Here's a familiar story. A marketing site ships with a gorgeous new heading font that looked perfect in every Figma frame. Two days after launch the client calls, annoyed, saying the headline "flickered" every time the page loaded. It isn't a bug in the traditional sense. It's FOUT, a flash of unstyled text, and nobody on the team thought about font loading until someone noticed it with their own eyes. Sizes that shift with the viewport make the swap more visible, so it helps to settle them first in the clamp preview.
That's usually how this gets fixed: after someone complains, not before. Font loading strategy sits in a blind spot between design and performance engineering, and it rarely gets budgeted time until it's visibly broken.
TL;DR: Font-display: swap now runs on roughly half of all pages, up from single digits a few years ago (HTTP Archive Web Almanac, 2025). Pair it with metric-matched fallbacks, one preloaded font file, and subsetting, and FOUT/FOIT stops costing you LCP or CLS points.
What's the Difference Between FOUT and FOIT?
FOUT (Flash of Unstyled Text) shows your fallback font right away, then swaps in the real one once it downloads. FOIT (Flash of Invisible Text) hides the text completely until the font arrives, sometimes for a full second on a throttled connection. One lets people read immediately. The other makes your hero section look like it's still loading, because it is.
Put plainly: FOUT is text flashing from one typeface to another mid-read, FOIT is text disappearing entirely until the file lands. Readers forgive the first far more often than the second.
Which one you get depends entirely on the font-display value in your @font-face rule. Leave it unset, or set to auto, and most browsers default toward FOIT-like behavior on custom fonts. That's the trap: doing nothing is the same as choosing invisible text.
Why Do Most Sites Still Get font-display Wrong?
Here's the good news first: font-display: swap is now the majority setting. The HTTP Archive's 2025 Web Almanac puts it at roughly half of all pages, both desktop and mobile, a huge jump from the low single digits it sat at only a handful of years earlier. The bad news? font-display: block is still the second most common value, and around 70 percent of that usage traces back to icon fonts, not body copy, which tells you plenty of teams are setting it once globally and forgetting it applies everywhere, icons included.
@font-face {
font-family: "Inter Variable";
src: url("/fonts/inter-variable-latin.woff2") format("woff2-variations");
font-weight: 100 900;
font-display: swap;
size-adjust: 100.2%;
ascent-override: 90%;
descent-override: 22%;
}
That size-adjust and ascent-override pairing is the part people skip. Without it, swap can still trigger a visible reflow the instant the real font lands, which is exactly what shows up as a CLS penalty in Lighthouse. Match the fallback's metrics first, then swap becomes genuinely invisible.
font-display: swap without metric-matched fallbacks just trades invisible text for a layout jump.
How Do You Preload Fonts Without Wasting Bandwidth?
Preload tells the browser "fetch this now, before you'd normally get to it." Used on one critical font file, it's one of the sharpest tools you have for cutting LCP. Used on four or five, it stops meaning anything, because the browser is now juggling priority requests that all claim to be urgent.
<link rel="preload" href="/fonts/inter-variable-latin.woff2"
as="font" type="font/woff2" crossorigin>
Only around 12 percent of pages use preload at all according to the same 2025 Web Almanac data, which is low given how cheap the fix is. Add this one tag for your body font (and a second heading weight only if it renders above the fold), and skip it for everything else. A page with six preload tags stacked up usually loads slower than it would with just one, because the preloads compete for the same bandwidth.
Does Subsetting Actually Cut File Size?
Subsetting strips glyphs your site never renders, extended Cyrillic, obscure symbols, ligature sets you don't use, out of the font file before it ever reaches a browser. Combined with a variable font, which we've already covered in our variable fonts breakdown, subsetting is where the real payload savings live.
pyftsubset inter-variable.ttf \
--output-file=inter-variable-latin.woff2 \
--flavor=woff2 \
--unicodes="U+0000-00FF,U+0131,U+0152-0153,U+2000-206F,U+2122,U+2212"
Subsetting a variable font to Latin with a command like that can shrink the file dramatically before font-display or preload enter the picture. Google Fonts users get a lighter version of the same idea through the ?subset=latin query parameter, less control, same core benefit.
How Do You Measure the Before and After?
Numbers make this concrete, so measure your own before and after. Run Lighthouse on a throttled 4G profile before touching anything, then again after each change. Four numbers tell the story: LCP, CLS, font payload and time to first visible text. Each fix moves a different one. Preload and subsetting shorten LCP and shrink the payload, swap removes the invisible-text window, and a metric-matched fallback keeps CLS close to zero when the real font arrives.
Will every site see big gains? No, a lot depends on how badly the starting point was configured. A site on font-display: auto, with no preload and a full glyph set, has the most to win. A site that already does half of this has less.
What Should You Ship This Week?
Open DevTools, check the Network tab, and look for your @font-face declarations. No font-display value, or auto? Add swap. More than two preload tags? Cut it to one. Full glyph set on a Latin-only site? Subset it.
That's three CSS changes and one build step, and none of it touches your actual type choices. Nobody screenshots a preload tag for a portfolio, but it's the half of typography work that decides whether your font choice actually shows up on time.
Work out the fallback numbers
The override values in the snippet above are a sketch, because the right ones depend on exactly which two fonts you pair. The calculator below opens on Inter over Arial, where the numbers come out at 107.12%, 90.44% and 22.52%. Swap in your own font, then read the verdict line before you paste anything.