Your site is fast on your wifi. That is not the test.

Speed tools contradict each other constantly. Here is which number to trust, and the two-minute test that shows what your customers really get.

May 2026 · Issue 1 · Joseph Deep

Website performance figures on a laptop screen

Welcome to the first issue. The plan is one email a month, and it will always be the same shape: one thing worth understanding, one thing worth doing, and no padding. If that stops being true, unsubscribe and I won’t be offended.

The thing worth understanding

Everyone tests their own website on a fast laptop, on their own wifi, with the site already cached in their browser. It loads instantly. It always does.

That is not the test. Your customer is on a phone, on mobile data, somewhere with two bars, and they have never visited before, so nothing is cached. That visit can easily take four times longer than yours did.

This is why speed tools seem to contradict each other. They aren’t measuring the same thing:

  • Lighthouse / PageSpeed Insights runs a simulated slow phone. It’s harsh on purpose, and it’s closest to your worst real visitor.
  • GTmetrix and friends often test from a fast server on a fast connection. Flattering, and less useful.
  • Your own browser is the least reliable of the three, because it’s cached.

If two tools disagree, the pessimistic one is usually describing a real person.

The number that actually matters

Ignore the overall score out of 100. It’s a blend, and it moves for reasons that don’t affect anybody.

Watch Largest Contentful Paint instead — how long until the biggest thing on screen has actually drawn. That’s the moment a visitor stops wondering whether the page is broken. Under 2.5 seconds is fine. Over 4 seconds and you are losing people who will never tell you they left.

The usual culprits, in the order I find them: enormous uncompressed images, a video header that loads before anything else, and five tracking scripts that each seemed harmless on their own.

The thing worth doing

Open your site on your phone, on mobile data, with wifi switched off. Not on your office wifi — actually turn it off.

Count out loud until the page is readable. That number is your real speed, and it’s the only test that doesn’t lie to you.

If the count gets past four, the fix is usually smaller than you’d think. It’s almost never “rebuild the site”. It’s normally three images and a script.

One thing I shipped this month

A client’s homepage was loading a 4MB background photo — a photo you could barely see behind the text. Compressed properly it came to 180KB, looked identical, and the page went from six seconds to under two on a phone. That was the entire fix.

Next month: the Google Ads settings that quietly spend your money on nothing.

Want this kind of thing about your business?

The newsletter is general. A look at your actual site and ads is not. Ask and you'll get a straight answer from the person who'd do the work.

Ask us