AI SEO TOOLS
SPEED REPORT 2026

The full leaderboard

Scores below come from Google's PageSpeed Insights API in mobile mode, using its Lighthouse lab engine. The archived summary reports medians from three requested runs per URL. A score of 90 or higher is a Lighthouse Performance tier. It is not a Core Web Vitals pass, which requires field LCP, INP, and CLS at the 75th percentile.

Click RERUN on any row to open PageSpeed Insights with that URL prefilled. A new result tests the current page and may differ from the May 5, 2026 archive.

Last run: May 5, 2026 - methodology - CC-BY-4.0

How was this speed report measured?

Every URL is tested with the same engine, same throttling, same device profile. The methodology stays constant across quarterly refreshes so historical comparisons remain valid.

The archived summary dataset includes tested URLs, median lab values, and collection timestamps. The May archive does not contain the individual run values, so it cannot prove run-to-run stability. The collector now preserves extracted values for each future run. Republish the summary under CC BY 4.0 with attribution.

What does a slow site actually cost?

Speed isn't a feature, it's a revenue lever. Google's own research and the Web Almanac's annual analysis converge on the same conclusion: every additional second on mobile loses customers.

• Bounce probability rises 32% when mobile load time goes from 1 to 3 seconds, and 90% from 1 to 5 seconds (Think with Google).
• Pages meeting Core Web Vitals thresholds show a 24% lower abandonment rate during loading (web.dev).
• Conversion rate drops by an average of 4.42% with each additional second of load time between 0 and 5 seconds (Portent industry benchmark study).

For a SaaS pricing page, that's the difference between a trial signup and a closed tab. For a publisher, it's the difference between a subscriber and a bounce. Half the tools in the table above are leaving money in the parking lot before the visitor even reads the headline.

What does Largest Contentful Paint measure?

LCP measures how long the largest above-the-fold element takes to render. Google's threshold for "good": 2.5 seconds or less. "Poor" starts at 4 seconds.

In this dataset 2 of 8 tools recorded a median lab LCP of 2.5 seconds or less. The three highest LCP medians range from 17.5 to 18.7 seconds. Those values show where further diagnosis is warranted, but the summary dataset alone does not identify the cause.

Use a fresh Lighthouse trace to inspect LCP subparts, then address the measured bottleneck. Common fixes include right-sizing the LCP image, serving an efficient format, prioritizing the LCP request, and reducing render-blocking work. Apply only the changes supported by the current trace.

Total Blocking Time is a lab diagnostic

TBT sums the blocking portion of long main-thread tasks between First Contentful Paint and Time to Interactive. It is useful for diagnosing JavaScript work in Lighthouse, but it is not a Core Web Vital and it is not field INP. The highest archived median in this dataset is 2,880ms.

Use the Lighthouse main-thread and long-task diagnostics to identify the responsible code before changing scripts. Third-party tags, application JavaScript, layout work, and parsing can all contribute, so the table alone cannot assign cause.

Why lead with mobile lab results?

Mobile emulation applies a consistent constrained profile that can expose work hidden by a fast desktop environment. Google primarily crawls with a smartphone user agent, but mobile-first indexing is not a separate ranking signal and a mobile Lighthouse score does not determine indexing.

This report ranks the archived mobile lab score and includes desktop for context. Search ranking and Core Web Vitals status require separate evidence, including relevance, indexability, and field LCP, INP, and CLS where CrUX coverage exists.

Why claude-seo.md ranks #1

The May 5 archive records a median mobile Lighthouse Performance score of 99 for claude-seo.md and 98 for MarketMuse. The other six stored medians are below 60. Because the legacy archive does not preserve individual run values, this page does not claim run-to-run stability. The implementation choices below are reproducible characteristics of this site, not proof that any one choice caused the score.

1. All CSS and JavaScript inlined. Zero external bundles. Zero render-blocking <link> tags beyond two web fonts loaded with display=optional. The browser has everything it needs to paint after the HTML arrives.
2. WebP for every image. Hero images use fetchpriority="high" and decoding="async". Below-fold images use loading="lazy".
3. Revalidating asset cache. Assets use a one-day browser cache plus a seven-day stale-while-revalidate window. Stable filenames can be updated without trapping visitors on a year-old response.
4. No analytics scripts. No chat widgets. No marketing automation. No A/B testing SDK. The page that's shown to a visitor is the page the developer wrote, not the page after the marketing team's fifteen pixels finish booting.
5. System fonts as the fallback path. If Google Fonts is slow, the browser renders Space Grotesk's system fallback in zero milliseconds. No FOIT, no FOUT, no waiting.
6. No JavaScript framework. Single-file static HTML. The entire site builds in zero seconds because there's nothing to build. Every byte is intentional.

The Claude SEO tool source is available under the MIT license. The website benchmark summary is available in the linked JSON dataset under CC BY 4.0.

Frequently asked questions