Core Web Vitals

"The site is slow" fixes nothing. Slow means different things to different people, and on the machine it was built on it's usually fast. A screen that paints in one second on office wifi and a new laptop takes four on a three-year-old phone underground.

Advertising is where that difference turns into money. You pay per click to bring someone in, and if they leave before the screen appears the spend already happened. Bring ten people and lose four that way and you met six while paying for ten.

Once these numbers became an input to search ranking, page speed moved from being an engineering concern to being marketing's problem. Core Web Vitals pins that speed to three numbers so everyone measures it the same way.

Three things get measured separately, and all three have to clear their threshold to pass.

1. Largest contentful paint (LCP) is how long until the largest piece of content is drawn. Under 2.5 seconds counts as good. The largest piece is usually the hero image or the headline.
2. Interaction to next paint (INP) is how long between a tap and the screen responding to it. The threshold is 200 milliseconds.
3. Cumulative layout shift (CLS) is how much already-drawn elements jump. Keep it under 0.1. A late-arriving image or ad pushing the button someone was aiming at further down the page gets caught here.

How the value is picked matters. It's not an average but the 75th percentile of what real users experienced — 75 of every 100 people had it better than that. Averages let numbers from fast devices hide the slow ones, so four people living with four seconds can leave the figure looking fine.

The place of measurement splits too. Running a tool on your own machine gives one number and collecting from actual visitors gives another, and the second is what counts. Which is also why the number takes weeks to move after a fix. Use the first to find what to fix and the second to confirm it improved.

"LCP is 4.1 seconds. Let's start with the ad landing page."

Fix the screen that paid-for visitors hit first.

"It's 1.2 seconds on my machine and 3.8 in the field."

Different places of measurement, different verdicts.

"Shrinking the image alone takes a second off LCP."

When the largest element is an image, that's the cheapest place to fix.

At Pagedeck, Ravi checked these numbers before increasing ad spend. The landing page's LCP was 4.1 seconds, with the hero image loading at full original size.

Theo resized it to the display size and changed the order fonts download in, bringing LCP to 2.3 seconds. The field number followed three weeks later.