Someone on our team ran Lighthouse against the homepage last quarter, saw a 98 performance score, and closed the ticket. Two weeks later a support email came in: the pricing page timed out for a prospect, three blog posts were 404ing from an old redirect that never got cleaned up, and none of it showed up anywhere because nobody had looked past the page Lighthouse happened to score.
That's not a Lighthouse bug. It did exactly what it's built to do: audit one page, thoroughly, right now. The gap was ours, for treating a single-page snapshot like it covered the site.
What Lighthouse is actually built for
Lighthouse is Google's open-source, single-page auditing tool. Per its own documentation, it runs a series of audits against one URL and generates a report on performance, accessibility, best practices, and SEO. You can run it from Chrome DevTools, the command line, or as a Node module, and it's genuinely good at what it does. Its performance scoring carries real weight, a composite of metrics like Largest Contentful Paint and Total Blocking Time, with a breakdown per metric detailed enough to actually act on. The lab environment is consistent too, simulating a fixed network and device profile so you get a repeatable number to compare against, rather than whatever your Wi-Fi happened to be doing. Its accessibility and best-practices checks catch real issues: missing alt text, insufficient contrast, missing viewport tags. It won't catch everything (it can't judge whether an alt attribute is actually accurate), but what it does check, it checks reliably. And setup is zero: it's built into every Chrome install, open DevTools, click the Lighthouse tab, done.
If the question is "how fast does this one page load and why," Lighthouse is still the right first stop, and PageSpeed Insights layers in real-world Chrome UX Report field data on top of Lighthouse's lab results for that same page. Nothing in this post is trying to replace that combination for page-level performance work.
It's also worth being fair about automation: Lighthouse CI exists, and teams do wire it into a pipeline to run against a list of URLs on every deploy and fail a build on regression. That's a real, working pattern for tracking performance scores over time on pages you already know matter. It still audits the URLs you hand it though; it doesn't discover the rest of your site, follow your links, or notice the page nobody remembered to add to the list.
What a single page can't tell you
The limitation isn't quality, it's scope. Lighthouse audits the URL you give it and nothing else, which means an entire category of problems is structurally invisible to it:
Whole-site crawling. Lighthouse has no concept of your other pages. It doesn't know your sitemap exists, doesn't follow links, and can't tell you that the page you tested is one of forty with the same duplicate title tag.
Broken links. A 404 two clicks from your homepage doesn't show up in a single-page audit of the homepage. You'd need to test every page individually, or crawl the site with something that follows links and checks each destination.
Security headers across pages. Lighthouse's best-practices category flags a few security-adjacent issues on the page it's testing, like missing HTTPS or a known-vulnerable JS library. It doesn't check whether your Content-Security-Policy or Strict-Transport-Security headers are set consistently across the site, because that requires visiting more than one URL.
Content and SEO consistency. One page might have a clean title tag and solid structured data. That tells you nothing about whether your blog, your docs, and your pricing page all follow the same pattern, which is usually where the real SEO debt lives.
Agent experience. This one didn't exist as a category when Lighthouse was designed. Whether your site has a valid llms.txt, whether your structured data is something an LLM can actually parse, whether AI crawlers can reach your content at all: none of it is in scope for a tool built to score DOM rendering and network timing on a single page.
Lighthouse vs squirrelscan
| Lighthouse | squirrelscan | |
|---|---|---|
| Scope | One URL per run | Crawls the whole site |
| Performance | Deep lab metrics (LCP, TBT, CLS) | Performance signals as part of a broader audit |
| Broken links | Not checked | Checked across every crawled page |
| Security headers | A few best-practices flags on one page | Header checks across the site |
| SEO consistency | Single-page checks | Cross-page consistency (titles, duplicate content, structure) |
| Agent experience | Not covered | llms.txt, structured data, AI crawler access |
| Output | Score + report for one page | Site-wide report, JSON, or LLM-ready text |
| Setup | Built into Chrome, zero install | curl -fsSL https://install.squirrelscan.com | sh |
When you need which one
Reach for Lighthouse (or PageSpeed Insights, if you want field data alongside the lab score) when you're optimizing a specific page and need a detailed breakdown of why it's slow. That's still the right tool for the job, and squirrelscan doesn't try to out-diagnose Lighthouse's performance category.
Reach for a site-wide audit when the question is bigger than one page: is the whole site healthy, are there broken links piling up, are your security headers consistent, can an AI agent actually read your content. That's a crawl problem, not a single-page problem, and it's where squirrelscan sits. Run squirrel audit https://example.com and it checks every page it finds against 249+ rules spanning SEO, performance, security, and agent experience, then hands back one report instead of one score per URL you remember to test.
Poke at the sample report to see the shape of a full-site result, check your own Core Web Vitals alongside the rest of the audit, or point the website audit tool at your URL directly. If you're already working inside a coding agent, the Claude Code integration runs the same audit without leaving your editor.