Export the report you actually tested
Open the page in Chrome, then open Developer Tools and its Lighthouse panel. Choose a navigation audit, select mobile or desktop, and run the categories you need. Save the finished report as JSON from the report menu. A Lighthouse command-line JSON export also works; a PageSpeed Insights API response must contain lighthouseResult. Select that file in the tool, or paste its JSON. The analyzer reads existing evidence: it does not visit the audited site or run a new test. The example button loads a reduced, real report from Dardo’s web-design page captured on 27 September 2026. Its original scores and check outcomes are preserved; screenshots and resource details are removed. That run reached its time limit, so results may be incomplete.
Read measurements before the overall score
A performance score summarizes several weighted measurements. Start with the actual metric and the user-visible problem. Largest Contentful Paint concerns when the largest content element appears; Total Blocking Time concerns blocking during the lab run; Cumulative Layout Shift describes unexpected layout movement. This tool displays milliseconds for timing metrics and a unitless value for CLS. Zero is a valid measurement, while an absent, invalid or errored measurement remains unavailable. A lab run cannot provide field INP or establish that a URL passes Core Web Vitals for its real visitors. Keep Search Console or other field observations beside the report when deciding what to improve.
Make each flag an inspectable task
The review groups flagged checks by performance, accessibility, SEO and best practices. It also counts manual, not-applicable, informational, unknown and error outcomes separately. Open the original report to locate the element or resource behind a flag; the review download deliberately excludes those raw details. For example, a contrast flag needs inspection of the actual foreground, background and interaction state. An unnamed-link flag needs a meaningful accessible name in its context. Assign the affected component, a person responsible and an acceptance check. These groups organize work; they do not estimate hours, commercial impact or the priority of an issue for your particular users.
Compare runs with matching conditions
Add a later report in the optional second input. The analyzer checks the recorded requested and final URLs, Lighthouse and browser versions, device, network and CPU settings, viewport, language and storage-reset settings. Other recorded run settings must also match. Missing conditions, a runtime error or a non-later timestamp withhold the deltas. Query strings remain part of the private URL comparison because they can change a page, although the displayed review removes them. When a comparison is available, each delta is later minus original. A negative timing value means a lower measurement in that run. Repeat tests under controlled conditions before attributing a change to your implementation; server load, extensions, content and hardware can still vary.
Finish with a real journey
After fixing a component, repeat its audit and complete a customer task on the deployed site: find the right service, follow its links, use the keyboard and submit a clearly identified test inquiry. Confirm receipt at the destination. Automated accessibility checks do not cover every requirement, and a high Lighthouse SEO score does not prove Google indexing, rankings or demand. Our Colombian web-provider study publishes a fixed cohort, unavailable attempts and the inspection method to show how we separate an observation from a wider claim. Apply the same discipline to your own before-and-after evidence. Use the Search Console planner separately to decide which existing search pages deserve attention.
Keep the handoff small and useful
Download the review JSON to keep the test date, available measurements, audit identifiers and comparison decision together. The file excludes screenshots, request details, custom audit titles, credentials, URL queries and fragments. URL paths remain and can contain private information, so inspect the download before sharing it. Your selected report is processed locally; no report payload is attached to the inquiry link. Share only the findings needed to agree on an implementation scope, who owns each change, and how the deployed result will be accepted. The first useful deliverable is a reproducible diagnosis and a working fix, followed by observation of actual search and inquiry data.
Sources & further reading
Sources behind this guide, with further detail from the original publishers.
- Chrome: Lighthouse overview and report export developer.chrome.com
- Chrome: performance scoring developer.chrome.com
- web.dev: lab and field data web.dev
- W3C: evaluating accessibility w3.org