Google Lighthouse is Google’s free, open-source auditing tool. It is built into Chrome DevTools and also powers the lab section of PageSpeed Insights. It loads one page under controlled conditions and scores it for Performance, Accessibility, Best Practices and SEO. The common mistake is to treat those scores as Google’s verdict on your rankings. They aren’t. The SEO score is a short checklist of indexing basics, and the performance score is a single lab run on an emulated phone. What Google’s ranking systems use is Core Web Vitals data from real Chrome users. This guide explains what each audit checks, how the scores are calculated, and which failures are worth fixing.
What Google Lighthouse Is, and What It Is Not
Lighthouse is a lab tool. It opens a URL in Chrome, records the load, and runs audits against the result. Besides the four scored categories, version 13.3 (May 2026) added an Agentic Browsing category to the default configuration. It checks how usable a page is for AI agents, covering the accessibility tree, WebMCP and llms.txt. It reports a count of passed checks rather than a 0–100 score, and Lighthouse describes it as “still under development and subject to change.”
Lighthouse 13, released in October 2025, was the last big structural change. According to the Chrome team’s release post, many older performance audits were merged into “insights” such as LCP breakdown and render-blocking requests. Performance scoring did not change. The only change to the SEO category was the removal of the font-size audit, because Google found “no signals” that it remains an SEO concern. If an agency report still lists “legible font sizes” as an SEO failure, it was produced with an older version.
It is not a crawler, a rank checker or a sample of real visitors: it tests one URL on one emulated device.
How to Run a Lighthouse SEO Audit in DevTools
The quickest way to run a Lighthouse SEO audit: open the page in an Incognito window so extensions stay out of the test, open DevTools, select the Lighthouse panel, choose Mobile or Desktop and the categories you need, then click Analyze page load. Google’s docs say a report takes about 30 to 60 seconds.
Lighthouse has three modes, and choosing the wrong one causes a lot of confusion:
- Navigation audits a single page load. It is the only mode that produces a performance score.
- Timespan records a period while you click and type. It has no overall score, but it is the only mode that measures Interaction to Next Paint (INP).
- Snapshot audits the page in its current state, for example with a menu open. It has no performance metrics.
The same engine runs elsewhere too. PageSpeed Insights runs it on Google’s servers, and the CLI lets you script it, for example npx lighthouse https://example.com --only-categories=seo --view. DevTools is the only option that can test local builds and pages behind a login. For the browser’s other diagnostic tools, see our Chrome SEO audit guide.
How the Lighthouse Performance Score Is Calculated
The Lighthouse performance score is a weighted average of five lab metrics. The insights below it don’t affect the score. These are the current default weights:
| Lab metric | Weight | Field metric Google uses |
|---|---|---|
| Total Blocking Time (TBT) | 30% | INP (TBT is only a rough proxy) |
| Largest Contentful Paint (LCP) | 25% | LCP |
| Cumulative Layout Shift (CLS) | 25% | CLS (field CLS also counts shifts after load) |
| First Contentful Paint (FCP) | 10% | Not a Core Web Vital |
| Speed Index | 10% | None |
Each value is mapped to 0–100 on a log-normal curve built from HTTP Archive data. The 25th percentile of real sites scores 50 and the 8th percentile scores 90. The bands are 0–49 red, 50–89 orange and 90–100 green. Because the curve flattens near the top, going from 95 to 100 takes far bigger gains than going from 50 to 70.
Two consequences matter for SEO. The biggest weight goes to TBT, which is not a Core Web Vital. And INP, the responsiveness metric Google does use, counts for nothing in a navigation run, because a page load involves no user interactions. A page can score in the 90s and still fail INP for real visitors.
Why Your Score Changes on Every Run
Google’s scoring documentation lists A/B tests, rotating ads, network routing, test hardware, extensions and antivirus software as causes. Two more are specific to Lighthouse:
- Simulated throttling. Lighthouse loads the page at full speed, then models a slower phone: about 150 ms of latency, roughly 1.6 Mbps download and a 4x CPU slowdown. The slowdown is relative to your machine, so a fast laptop and an old office PC produce different “mobile” results.
- Where the test runs. DevTools uses your computer and network. PageSpeed Insights uses Google’s data centres, and its Lighthouse version can differ from the one in your Chrome.
The Lighthouse team’s variability guide says the median of five runs is about twice as stable as one run. Never report a single run, and never compare a DevTools score with a PSI score.
Lab vs Field: Which Numbers Google Ranks On
Google’s page experience documentation says Core Web Vitals “are used by our ranking systems,” but also that Google “always seeks to show the most relevant content, even if the page experience is sub-par.” Search Console Help says its Core Web Vitals report comes from the Chrome UX Report (CrUX), built from “actual users visiting your URL.” PageSpeed Insights shows the same data over 28 days at the 75th percentile, against these thresholds: LCP 2.5 s or less, INP 200 ms or less, and CLS 0.1 or less.
When the two sources disagree:
- Field passes, lab is orange: real users are fine. Don’t spend a sprint on a lab score.
- Field fails, lab is green: the lab isn’t reproducing your users’ conditions, such as slower phones or interactions after load. Investigate with timespan mode.
- No field data: the page has too little Chrome traffic for CrUX. Track the lab median over time instead.
See our field vs lab data guide for the details and the Core Web Vitals guide for fixes to each metric.
The SEO Category, Audit by Audit
The SEO category has ten scored audits and one manual reminder:
| Audit | Fails when | Doesn’t check |
|---|---|---|
| Not blocked from indexing | A noindex in the robots meta tag or X-Robots-Tag header, or a robots.txt rule, blocks all crawlers |
Whether Google actually indexed the page |
| Successful HTTP status | The page returns a 4xx or 5xx status | Redirect chains, soft 404s |
| robots.txt is valid | Syntax errors, or the file returns a 5xx or 429 status | Whether your rules block the right URLs |
| Links are crawlable | Anchors have no usable href (missing, empty or a script placeholder) |
Orphan pages, click depth |
| Document has a title | The <title> is missing or empty |
Length, relevance, duplicates |
| Has a meta description | Missing or empty | Quality |
| Descriptive link text | Generic anchors such as “click here” or “read more” | Anchor relevance or distribution |
| Images have alt text | An <img> has no alt attribute |
Whether the alt text is useful |
| Valid hreflang | An unknown language code or a relative URL | Return links, consistency across pages |
| Valid rel=canonical | Conflicting, invalid or relative canonicals, or a deep page canonicalised to the homepage | Whether the target is indexable |
| Structured data (manual) | Never fails; it only reminds you | Everything. Use a validator |
Every check covers only the page being tested. If your canonical points to a URL that returns 404, the audit still passes because the tag itself is well formed. For cross-page checks, use the free Canonical Tag Checker, Hreflang Checker and Robots.txt Tester.
The Weighting Most Guides Get Wrong
Many guides say every SEO audit carries equal weight. Lighthouse’s source code says otherwise. In the default configuration, the indexing audit has a weight of 93/23 (about 4), and each of the other nine has a weight of 1. A code comment says the weight was chosen so that this one failure makes the whole category fail. With the current weights:
- A
noindexpage that passes everything else scores 69. It shows orange, not red, so a staging tag left on a live template can look minor on a dashboard when it is the worst failure possible. - A page whose only problem is a missing meta description scores 92. It stays green.
Audits that don’t apply are also removed from the calculation. A page with no canonical tag at all is marked “not applicable”, not failed, so it can still score 100, while a page with a slightly malformed canonical loses points. A 100 doesn’t mean your canonicals are in place. It means nothing Lighthouse looked at was broken. See how the canonical tag actually works.
What a Perfect SEO Score Can’t Tell You
Lighthouse describes its SEO category as “basic search engine optimization advice” and says “there are many additional factors Lighthouse does not score here.” A 100 tells you nothing about:
- The rest of the site: duplicate titles, orphan pages and noindexed sections only show up in a crawl.
- Content and intent: whether the page answers the query better than the pages ranking above it.
- What Googlebot sees: Lighthouse renders the page in your own Chrome. Use Search Console’s URL Inspection to see Google’s version.
- Structured data: see our structured data testing guide or use the free Schema Markup Validator.
The Accessibility and Best Practices categories are worth a quick look, though. Missing alt text, unnamed links, pages not served over HTTPS and console errors are all real defects, and a console error is often the first sign that a script responsible for rendering content is failing. Our accessibility and SEO guide explains where the two genuinely overlap.
Whole-site issues are why SEO Rocket’s technical audit uses a real crawler (backed by Screaming Frog, with JavaScript rendering) instead of scoring one URL. It finds 404s, redirect chains, broken links and soft-404 candidates on pages a Lighthouse run never loads. Our technical SEO audit guide covers the full process.
A Triage Workflow: From Report to Fix List
- Fix indexing blockers first. If the indexing, HTTP status or robots.txt audits fail, nothing else matters yet.
- Check field data before working on performance. If CrUX shows LCP, INP and CLS passing, note the lab score and move on.
- If a field metric fails, open the matching insight. For LCP, use the LCP breakdown and LCP request discovery insights. For INP, use timespan mode or the DevTools Performance panel, which shows local LCP, CLS and INP live as you interact and can compare them with CrUX. For CLS, check the layout shift culprits.
- Fix the template, not the URL. A heavy hero image on the product template slows every product page, so one fix improves all of them.
- Re-test with a median, then give the 28-day CrUX window time to reflect the change before reporting success.
In SEO Rocket, steps 2 and 3 happen in chat. Ask it to check a page’s speed and it runs PageSpeed Insights, then returns the lab score next to real-user LCP, INP and CLS. It tells you when a page has no field data and lists the opportunities with the largest time savings. For a quick standalone check, try the free Website Speed Test.
FAQ
Is the Lighthouse score a Google ranking factor?
Not directly. Google says its ranking systems use Core Web Vitals, and the Core Web Vitals data it reports to site owners comes from CrUX field measurements. A Lighthouse score is a lab estimate from one test, useful for diagnosing those metrics rather than standing in for them.
Why does PageSpeed Insights show a different score from DevTools?
Same engine, different hardware, network and sometimes Lighthouse version. Compare results from one tool only.
What is a good Lighthouse SEO score?
Every applicable audit should pass, which is easy because the checks are basic. A score below 70 often means the indexing audit failed, so check that first.
The Short Version
Google Lighthouse is an excellent diagnostic tool and a poor scoreboard. Use the SEO category for indexing basics, remembering that a noindex page still scores 69. Use the performance insights to explain a failing field metric, not to chase 100. Judge success by real-user Core Web Vitals and whether the page ranks.
