Running Lighthouse across an entire website

Lighthouse audits one URL per run and ships no crawler, so auditing a 40-page site means 40 separate runs and no aggregate view. There are four ways around that: script the CLI in a loop, wire up Lighthouse CI, script the PageSpeed Insights API, or use a desktop tool that crawls and runs Lighthouse per page. Audra is the last of those — it drives your installed Chrome across a whole crawl, then merges every page's scores with SEO, accessibility, link and AI-visibility checks into one report.

Last updated 2026-07-28

Can Lighthouse crawl an entire website?

No. Lighthouse audits a single URL per run and ships no crawler — this is confirmed in Google's own documentation, and it is a design decision rather than an oversight. Lighthouse is a page-level diagnostic.

The consequence is that there is no built-in way to answer "which of my 60 pages is slowest?" without running it 60 times and pasting the results into a spreadsheet. Everything that makes Lighthouse work site-wide is orchestration you either build or buy: discovering the URL list, queueing the runs, storing the JSON, and aggregating it into something readable.

The ecosystem of wrappers around it — Lighthouse CI, Unlighthouse and others — is itself the evidence. Audra uses Unlighthouse internally for exactly this reason: it is the fan-out layer Lighthouse deliberately does not include.

What are the real options for running Lighthouse site-wide?

There are four, and they suit genuinely different people: the CLI in a loop for one-off scripting, Lighthouse CI for build pipelines, the PageSpeed Insights API for cloud automation, and a desktop crawler for producing a report someone will read. Three of the four are free.

What does Audra add on top of Lighthouse?

Three things: it finds the pages, it runs four more check layers on each one, and it reduces the result to a single number with a document attached.

Discovery is sitemap parsing and link crawling running together, capped at 100 pages by default and adjustable to 1,000, with include and exclude path patterns and robots.txt respected while crawling. Each page is then audited at a 412×823 mobile viewport and a 1366×900 desktop viewport, driving the Chrome already installed on your computer — Audra never bundles or downloads a browser, which is also why the installer is around 165 MB rather than the several hundred a bundled Chromium would add.

On top of the Lighthouse categories, each page gets 14 deterministic HTML SEO checks with real thresholds — title flagged outside 50–60 characters, meta description outside 120–158, exactly one H1, no skipped heading levels, an absolute canonical, valid hreflang, complete Open Graph, and thin content under 200 words — plus the full axe-core 4.12 accessibility ruleset weighted by impact, internal link checking that flags broken links as errors and redirect chains over one hop as warnings, and 8 AEO checks covering whether AI answer engines can retrieve and cite the page at all.

The composite Audra Score weights Performance 30%, SEO 30%, Accessibility 25% and Best Practices 15%. It is Audra's own number, not a Lighthouse score — the underlying Lighthouse category scores are shown unmodified alongside it, so nothing is hidden. If you want to understand what those underlying numbers mean before trusting a composite of them, read what counts as a good Lighthouse score.

Why can PageSpeed Insights not audit my staging site?

Because PageSpeed Insights runs on Google's infrastructure rather than your machine — Google's documentation states it runs in a Google datacenter that varies with network conditions. It therefore needs a publicly reachable URL, and cannot see localhost, a staging environment behind authentication, an intranet site, or anything behind a VPN.

This is the clearest structural advantage of running Lighthouse locally, whether through the CLI, DevTools or a desktop tool. Audra drives Chrome on your own machine, so a staging site audits exactly like production and no client URL is ever transmitted to a third-party service. For agency work under NDA, that removes a conversation entirely.

The Chrome DevTools Lighthouse panel has the same advantage and costs nothing. If you only need to check one authenticated page, use DevTools rather than buying anything — that is the honest recommendation.

What does Lighthouse still do better?

Lighthouse is free, it is the reference implementation every other tool reports numbers from, and PageSpeed Insights shows you something no purely local tool can: real-user field data.

• Field data. PageSpeed Insights layers CrUX real-user measurements over the lab result. Audra reports lab data only, which is a simulation of one visit under one set of conditions. When the two disagree, believe the field data — see lab versus field data in the Core Web Vitals guide.
• It costs nothing, in every form: Chrome DevTools, the CLI, the Node module and PageSpeed Insights.
• For a single landing page it is the fastest possible answer. Opening DevTools beats installing an application.
• Lighthouse CI enforces performance budgets on every commit and fails the build when a threshold regresses. That is a different job from auditing, and Audra does not attempt it.
• It is the canonical source. Every third-party tool, Audra included, ultimately reports numbers that came from Lighthouse.

Why do Lighthouse scores change between runs?

Because the performance score is measured rather than calculated — network variance, CPU contention on the machine running the test, and third-party scripts responding at different speeds all move it. A single run is a sample, not a measurement.

Two further causes catch people out: A/B tests or personalisation serving different content to different runs, and Chrome or Lighthouse version changes that alter metric weights or add new audits. A score can drop after a browser update with no change to your site at all.

The practical consequence is that a trend across many pages, or a before-and-after on the same machine, is far more meaningful than any one number. This is one of the better arguments for auditing a whole site rather than a single page: 40 pages measured once tells you much more about a template-level problem than one page measured 40 times.