How to analyze and improve website performance: define metrics, pinpoint issues, and verify results

ProcrustesPR

When analyzing website performance, first identify whether content appears slowly, users wait too long after taking an action, or the layout shifts during loading. Then set the test location, device, browser, and connection conditions. Start with PageSpeed Insights (PSI) for page diagnostics and improvement recommendations, and use WebPageTest when you need more detailed data. Depending on the issue, examine images, JavaScript, caching, or the server, and retest after making changes rather than chasing a single score.

ProcrustesPRProcrustesPRHow Should a CMS Handle Images? From Upload Optimization to Descriptions and URL Management

Define the performance problem before choosing metrics

For users, performance is about more than how many seconds a page takes to load. Slow-loading content, delayed responses to actions, and elements shifting during reading are distinct problems. An overall score cannot replace an assessment of each.

  • Content rendering: LCP (Largest Contentful Paint) describes how long it takes for the largest content element to appear and can help assess how quickly the main content loads.
  • Interaction experience: Check whether users have to wait after taking an action; slow responses may cause visitors to leave.
  • Visual stability: CLS (Cumulative Layout Shift) measures unexpected layout shifts; a stable layout makes reading and interaction easier.

Earlier articles described LCP, FID, and CLS as the Core Web Vitals of their time, but that list should not be treated as the complete current specification. This article focuses on loading, interaction, and stability to define performance problems rather than adopting the older articles’ full explanation of the metrics.

Test conditions should reflect your readers; defaults are only a starting point

PSI can assess desktop and mobile pages separately, making it useful for identifying areas worth investigating. To compare different usage scenarios, WebPageTest lets you configure the location, device, browser, and connection type, helping identify regions or conditions where performance is poor.

If you are unsure how to configure your first test, use the tool’s recommended defaults as an initial baseline. The results reflect only the selected conditions, not every visitor’s experience. For example, a website serving multiple countries should compare results from its target regions rather than infer the global experience from a single location. After making changes, retest under the same conditions so you can compare the results.

Choose improvements based on the diagnosis, then verify the changes

Once you have a report, identify which pages the issues affect, then plan the fixes. In its 2024 explanation of Site Audit, Semrush categorizes issues as errors, warnings, and notices, recommending that the more serious errors be addressed first. This classification is specific to that tool, not a universal rule across performance tools.

  • Images: Large images may slow down pages; check compression, file formats, and dimensions.
  • Caching and hosting: Browser caching can reduce repeated downloads, particularly for returning visitors; servers and databases can also be examined.
  • JavaScript: Check its impact on content rendering, but do not treat disabling scripts as the solution.

In one Semrush test, DebugBear was used to measure ten URLs on different domains. Disabling JavaScript improved mobile LCP by at least 30% on every page, but the test also showed that disabling it could remove main content and calls to action. This finding is a starting point for troubleshooting, not a guarantee of similar gains on other websites. After making fixes, rerun the tests to check whether performance has improved and important content still appears.

ProcrustesPRProcrustesPRHow should images of different sizes on the same website be named?

Website performance: perspectives from across the web

Can disabling JavaScript be a way to improve website performance?

Do not treat disabling JavaScript as the solution to performance problems. In Semrush’s mobile tests of ten URLs on different domains, disabling it improved LCP by at least 30% on every page, but some pages also lost their main content and calls to action. This finding can help you investigate the impact of scripts, but it does not guarantee similar gains on other websites. The article recommends placing main content and important details in the core HTML while also considering loading speed.

If your target audience is in Taiwan, must your website also be hosted in Taiwan?

Not necessarily. Compare actual connection test results from the locations of your target readers. An article on hosting emphasizes connection speeds in the target market. In its example, the connection time was 46 milliseconds from Taiwan and 1 millisecond from Singapore. The author therefore concluded that hosting in Singapore was not ideal for a website serving Taiwan. But this is a single example and the author’s judgment. It does not justify recommending Taiwan-based hosting for every website serving readers in Taiwan, nor does it establish a hosting-location rule for all websites.

Can you ignore performance audit notices if they do not affect the health score?

Not necessarily. In Semrush’s 2024 introduction to Site Audit, notices are classified as less serious issues. Although they do not affect that tool’s Site Health score, addressing them may still be useful. If resources are limited, follow its recommendation to address errors first, then warnings, and evaluate notices afterward. This relationship between severity and scoring is specific to that tool’s classification and cannot be directly applied to other audit tools.

ProcrustesPRHow many image variants should you prepare, and how should you manage them?

References

  1. Core Web Vitals 檢測工具推薦tenten.co · 2021-10-18T14:22:31.000Z — Source used in the article above.
  2. SEO操作重要關鍵:分析你的競爭對手www.da-vinci.com.tw · 2023-04-09T20:04:19+08:00 — Discusses: If your target audience is in Taiwan, must your website also be hosted in Taiwan?
  3. 12 Best Website Optimization Tools in 2025 (Free + Paid)www.semrush.com · 2023-09-13T13:31:00Z — Source used in the article above.
  4. JavaScript Explained for Beginners: What It Is and Why It Matterswww.semrush.com · 2021-09-01T13:57:00Z — Discusses: Can disabling JavaScript be a way to improve website performance?
  5. 技術性 SEO:全面指南(19個最佳實踐)tenten.co · 2024-04-28T15:21:55.000Z — Source used in the article above.
  6. SEO and CRO: How to Combine Strategies for Better Resultswww.semrush.com · 2024-03-28T09:25:00Z — Discusses: Can you ignore performance audit notices if they do not affect the health score?

Related articles

How Should a CMS Handle Images? From Upload Optimization to Descriptions and URL ManagementHow should images of different sizes on the same website be named?How many image variants should you prepare, and how should you manage them?How do visual and short-form video trends in search results affect SEO?How can photography websites use SEO to increase visibility and attract clients?How can service and e-commerce websites improve search visibility? Start by distinguishing local needs from buying intentWhy optimize a website, and which tasks can be done for free?How can you use search results page features to attract traffic?
ProcrustesPR
ProcrustesPR