Original research
We Ran PageSpeed on 1,080 Insurance Agency Websites
Everyone claims agency websites are slow. We measured exactly how slow, with Google's own tool, across the whole industry sample — and published the data.
10 min read
On August 18, 2026 we ran Google's PageSpeed Insights API against 1,080 insurance agency homepages on the mobile strategy, then filtered to the 1,046 substantive agency sites (parked pages and list contamination removed)1. The median performance score is 61. Only 12 sites — 1.15% — reach 99+, and when we fingerprinted those twelve, ten are static or website-builder static output; just two run WordPress, one of them only behind an aggressive optimization layer. One in five sites scores below 50. The bottleneck is paint speed, not layout: the median site takes 7.9 seconds to render its largest element against Google's 2.5-second "good" threshold2, while 88% already pass the layout-shift bar. And a quieter finding: more than 90% of agency sites have so little traffic that Chrome has no real-user field data for them at all.
What we did
This is the companion to our tracking-pixel study, run against the same working list of U.S. insurance agency domains. For each of 1,350 domains we requested a mobile PageSpeed Insights analysis of the homepage through Google's public API; 1,080 returned a Lighthouse performance score (the rest were unreachable, blocked automated analysis, or errored). Every number below comes from Google's own measurement pipeline, and the aggregate dataset is published CC01.
The distribution
| Mobile performance score | Sites | Share of 1,046 |
|---|---|---|
| 99–100 (the tier we build to) | 12 | 1.15% |
| 90–98 | 49 | 4.7% |
| 50–89 | 763 | 72.9% |
| Below 50 | 222 | 21.2% |
Median 61; the middle half of the industry sits between 52 and 70. And the 99+ tier has a telling composition: we fingerprinted all twelve sites from their served HTML, and ten are static output — hand-built pages or website-builder static export. Only two run WordPress, and one of those gets there via NitroPack, an optimization layer that effectively staticizes the page. Reaching 99+ on a dynamic stack without such a layer is, in this sample, a one-in-a-thousand event. For context on what a score "needs" to be, our answer page on what PageSpeed score an agency site actually needs separates the ranking question from the human one — but wherever you draw the line, 61 is on the wrong side of it.
LCP is the problem, not CLS
The industry has largely solved layout stability: 88.2% of scanned sites pass Google's CLS threshold and nearly half measure a perfect 0. What it has not solved is paint speed. The median lab Largest Contentful Paint is 7.9 seconds, more than three times Google's 2.5-second "good" bar2, and only 4.6% of sites come in under it. In practice that means the typical agency homepage shows a mostly blank or half-built screen for most of eight seconds on a phone — the exact window in which a just-clicked prospect decides whether to wait.
The field-data finding nobody talks about
Only 102 of 1,046 sites — under 10% — have Chrome UX Report field data at all, because CrUX requires a minimum volume of real Chrome visitors before Google will publish real-user metrics3. Of those 105, just 25 rate FAST overall. Two implications: first, for 90%+ of agencies, the "field vs lab" debate is moot — lab data is the only data that exists for your site (our Lighthouse-vs-PSI answer page explains the difference). Second, the sites that DO have field data are disproportionately the higher-traffic ones — and even among them, only 24 of 102 rate FAST overall.
What 99+ actually takes
The 1.15% is not a lottery. Our own client builds measure 99–100 on mobile from clean builds — repeatedly, across different agencies — and the pattern behind those scores is architectural, not cosmetic: static rendering, no render-blocking third-party JavaScript, images sized and prioritized deliberately, and every "best practice" measured before it ships (we've reverted a font preload that made LCP a second worse). The full breakdown lives in can a real business website score 99+ on mobile. A slow stack with a caching plugin does not get there; that's why 98.2% of the industry doesn't.
Methodology and limitations
Sample: the same 1,350-domain agency working list as the tracking-pixel study (an alphabetical batch spanning roughly a–o); 1,080 homepages returned scores. Substance filter: headline figures use the 1,046 sites with at least 100 words of homepage content after removing three non-agency domains that had leaked into the list — because a parked or near-empty page scores 100 trivially and would flatter the top bucket (the raw, unfiltered numbers are in the published CSV for transparency). Collection: Google PageSpeed Insights API v5, mobile strategy, August 18, 2026, one run per site. Stack fingerprinting: served-HTML signatures (generator tags, wp-content paths, framework markers) on the twelve 99+ sites. Limitations: homepage only; a single Lighthouse run per site (lab scores vary run to run — medians across a thousand sites smooth this, individual rows would not); lab data reflects a simulated mid-tier device and connection; 270 domains couldn't be scored and are excluded rather than assumed slow. Per-domain results are unpublished by design; the aggregate CSV is CC01.
Questions about this study
How were the scores collected?
Through Google's own PageSpeed Insights API, version 5, mobile strategy, run against each site's homepage on August 18, 2026. The performance score, lab LCP, CLS, and TBT come from the Lighthouse result; the field metrics come from the Chrome UX Report data the API returns when a site has enough real-user traffic to qualify.
Is a score of 61 actually bad?
It means the page is slow for real visitors on ordinary phones — the median site in this sample takes 7.9 seconds to paint its largest element on a mobile connection, against Google's 2.5-second threshold for 'good.' Whether that costs rankings is debated; that it costs waiting humans is not.
Why do only 102 of 1,046 sites have field data?
Chrome's UX Report only accumulates field data for sites with sufficient real-user traffic. More than 90% of agency sites in this sample don't clear that bar — which is itself a finding: most agency websites have too few visitors for Google to even measure their real-world experience.
Is the 99+ club really only 1.15%?
Yes, and the raw number was even more misleading before we cleaned it: 19 of 1,080 scored 99+, but seven of those were parked or near-empty pages (a blank page scores 100 trivially) or non-agency domains. Among the 1,046 substantive agency sites, 12 reach 99+ — and ten of the twelve are static builds. It is achievable, but by architecture chosen for speed, not a plugin added to a slow stack.
Where is the underlying data?
The aggregate dataset is published at /research/agency-core-web-vitals-2026-08.csv under CC0, free to cite. As with our tracking-pixel study, per-domain results are deliberately unpublished; researchers can contact us for verification.
- Strategic AI Architects. "Insurance Agency Website Core Web Vitals Scan, August 2026," aggregate dataset, CC0. strategicaiarchitects.com/research/agency-core-web-vitals-2026-08.csv. Collected via the Google PageSpeed Insights API v5.
- Google / web.dev. "Largest Contentful Paint (LCP)" — the 2.5-second "good" threshold, assessed at the 75th percentile. web.dev.
- Google. "Chrome UX Report" — CrUX publishes field data only for origins meeting a minimum popularity/traffic threshold. developer.chrome.com.
Find out your own score — and what 99+ would take
Run the free Audit against your domain, or see how we build to 99–100 on mobile.
Related reading: the tracking-pixel study · what is Largest Contentful Paint