Keep the test conditions comparable
In Performance Investigator, enter the affected public URL and choose mobile or desktop. An optional comparison page uses the same profile. The report describes this test server and observation window. Visitor devices, network conditions and page states can differ.
Find the likely work behind the wait
Inspect the observed Largest Contentful Paint element and resource, then review costly requests and third-party hosts. A long request and a late paint can be related, but their timing alone does not prove causation. Cross-origin response sizes without timing permission remain unknown; zero must not be interpreted as a free request.
Example: a delayed hero image
This illustrative case uses a synthetic page with a large hero image. The image becomes the observed LCP element. Its request completes late, so the next useful experiment is to reduce its transfer size or remove an unnecessary loading delay. Repeat the test with the same screen size. Compare the run range as well as the reported central value; a single quicker run is weak evidence.
Bring a browser capture when the problem is local
If the issue occurs only in your browser, export a HAR from the affected state and compare it with a successful capture in Private Browser Network Analyzer. The files stay in your browser. Match the page state and time window. Requests with the same path can still differ by host, query or occurrence.
Retest a specific hypothesis
- Save the original report and the exact test settings.
- Change one relevant resource or loading behavior.
- Run the same test and compare compatible results.
- Check the page and its main user journey for regressions.
A short lab run cannot replace field data from real visitors. Use missing or incomplete evidence as a reason to investigate further, not as a successful performance result.
Reference: Largest Contentful Paint.