Loading time

Loading time sounds like a single number, but it is a sequence. First the server answers, then the text arrives, then come images, fonts and everything loaded in from outside providers. For the visitor only one point in that sequence counts: the moment they see what they clicked for. Whatever loads in after that, they usually do not even notice.

It is exactly this moment that is measured by the value everyone talks about today. In the jargon it is called Largest Contentful Paint, roughly: the point at which the largest visible element is in place — as a rule the main image or the headline at the very top. Google draws the line at 2.5 seconds. Below that the value counts as good, from 4.0 seconds as poor, and in between lies the range where improvement is due.

Why this counts for a business has less to do with technology than with patience. Someone searching for a tradesman on their phone over a mobile connection has three more hits sitting in the list. A page that stays blank for a while loses the visitor before it has even shown what it can do. Google additionally counts loading time among the signals that feed into its assessment — as one of many, not as the decisive one.

The causes are almost always the same three. First, images uploaded at full camera resolution and only scaled down in the browser. Second, typefaces fetched from someone else's server before a single letter appears. Third, everything that comes along on top: map views, chat windows, review stars from outside portals, tracking tools. Each single piece looks harmless; together they decide the seconds.

In measuring there lies a trap that surprises many: the same page does not get the same score from two measurements taken shortly after one another. Network, server load and caches fluctuate, and the common testing tools extrapolate part of the result instead of measuring it. Anyone wanting to make a statement measures several times and states the range instead of a single number. How that is done cleanly is set out in our guide to measuring a website.

What actually helps is unspectacular. Deliver images at the size at which they are displayed, and in a modern format. Supply fonts with the site instead of loading them in. Reduce third-party scripts to what is genuinely needed — and load the rest only once the visitor has consented. Added to that is where the site sits: a server in Germany answers German visitors faster than one on another continent, see hosting.

What we advise against is chasing the score. It is possible to trim a page to a high score that still feels sluggish to use — because content only appears on scrolling and jumps about as it does. Conversely there are pages with a middling score that work perfectly well. The score is a pointer to possible weak spots, not a picture of the experience.

For you that means: loading time is not a one-off project but a property that is at stake again with every change. A chat window added later can undo months of work. We therefore measure before and after and disclose both values. We do not promise a particular score, because it depends on things outside our hands.