• Oct 22, 2026
  • 3 min read

Lighthouse Numbers Vary: Making Speed Claims You Can Defend

My outreach emails state one measured problem per website (the approach), and the most common one is speed: “your homepage takes over 12 seconds to show its main content on a phone”. That sentence has to survive the recipient checking it.

Then I audited the same site three times and got 4.5 s, 11.8 s and 12.9 s.

Why it varies

Lighthouse’s mobile test doesn’t wait on a real slow network. It loads the page on your machine and then simulates a mid-range phone on a slow connection, extrapolating from what it observed. So anything that disturbs the observation moves the result:

  • CPU contention. I was running two audits in parallel (two Chrome instances) on a laptop with a load average of 6.6. The simulation scales up whatever slowness it sees.
  • The site itself. Cold caches, a slow first byte from the host, a third-party script that’s sometimes slow.
  • Your network on that particular minute.

Run alone, the same site measured 12.3 s and 13.2 s. The 4.5 s was the outlier.

Three rules

1. Measure twice, keep the faster run. If the finding is about speed, the batch audits again and keeps the run with the lower LCP, for both the sentence and the PDF, so they never disagree:

if (f && /load time|MB homepage/.test(f.short)) {
  const again = await audit(lead, '.run2');
  if (!again.error && lcp(again) < lcp(res)) {
    // replace the first run's JSON and PDF with the faster one
    res = again; f = finding(res.json);
  }
}

Choosing the faster run is deliberate: under-claiming costs nothing, over-claiming costs the conversation.

2. Round down. “Over 12 seconds”, not “12.9 seconds”. It reads more naturally, and it leaves a margin.

3. Point to an independent source and check it agrees. The email names pagespeed.web.dev, Google’s own test, because that’s where a curious recipient will go. I checked two sites against it:

SiteMy auditGoogle PageSpeed (mobile)
Example A12.9 s12.3 s
Example B16.3 s16.2 s

Within about half a second, and “over 12 seconds” holds either way.

What the number doesn’t tell you

Neither my audit nor PageSpeed’s lab test is what real visitors experience. Real-user data (Chrome’s UX report) only exists for sites with enough traffic, and none of the small business sites I checked had any. So the claim is carefully worded: “on a phone”, “main content”, Google’s target, and a link to check. Not “your customers wait 12 seconds”.

Takeaways

  • Lab speed numbers vary; parallel runs on a busy machine make it worse.
  • Measure twice and keep the faster result when the number goes in front of someone.
  • Round down and name a source they can check.
  • Word the claim as what was measured, not as what users experience.