Lighthouse is Google's open-source tool for checking aspects of webpage quality, including performance, accessibility, best practices and SEO. To improve its SEO score, fix the actual failed checks in your report, then inspect the live page again. A score of 100 does not establish that the page is indexed, ranks well or contains accurate information.
It is useful for testing representative pages and detecting regressions. A single report describes the page and conditions tested; it does not certify every URL, every visitor's experience or the website's complete SEO strategy.
Read the categories separately: a real report example
This is a mobile Lighthouse lab report for Guangsuan's /seo/ page, run on September 27, 2026: Performance 75, Accessibility 100 and SEO 100; LCP 3.3 seconds, TBT 650 milliseconds and CLS 0.002. The SEO checks passed while performance still had room for improvement. These are historical measurements of our own site, not today's scores or a customer case study.

Total Blocking Time, shown in this screenshot, is a lab metric. It is not a measurement of real users' Interaction to Next Paint. A local loading test cannot establish that all visitors pass Core Web Vitals.
How to run a useful Lighthouse SEO check
- Open the URL you intend to test. Check the final address, page content, login state and consent state.
- Open Chrome DevTools and select the Lighthouse panel. Choose the applicable mode, device and audit categories.
- Run the audit and save the report, including its time, Lighthouse version, final URL, network and CPU settings.
- Read failed audits, diagnostics and manual checks. Open affected elements and confirm the finding on the actual page.
- Fix the responsible CMS field, template, content or configuration.
- Retest under comparable conditions and verify the published output. Preserve multiple measurements when investigating variable performance results.
The official overview also explains command-line and programmatic use. Features and audit sets change across versions and entry points, so do not rely on a permanent count of categories or SEO checks taken from an older tutorial.
Turn common failed checks into the right repair
| Finding | What to inspect and change | What a pass still does not prove |
|---|---|---|
| Missing title or description | Correct the page's CMS field or template; check the final HTML for empty or duplicate tags | That the wording is useful or will appear unchanged in search |
| Page blocked from indexing | Confirm the URL should be public; inspect robots meta and HTTP headers before removing an unintended restriction | That Google has crawled or indexed the page |
| HTTP status or robots.txt problem | Check the real response and rule; repair unintended access failures | That every removed URL should return 200 |
| Uncrawlable or unclear link | Use the correct href and descriptive wording for navigation | That the destination is relevant or the site has a complete internal-link structure |
| Missing image alt | Describe information images appropriately; use empty alt for genuinely decorative images | That the description is accurate or the image is licensed |
| Canonical or language-version issue | Inspect valid URLs and the relationship between the actual pages | That the chosen canonical or international architecture is correct for the whole site |
The exact checks depend on your report. For example, the meta-description audit fails when the tag is absent or its content is empty. The documentation explicitly says Lighthouse does not assess description quality. Filling the field with arbitrary text may pass that audit without improving the answer shown to a searcher.
For the writing itself, see titles and descriptions (Chinese). Google may generate different search presentation text; passing Lighthouse does not guarantee an untruncated title or a fixed click-through-rate increase.
What does SEO 100 leave unchecked?
Lighthouse does not perform a complete content strategy or factual review. It cannot verify a product specification, decide whether an article answers its reader's question fully, or establish that a photograph shows the claimed event. Those tasks need appropriate documents and editorial judgment.
There is also no Lighthouse rule requiring every informational page to exceed 800 words. Decide article length from the question and necessary evidence. Check internal links, useful alt text and page structure in context rather than relying on a fixed word or link count.
Structured-data checks require the relevant schema and search-feature tools. Valid JSON-LD does not guarantee rich results, and repeated references to the same entity are not automatically a duplication error. Compare the data with visible content and the requirements for the specific feature.
Indexing controls also need an intentional decision. A canonical can point to another genuinely equivalent page. A permanently removed page without a suitable replacement may correctly return 404. Redirecting every such URL to the homepage can hide a warning without helping the visitor.
If performance is low, investigate the actual metric
The performance-scoring documentation explains how measured metrics are converted to scores and combined. Weights and distributions can change by version. The Opportunities section is not a list of fixed point deductions, and these weights are not Google's search-ranking formula.
- Late LCP: identify the element and distinguish server response, resource discovery, downloading and rendering delays. If text is the LCP element, image compression alone may not address it.
- High TBT: inspect long tasks and the scripts involved, then test the interactions visitors actually use.
- High CLS: find the moving elements and check reserved space for media, font behavior and late-inserted content.
For images, compare format, compression and visible quality. For layout shifts, follow a CLS investigation (Chinese). Analytics, advertising and support scripts have business functions; establish those requirements before changing when they run.
Test conditions, caches, ads and third-party responses can change between runs. Compare the samples and metric values instead of treating a one-point increase as a proven improvement for real visitors. Lighthouse findings and page-speed priorities (Chinese) should lead to a measured repair.
Accessibility and mobile usability still need hands-on checks
Automated checks can identify some name, contrast and structure problems. They cannot establish full accessibility. Check keyboard order, visible focus, menus and forms, zoom, touch targets and relevant assistive-technology behavior.
Low visual contrast can make text hard to read for sighted users; it does not mean a screen reader cannot extract text because the color is pale. Likewise, a small tap target is a usability issue, not a direct measurement of INP or a fixed ranking penalty. Keep the problem and its evidence aligned.
Combine the report with indexing and search evidence
PageSpeed Insights combines Lighthouse lab diagnostics with Chrome UX Report field data when sufficient data is available. Inspect whether that field data applies to the URL or origin, and whether it covers the relevant device group. It is not an unrelated alternative algorithm to Lighthouse.
Use Search Console for the site's indexing and search observations, and a crawler for relationships across URLs. Research tools such as Semrush can help with queries and comparison pages, but their estimates do not replace the site's own analytics or business records.
A useful completion record includes the original failed check, its cause, the exact change, the new report and a live-page check. Record technical passes, content review, actual indexing, search traffic and genuine inquiries as separate outcomes.