Start with the pages and actions a customer needs to complete an enquiry. Choose a destination page, a search or offer page and the final enquiry step. This gives a performance review a business purpose beyond improving a homepage score.

Record the exact addresses, device category and journey you are testing. Keep that set for later comparisons so an apparent improvement is not simply the result of testing a different page.

Understand the three measures

Largest Contentful Paint, or LCP, concerns when the largest eligible visible content renders. Google’s good threshold is 2.5 seconds or less. Interaction to Next Paint, or INP, concerns responsiveness to qualifying interactions, with a good threshold of 200 milliseconds or less. Cumulative Layout Shift, or CLS, measures unexpected visual movement, with a good threshold of 0.1 or less. Assess these at the 75th percentile of visits, with mobile and desktop considered separately. Source: web.dev: Largest Contentful Paint Source: web.dev: Interaction to Next Paint Source: web.dev: Cumulative Layout Shift

Use the measurements to ask different questions: what appears late, which action responds slowly and what moves unexpectedly? Keeping the questions separate makes the improvement brief more precise.

Separate a test run from visitor experience

PageSpeed Insights combines laboratory testing with real-user information where available. Its field data covers a trailing 28-day period. A page without sufficient data may not have a page-level field result, so check whether the report is showing the page or the wider origin. Source: Google: About PageSpeed Insights

Save the report and its scope. Treat a laboratory result as a controlled diagnostic and the field result as evidence from the measured visits. After a change, do not expect a historical collection period to describe only the new version.

If field data is unavailable, say so in the review. Continue with controlled tests and observed customer journeys rather than inventing a real-user score.

Exercise the controls people need

Use a fictional enquiry to open the date picker, change traveller numbers, apply a filter and submit valid test details. Note any delay, unexpected movement or lost selection. These observations help explain what a metric might mean in a practical journey.

For illustration, distinguish a date control that reacts slowly from a search that acknowledges the action promptly but waits for a supplier result. INP does not measure the entire supplier operation. Source: web.dev: Interaction to Next Paint

Ask the team to investigate each issue with the exact steps and evidence. Avoid prescribing a technical fix from the score alone, because the measured element or interaction should guide the diagnosis.

Prioritise a small, testable change

For each issue, record the affected page, the customer action, the observed problem and the intended outcome. Assign an owner and agree how the change will be checked. This makes the work easier to evaluate than a broad request to improve speed.

Repeat the same journey after the fix. Check that images remain accurate, filters preserve the selected details and the enquiry reaches its intended destination. A performance improvement should be reviewed alongside the task the page serves.

Keep search claims proportionate

Google says its ranking systems use Core Web Vitals, but good scores do not guarantee high rankings. Record technical improvements separately from search impressions and enquiries, then assess those outcomes with the appropriate reporting. Source: Google Search Central: page experience

Keep the baseline, change and retest together. That record gives your next review a useful starting point and helps you decide which remaining issue deserves attention.

Written and updated by Travelgenix. Sources checked on 16 October 2026. Worked examples and checklists are editorial illustrations and recommendations, not customer results.