THE SHORT ANSWER
Use PageSpeed Insights to find performance problems, then check those problems against the actual customer experience. Fix what gets in the way, retest it, and ask what the work changed. Don't buy a whole rebuild just because a report looks intimidating.
Start with a task, not the score
Pick a task that matters to your business. For a salon, that might be finding a service and checking appointment availability. For a plumber, it could be finding the service area and opening the quote form. These are examples to test, not results from a client project.
Use the page a customer would actually land on. An ad might send them straight to a service page, so testing only the homepage would miss that part of the visit.
Write down where you hesitate. Maybe the main image appears late. Maybe a button moves just as you reach for it. Or maybe the page loads quickly and you still can't find the phone number. Keep those observations separate. They need different fixes.
If you test a live form, coordinate with whoever receives it and clearly mark the submission as a test. Don't accidentally fill the team's inbox with pretend leads.
PageSpeed is showing more than one kind of evidence
Google's PageSpeed Insights combines a simulated Lighthouse test with real visitor data when enough eligible data is available. The simulated test is useful for investigating a page under test conditions. The visitor data describes experiences already recorded from eligible Chrome users.
They can disagree. Your page may behave differently across devices and connections. Missing visitor data also doesn't establish that your website is broken. A newer page or one with too few samples may not have enough data to show.
Enter the exact public page address into PageSpeed Insights and review the mobile result. Keep a note of which page you tested and when. Then compare its findings with the task you just tried on your phone. Don't treat your own phone visit as a substitute for broader visitor data either.
What the labels mean in plain English
Core Web Vitals are measurements of parts of the page experience. You don't need to memorize the abbreviations to ask useful questions about them.
- Largest Contentful Paint, or LCP, concerns when the largest visible image or text block appears. Ask what is holding up that content.
- Interaction to Next Paint, or INP, concerns how promptly the page responds visually to interactions. Ask which interactions are slow and what is delaying the response.
- Cumulative Layout Shift, or CLS, concerns unexpected movement of page elements. Ask what is moving and why space wasn't reserved for it.
Turn the report into a useful conversation
Send your developer the page address, the report, and the specific problem you noticed. "The quote button moves when the photo loads" gives them something to investigate. "Make the site faster" leaves a lot open.
Ask them to explain the cause before approving the work. A large image might need attention. A delayed booking tool might need a different investigation. Don't assume either is the cause just because it sounds plausible.
Ask what they expect the change to improve, what it could affect elsewhere, and how they'll check the result. Keep a copy of the page before the change so you can compare the customer task afterward.
For paid work, get the scope in writing. Our pricing page explains our current options, but a report alone can't establish what a particular site needs.
Retest without turning it into a points contest
Lighthouse scores can vary because the underlying testing conditions change. Chrome's documentation recommends thinking about a range of performance results rather than treating a single number as the whole story.
Keep your comparisons sensible. Check the same page and device category. After a change, repeat the customer task as well as the report. Confirm the menu still opens, the form still works, and the booking route still goes where it should.
A report can't decide how much your next improvement is worth to your business. Ask for a clear reason to keep going, especially when the remaining work is about a small score difference rather than a problem someone can demonstrate.
A better score is not a ranking guarantee
Google says Core Web Vitals are used by its ranking systems, but good scores do not guarantee top search rankings. Its guidance also cautions against chasing a perfect score only for SEO.
Take performance seriously without accepting promises the report cannot support. If someone says a score increase will put your business above a competitor, ask for the basis of that claim.
For your next website conversation, bring the actual page address and a short note about the task that gave you trouble. "I couldn't book without the page jumping" is a useful place to start.