.1s
LaunchDarkly 0.1%
454.1s
67254ms
710.13
453.0s
541.2s
Split.io 0%
573.6s
50295ms
430.22
432.9s
571.5s
Optimizely 0.1%
682.9s
61239ms
780.09
612.2s
441.2s
Monetate 0%
782.4s
45230ms
780.07
701.9s
391.1s
60%95%+ passing Cell: pass rate, small number = p75 · faded rows: under 100 sites

One row per provider, one column per vital - the cell is the share of sites passing, the small number the p75. No toggling needed to see where the category actually differs.

Google Optimize leads on INP: 94% of its sites pass. Optimizely trails at 61%. computed

The State of Web Vitals · Q2 2026 · 189,915 sites · phone field datacorewebvitals.io/state-of-cwv
Related signals Maps (4) → CDN (16) → Chat & messaging (17) → Bot detection (8) → Chrome field data from 189,915 sites, representing millions of real page loads · How we measured
Live queries (3) — admin only
Query #1: 36.6 ms 
SELECT 1 AS ok FROM site_providers LIMIT 0;
Query #2: 460.2 ms 
SELECT CAST(p.key AS VARCHAR) AS e, COUNT(*) AS sites,
  COUNT(*) FILTER (WHERE s.crux."all".lcp IS NOT NULL) AS all_lcp_n,
  quantile_disc(s.crux."all".lcp, 0.5) AS all_lcp_median,
  quantile_disc(s.crux."all".lcp, 0.75) AS all_lcp_p75,
  COUNT(*) FILTER (WHERE s.crux."all".lcp <= 2500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".lcp IS NOT NULL), 0) AS all_lcp_good,
  COUNT(*) FILTER (WHERE s.crux."all".lcp > 2500 AND s.crux."all".lcp <= 4000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".lcp IS NOT NULL), 0) AS all_lcp_ni,
  COUNT(*) FILTER (WHERE s.crux."all".lcp > 4000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".lcp IS NOT NULL), 0) AS all_lcp_poor,
  COUNT(*) FILTER (WHERE s.crux."all".inp IS NOT NULL) AS all_inp_n,
  quantile_disc(s.crux."all".inp, 0.5) AS all_inp_median,
  quantile_disc(s.crux."all".inp, 0.75) AS all_inp_p75,
  COUNT(*) FILTER (WHERE s.crux."all".inp <= 200) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".inp IS NOT NULL), 0) AS all_inp_good,
  COUNT(*) FILTER (WHERE s.crux."all".inp > 200 AND s.crux."all".inp <= 500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".inp IS NOT NULL), 0) AS all_inp_ni,
  COUNT(*) FILTER (WHERE s.crux."all".inp > 500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".inp IS NOT NULL), 0) AS all_inp_poor,
  COUNT(*) FILTER (WHERE s.crux."all".cls IS NOT NULL) AS all_cls_n,
  quantile_disc(s.crux."all".cls, 0.5) AS all_cls_median,
  quantile_disc(s.crux."all".cls, 0.75) AS all_cls_p75,
  COUNT(*) FILTER (WHERE s.crux."all".cls <= 0.1) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".cls IS NOT NULL), 0) AS all_cls_good,
  COUNT(*) FILTER (WHERE s.crux."all".cls > 0.1 AND s.crux."all".cls <= 0.25) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".cls IS NOT NULL), 0) AS all_cls_ni,
  COUNT(*) FILTER (WHERE s.crux."all".cls > 0.25) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".cls IS NOT NULL), 0) AS all_cls_poor,
  COUNT(*) FILTER (WHERE s.crux."all".fcp IS NOT NULL) AS all_fcp_n,
  quantile_disc(s.crux."all".fcp, 0.5) AS all_fcp_median,
  quantile_disc(s.crux."all".fcp, 0.75) AS all_fcp_p75,
  COUNT(*) FILTER (WHERE s.crux."all".fcp <= 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".fcp IS NOT NULL), 0) AS all_fcp_good,
  COUNT(*) FILTER (WHERE s.crux."all".fcp > 1800 AND s.crux."all".fcp <= 3000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".fcp IS NOT NULL), 0) AS all_fcp_ni,
  COUNT(*) FILTER (WHERE s.crux."all".fcp > 3000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".fcp IS NOT NULL), 0) AS all_fcp_poor,
  COUNT(*) FILTER (WHERE s.crux."all".ttfb IS NOT NULL) AS all_ttfb_n,
  quantile_disc(s.crux."all".ttfb, 0.5) AS all_ttfb_median,
  quantile_disc(s.crux."all".ttfb, 0.75) AS all_ttfb_p75,
  COUNT(*) FILTER (WHERE s.crux."all".ttfb <= 800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".ttfb IS NOT NULL), 0) AS all_ttfb_good,
  COUNT(*) FILTER (WHERE s.crux."all".ttfb > 800 AND s.crux."all".ttfb <= 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".ttfb IS NOT NULL), 0) AS all_ttfb_ni,
  COUNT(*) FILTER (WHERE s.crux."all".ttfb > 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."all".ttfb IS NOT NULL), 0) AS all_ttfb_poor,
  COUNT(*) FILTER (WHERE s.crux."phone".lcp IS NOT NULL) AS phone_lcp_n,
  quantile_disc(s.crux."phone".lcp, 0.5) AS phone_lcp_median,
  quantile_disc(s.crux."phone".lcp, 0.75) AS phone_lcp_p75,
  COUNT(*) FILTER (WHERE s.crux."phone".lcp <= 2500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".lcp IS NOT NULL), 0) AS phone_lcp_good,
  COUNT(*) FILTER (WHERE s.crux."phone".lcp > 2500 AND s.crux."phone".lcp <= 4000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".lcp IS NOT NULL), 0) AS phone_lcp_ni,
  COUNT(*) FILTER (WHERE s.crux."phone".lcp > 4000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".lcp IS NOT NULL), 0) AS phone_lcp_poor,
  COUNT(*) FILTER (WHERE s.crux."phone".inp IS NOT NULL) AS phone_inp_n,
  quantile_disc(s.crux."phone".inp, 0.5) AS phone_inp_median,
  quantile_disc(s.crux."phone".inp, 0.75) AS phone_inp_p75,
  COUNT(*) FILTER (WHERE s.crux."phone".inp <= 200) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".inp IS NOT NULL), 0) AS phone_inp_good,
  COUNT(*) FILTER (WHERE s.crux."phone".inp > 200 AND s.crux."phone".inp <= 500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".inp IS NOT NULL), 0) AS phone_inp_ni,
  COUNT(*) FILTER (WHERE s.crux."phone".inp > 500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".inp IS NOT NULL), 0) AS phone_inp_poor,
  COUNT(*) FILTER (WHERE s.crux."phone".cls IS NOT NULL) AS phone_cls_n,
  quantile_disc(s.crux."phone".cls, 0.5) AS phone_cls_median,
  quantile_disc(s.crux."phone".cls, 0.75) AS phone_cls_p75,
  COUNT(*) FILTER (WHERE s.crux."phone".cls <= 0.1) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".cls IS NOT NULL), 0) AS phone_cls_good,
  COUNT(*) FILTER (WHERE s.crux."phone".cls > 0.1 AND s.crux."phone".cls <= 0.25) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".cls IS NOT NULL), 0) AS phone_cls_ni,
  COUNT(*) FILTER (WHERE s.crux."phone".cls > 0.25) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".cls IS NOT NULL), 0) AS phone_cls_poor,
  COUNT(*) FILTER (WHERE s.crux."phone".fcp IS NOT NULL) AS phone_fcp_n,
  quantile_disc(s.crux."phone".fcp, 0.5) AS phone_fcp_median,
  quantile_disc(s.crux."phone".fcp, 0.75) AS phone_fcp_p75,
  COUNT(*) FILTER (WHERE s.crux."phone".fcp <= 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".fcp IS NOT NULL), 0) AS phone_fcp_good,
  COUNT(*) FILTER (WHERE s.crux."phone".fcp > 1800 AND s.crux."phone".fcp <= 3000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".fcp IS NOT NULL), 0) AS phone_fcp_ni,
  COUNT(*) FILTER (WHERE s.crux."phone".fcp > 3000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".fcp IS NOT NULL), 0) AS phone_fcp_poor,
  COUNT(*) FILTER (WHERE s.crux."phone".ttfb IS NOT NULL) AS phone_ttfb_n,
  quantile_disc(s.crux."phone".ttfb, 0.5) AS phone_ttfb_median,
  quantile_disc(s.crux."phone".ttfb, 0.75) AS phone_ttfb_p75,
  COUNT(*) FILTER (WHERE s.crux."phone".ttfb <= 800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".ttfb IS NOT NULL), 0) AS phone_ttfb_good,
  COUNT(*) FILTER (WHERE s.crux."phone".ttfb > 800 AND s.crux."phone".ttfb <= 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".ttfb IS NOT NULL), 0) AS phone_ttfb_ni,
  COUNT(*) FILTER (WHERE s.crux."phone".ttfb > 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."phone".ttfb IS NOT NULL), 0) AS phone_ttfb_poor,
  COUNT(*) FILTER (WHERE s.crux."desktop".lcp IS NOT NULL) AS desktop_lcp_n,
  quantile_disc(s.crux."desktop".lcp, 0.5) AS desktop_lcp_median,
  quantile_disc(s.crux."desktop".lcp, 0.75) AS desktop_lcp_p75,
  COUNT(*) FILTER (WHERE s.crux."desktop".lcp <= 2500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".lcp IS NOT NULL), 0) AS desktop_lcp_good,
  COUNT(*) FILTER (WHERE s.crux."desktop".lcp > 2500 AND s.crux."desktop".lcp <= 4000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".lcp IS NOT NULL), 0) AS desktop_lcp_ni,
  COUNT(*) FILTER (WHERE s.crux."desktop".lcp > 4000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".lcp IS NOT NULL), 0) AS desktop_lcp_poor,
  COUNT(*) FILTER (WHERE s.crux."desktop".inp IS NOT NULL) AS desktop_inp_n,
  quantile_disc(s.crux."desktop".inp, 0.5) AS desktop_inp_median,
  quantile_disc(s.crux."desktop".inp, 0.75) AS desktop_inp_p75,
  COUNT(*) FILTER (WHERE s.crux."desktop".inp <= 200) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".inp IS NOT NULL), 0) AS desktop_inp_good,
  COUNT(*) FILTER (WHERE s.crux."desktop".inp > 200 AND s.crux."desktop".inp <= 500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".inp IS NOT NULL), 0) AS desktop_inp_ni,
  COUNT(*) FILTER (WHERE s.crux."desktop".inp > 500) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".inp IS NOT NULL), 0) AS desktop_inp_poor,
  COUNT(*) FILTER (WHERE s.crux."desktop".cls IS NOT NULL) AS desktop_cls_n,
  quantile_disc(s.crux."desktop".cls, 0.5) AS desktop_cls_median,
  quantile_disc(s.crux."desktop".cls, 0.75) AS desktop_cls_p75,
  COUNT(*) FILTER (WHERE s.crux."desktop".cls <= 0.1) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".cls IS NOT NULL), 0) AS desktop_cls_good,
  COUNT(*) FILTER (WHERE s.crux."desktop".cls > 0.1 AND s.crux."desktop".cls <= 0.25) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".cls IS NOT NULL), 0) AS desktop_cls_ni,
  COUNT(*) FILTER (WHERE s.crux."desktop".cls > 0.25) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".cls IS NOT NULL), 0) AS desktop_cls_poor,
  COUNT(*) FILTER (WHERE s.crux."desktop".fcp IS NOT NULL) AS desktop_fcp_n,
  quantile_disc(s.crux."desktop".fcp, 0.5) AS desktop_fcp_median,
  quantile_disc(s.crux."desktop".fcp, 0.75) AS desktop_fcp_p75,
  COUNT(*) FILTER (WHERE s.crux."desktop".fcp <= 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".fcp IS NOT NULL), 0) AS desktop_fcp_good,
  COUNT(*) FILTER (WHERE s.crux."desktop".fcp > 1800 AND s.crux."desktop".fcp <= 3000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".fcp IS NOT NULL), 0) AS desktop_fcp_ni,
  COUNT(*) FILTER (WHERE s.crux."desktop".fcp > 3000) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".fcp IS NOT NULL), 0) AS desktop_fcp_poor,
  COUNT(*) FILTER (WHERE s.crux."desktop".ttfb IS NOT NULL) AS desktop_ttfb_n,
  quantile_disc(s.crux."desktop".ttfb, 0.5) AS desktop_ttfb_median,
  quantile_disc(s.crux."desktop".ttfb, 0.75) AS desktop_ttfb_p75,
  COUNT(*) FILTER (WHERE s.crux."desktop".ttfb <= 800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".ttfb IS NOT NULL), 0) AS desktop_ttfb_good,
  COUNT(*) FILTER (WHERE s.crux."desktop".ttfb > 800 AND s.crux."desktop".ttfb <= 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".ttfb IS NOT NULL), 0) AS desktop_ttfb_ni,
  COUNT(*) FILTER (WHERE s.crux."desktop".ttfb > 1800) * 1.0 / NULLIF(COUNT(*) FILTER (WHERE s.crux."desktop".ttfb IS NOT NULL), 0) AS desktop_ttfb_poor
                    FROM site_providers p
                    JOIN sites s ON s.origin = p.origin
                    WHERE p.category = 'ab_testing'
                      AND p.key IS NOT NULL AND CAST(p.key AS VARCHAR) <> ''
                    GROUP BY p.key;
Query #3: 33.4 ms 
SELECT COUNT(*) AS n FROM sites;
JSON file lookups (5)
20260625/meta.json 0.03 ms
20260625/types.json 0.51 ms
20260625/loaf-scripts.json 0.39 ms
20260625/menu.json 0.12 ms
20260625/providers.json 5.37 ms