Do font preloads make a website faster?

Sometimes — and sometimes they make it measurably slower: when we tested adding font preloads to one of our own production insurance sites, mobile LCP regressed by roughly one second, so that site ships without them (SAA production test, PageSpeed Insights mobile, 2026).

The qualifier: this is not an argument that preloads are bad — it is an argument that they are a bandwidth-priority decision, not a free win. Plenty of pages do benefit, which is why the advice survives. But "add font preloads" sits on nearly every performance checklist as if it had no cost, and it has a specific one that Google's own documentation names: preloaded resources are fetched early at elevated priority, competing with everything else the page needs first. Whether that helps or hurts depends on what your page's LCP element is.

What our test showed, and the mechanism behind it

• The change: standard <link rel="preload" as="font"> hints for the site's web fonts, added to a production insurance agency site that was already fast. (SAA production test, 2026.)
• The result: mobile LCP regressed by roughly one second in PageSpeed Insights. Desktop, with bandwidth to spare, did not show the same penalty. The preloads were removed; the site ships without them.
• The mechanism: the page's LCP element is the hero image. On PSI's throttled mobile connection, the preloaded font bytes moved to the front of the queue and the hero image bytes moved back — the browser painted text sooner and the largest element later. That is the trade the metric recorded.
• The corroborating guidance: web.dev's preload documentation warns that unnecessary preloads delay other resources, and its font best-practices guidance flags that preloading fonts can take bandwidth from higher-priority resources — the caveat is in the fine print of the same sources the checklists cite. (web.dev, "Preload critical assets" and "Best practices for fonts," verified live 2026-08-18.)

The honest lesson we took from it: best practices are hypotheses, and your page is the experiment. A recommendation tuned on pages where text is the LCP element inverts on pages where an image is. This is why every optimization on our builds — which measure 99-100 on mobile PageSpeed — gets a before-and-after measurement in the same tool under the same lab conditions before it ships, and why "it's a best practice" is never, on its own, a reason to keep a change.