Lighthouse scores: what they mean and what they do not
What the four Lighthouse scores measure, why they vary between runs, how they relate to real visitor experience, and how to use them without chasing them.
The short answer
Lighthouse is a tool built into the Chrome browser that loads a page once under simulated conditions and produces four scores out of a hundred: performance, accessibility, best practices and search basics. It is a lab test. The performance score is an estimate from a single load on a simulated mid-range phone and slow connection, and it varies between runs and machines by several points for no reason that matters. It is not what search engines use for ranking; they use measurements from real visitors. A page can score well in the lab and feel slow to visitors, or the reverse, depending on devices, connections and what visitors actually do on it. The accessibility score covers only what automation can detect, which is a fraction of the standard, so a hundred means no automated failures were found, not that the site is accessible. Lighthouse is valuable for what it is good at: pointing at specific problems on a specific page, and tracking a page against itself over time. For knowing how the site performs, use real-user data.
The four scores
| Score | What it measures | What it misses | How to use it |
|---|---|---|---|
| Performance | Lab estimates of loading, interactivity and stability metrics on a simulated device | Real devices and connections; interaction after load; pages not tested | Diagnose specific issues from the audit list; compare a page over time |
| Accessibility | Automated checks: contrast, labels, names, roles, some structure | Keyboard behaviour, screen reader experience, meaning, focus order, most of the standard | Fix everything it flags, then do the manual tests |
| Best practices | HTTPS, console errors, deprecated APIs, image aspect ratios, some security basics | Most security; code quality; architecture | Clear the list; it is mostly hygiene |
| Search basics | Metadata present, crawlable, mobile-friendly, valid structured data | Content quality, rankings, backlinks, intent | Treat as a minimum, not a search strategy |
Using Lighthouse well
- Run it several times on the same page and machine and take the median.
- Read the audit list, not the score; the list names the specific images, scripts and resources causing the estimate.
- Test the pages that matter: home, key service pages, an article, a form, not only the home page.
- Fix what it flags in accessibility and best practices; they are almost always genuine.
- Then do the manual accessibility tests, keyboard and screen reader, which the score cannot do.
- Track over time in the pipeline, with budgets that fail a build when a page regresses.
- Judge the outcome by field data in the search console and real-user monitoring.
What a good score does and does not tell a business
A performance score in the nineties on key pages, consistently, says the pages are built without obvious waste. It does not say visitors find the site fast; the field data says that. An accessibility score of a hundred says the automated checks pass. It does not say a blind visitor can use the site; the manual tests say that. Best practices and search basics at a hundred say the hygiene is done. None of the four says anything about whether the content is right or whether the site does its job.
What this means for you
Lighthouse is a diagnostic tool, not a verdict: a lab estimate that varies, is not used for ranking, and covers only part of accessibility. Use its audit lists to find and fix specific problems, track pages against themselves in the pipeline, and do the manual accessibility tests it cannot. Judge how the site actually performs by real-user data, which is what visitors and search engines experience.
Frequently asked questions
Our developer says the site scores 98. Is it fast?
It is probably fast in the lab, on the conditions Lighthouse simulated, on that run. Whether it is fast for your visitors depends on their devices, connections and what they do on the page, which is what real-user measurements report. A high lab score with poor field data usually points to something the single lab load does not exercise: heavy interaction, third-party scripts that load late, or pages other than the one tested.
Why does the score change every time we run it?
Because it is a single measured load on whatever machine ran it, with simulated throttling, and small differences in timing, caching, background activity and network produce different numbers. Variation of several points between runs is normal. Compare medians of several runs, on the same machine, against the same page, and treat single scores as indicative rather than exact.
Does Google use the Lighthouse score for ranking?
No. Search ranking uses field data from real visitors, the Core Web Vitals, assessed over a period, not lab scores. Lighthouse reports lab estimates of the same metrics and is useful for diagnosis, but a page's ranking signal comes from how it performed for actual users. That is why field data is the number to watch and Lighthouse the tool to fix things with.
Sources
- Chrome for Developers: Lighthouse performance scoring (accessed 2026-09-12)