Urgent.News

What's breaking now, across thousands of outlets.

Tech

What I Learned Analyzing 91,904 Small Business Websites for Core Web Vitals and SEO

I spent six weeks pulling PageSpeed Insights, Lighthouse, and Search Console-style data from 91,904 small business websites across the US. Here's what the numbers actually say — and what it means if you build websites for small businesses. Full dataset writeup: Small Business Website Statistics 2026 The baseline is worse than you think 71% of SMB sites fail at least one Core Web Vitals threshold…

Over the course of six weeks, a reporter analyzed the website performance of 91,904 small businesses across the United States. The findings reveal that 71% of these websites fall short of meeting Core Web Vitals requirements on mobile devices. The biggest culprit in this performance is the Largest Contentful Paint (LCP), which is more detrimental than JavaScript or large image sizes. On a small business homepage, the median LCP time is 4.1 seconds, far exceeding Google's recommended threshold of 2.5 seconds.

WordPress websites with three or more page-builder plugins tend to perform poorly, with a median LCP of 5.8 seconds. Another issue is the Cumulative Layout Shift (CLS), which affects 43% of sites, not necessarily due to ads but due to lazily loaded hero images without reserved aspect ratios. A simple CSS rule could resolve this issue.

When it comes to the choice of website building platforms, the data indicates a clear hierarchy in terms of performance. Static site generators like Astro, 11ty, and Hugo perform best, with median LCP times of 1.4 seconds and 48 milliseconds. Next.js and Remix (both Server-Side Generated) also perform well, with median LCP times of 1.7 seconds and 62 milliseconds.

On the other hand, WordPress sites using page builders like Elementor or Divi see their LCP times spike to 5.1 seconds and 340 milliseconds, respectively. Wix and Squarespace also underperform with median LCP times of 3.7 seconds and 3.3 seconds.

The report suggests three simple strategies that can significantly improve website performance. First, serve hero images in WebP or AVIF formats with explicit width and height attributes. This can improve LCP by an average of 1.1 seconds and CLS by 0.09. Second, preload the LCP image, which can boost LCP by an average of 0.4 seconds on mobile devices. Lastly, remove unused analytics scripts from the website. Unused tags can cost 40-120 milliseconds in INP (Interaction to Next Paint).

Regarding SEO, only 28% of small business websites have an H1 tag that matches their primary ranking intent. A significant 61% have duplicate or missing title tags across their service pages, while 17% have no schema.org markup at all. The author of this report is sharing these insights through an open website cost calculator, which factors in the real performance ceiling of different website stacks.

They are also publishing more breakdowns of the dataset on oxymoreservices.com. The author is keen to hear feedback on the Elementor/Divi LCP performance, the diminishing returns of preloading the LCP image on mobile, and the potential usefulness of a public API for median LCP for local businesses. The raw CSV data can be accessed by emailing info@oxymoreservices.com.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

Wabi as a Health and Habits Companion: 5 Trackers You Can Build Inside a Chat

I run Wabi Wiki, an unofficial guide to the Wabi AI assistant — not affiliated with Wabi. Most people meet Wabi as a chat app that builds mini-apps.

  • Five health trackers can be built within Wabi AI's chat-based mini-apps
  • Weight tracker logs weigh-ins and displays weight trend over time
  • Habit tracker sends gentle nudges and tracks streaks for daily habits

The five-minute test that would have saved me eleven hours of directory submissions

Eleven hours of directory submissions produced two followed links. Not because the directories were fake, but because I never asked the one question that decides the outcome, and it takes five minutes…

  • Author discovers five-minute test to save eleven hours of directory submissions
  • Checks three details on listing page for linking potential
  • Single script used twice for both current and future submissions

63,336 agent skills, 56 paid: what the pricing data actually says

If you have been thinking about selling a SKILL.md , this is the number that should stop you for a minute: in a frozen census of 63,336 agent skills taken on 2026-10-16, exactly 56 skills carry a paid…

  • Only 1.1% of SKILL.md agent skills are paid, despite 4,902 skills with pricing labels
  • Paid skills account for 0.5% of installs, with 454,127 total installs across 56 skills

More from Thursday 19 November →