Half the sites we audited ship images without dimensions — and their CLS score says they're fine

The oldest fix in web performance, still not applied

An image without width and height is a hole in the layout. The browser has no idea how much room to leave, so it leaves none, paints the text — and then shoves everything down when the file finally lands. Reserving that space is one attribute pair, it has worked in every browser for decades, and it costs nothing at runtime.

So the question isn’t whether people know. It’s how often the markup that actually reaches a browser carries it. Across 162 audited sites, 86 of them — 53% — ship at least one image the browser cannot size in advance. Between them we had to measure and correct 1,010 individual images: a median of 10 on each affected site, and 24 on the worst.

The score that should catch it, doesn’t

Here is the part we did not expect. Of those 86 sites, 82 — 95% — score “good” on Cumulative Layout Shift, at or under the 0.10 threshold. 26 of them score exactly zero. Across the whole base, a third of the sites report a mobile CLS of precisely 0.

Not because the images are sized. Because of how the measurement works.

A lab run like Lighthouse (the engine behind PageSpeed Insights) loads the page once, in a simulated browser, at a fixed viewport — and then stops. It doesn’t scroll. It doesn’t click. A shift only counts if it happens while the tool is watching, and most of these images are below the fold, so the shift they cause happens on a scroll that never occurs.

This isn’t our reading of it — Google documents the gap directly: lab CLS “only considers layout shifts that occur above the fold and during load”, while the field metric “considers all unexpected layout shifts that occur throughout the lifespan of the page, including content that shifts as the user scrolls”. Our numbers just say how often that gap has something real hiding in it.

The defect is real and sitting in the markup. The number people check to find it is looking the other way.

It tracks the platform, not the team

Split the same 162 sites by what built them and the pattern is not about diligence:

Platform	Sites shipping unsized images	Images we fixed
Duda	41 of 48 — 85%	452
Bitrix	3 of 4 — 75%	56
Shopify	3 of 6 — 50%	48
WordPress	23 of 48 — 48%	275
Tilda	0 of 6 — 0%	0
Next.js	0 of 6 — 0%	0

The two platforms with a perfect record are the two that build the image tag for you and refuse to emit one without dimensions. The ones at the top hand that job to a theme, a widget or a content editor — and it gets dropped, over and over, on sites that are otherwise professionally maintained.

Treat the small counts with care: six Tilda sites and six Next.js sites are not proof of a guarantee, only an absence of counter-examples in this base. The Duda and WordPress rows, at 48 sites each, are the two we would defend.

What to do with this

If you own a site, the useful conclusion is not “add width and height” — you already knew that. It’s that a green CLS score is not evidence you did. Check the markup, not the metric: open any page, look at a handful of <img> tags below the first screen, and see whether they carry dimensions or a CSS aspect-ratio.

If they don’t, the fix is per-image and permanent, and the official advice is one line: always include width and height on images and video, or reserve the space with CSS aspect-ratio. Where it gets awkward is when the markup isn’t yours to change — a builder’s template, a theme, a plugin’s gallery widget. That is the case we built WebSpeed for: it reads each image’s real dimensions from the file header and writes the ratio into the page as it’s served, so the browser reserves the right box without anyone guessing a 1/1 and squashing a portrait crop.

Either way, the check is the same and it takes a minute: look at what your pages actually ship, rather than at the score that failed to mention it.