Mobile performance across ten client websites

Pilot results

Lighthouse scores out of 100, rounded to the nearest 5. LCP in seconds, rounded to 0.5 s. Ties round upwards. These are not exact medians and cannot classify results at scoring thresholds.

What these figures tell us

Technical SEO checks and mobile performance describe different aspects of a website. An SEO score does not measure speed or Google rankings. The difference between scores is not a percentage improvement.

LCP measures when the largest content element appears, not total page load time. Here it comes from a laboratory test, not real visits. These aggregates cannot show whether the websites with better SEO are also those that load more slowly.

How we measured

1. We selected the ten external websites with Lighthouse monitoring available in our catalogue. We excluded Kiwop’s website and the demo and checked for a distinct project and domain per observation. The unit is a website, not an independent corporate group.
2. We fixed protocol 2026-09-29-p1 before extracting the figures. We took the single mobile observation from each connector’s latest active record, synchronised on 29 September 2026 (Europe/Madrid). One URL per website, without rerunning tests to pick a better result.
3. The source is PageSpeed Insights v5. We use Lighthouse laboratory scores and LCP, excluding CrUX and all real-user data. All five metrics have ten valid values. A missing value would have suppressed the entire metric, without replacing it with zero.
4. We calculated each metric’s median and applied the stated rounding. An independent calculation and tests for missing values, duplicates and invalid inputs verified the results.

What cannot be concluded

• This is an availability-based sample of Kiwop clients, with whom we have a commercial relationship. It is not a random sample or an independent market audit. It cannot compare the public and private sectors.
• One URL and one observation per website do not describe typical performance or all pages. The Lighthouse version, execution time and full environment were not retained. The date is the synchronisation date. This is not a controlled benchmark.
• We calculate no trends, confidence intervals, causal relationships or sales impact. Automated accessibility does not certify legal compliance, and laboratory LCP does not establish a Core Web Vitals pass.

What data we share

We publish only rounded medians for the whole sample. We omit names, domains, identifiers, individual rows, sectors, locations and extremes. We use no personal browsing data, conversations, sales or case files.

Data and definitions

• Download aggregates as JSON
• Download the table as CSV

Data dictionary

Field names remain in English so all seven editions share the same dataset.

protocol_version / period

Method version and edition month.

strategy / sample_unit / n_sites

Mobile; website as the unit; total websites included.

statistics

The five table metrics: performance, seo, accessibility, best_practices and lcp_seconds.

status / n_sites

Availability and valid observation count for each metric.

median_rounded / rounding_step / unit

Rounded median, rounding step and unit: score_0_100 or seconds.

limitations

Dataset limitations. The CSV has one row per metric, not per client.

The method allows the calculation to be repeated in the authorised environment. Downloadable aggregates allow the report’s consistency to be checked, but cannot reconstruct exact medians without the private observations.

How to cite this study

Kiwop Labs (2026). Mobile performance across ten client websites. 2026-09. https://www.kiwop.com/en/labs/client-mobile-performance-2026

Text and aggregates under CC BY 4.0, with attribution to Kiwop Labs. CC BY 4.0.

Version and corrections

Edition 1.0, September 2026. No corrections recorded. Any correction will state here what changed and when.

Method references

• PageSpeed Insights: laboratory and field data
• Lighthouse: performance scoring and variability