What slowness costs
Last updated
A slow page loses people before they've seen what you sell. Speed shows how fast your pages load for real visitors, and how many conversions slow loads cost you.
Speed works on the web only. The web SDK measures every page load on its own. You don't need to tag anything.
Open Speed
In your app, select Speed. The top shows the conversions lost to slow loads, and the share of sessions that passed Core Web Vitals. Converting means reaching any goal. Choose a saved task to see task success by speed.
What's measured
| Measure | What it means | Good | Poor |
|---|---|---|---|
| Largest Contentful Paint (LCP) | How long until the main content shows | 2.5s or less | Over 4s |
| Interaction to Next Paint (INP) | How long the page takes to react to a click, tap or key | 200ms or less | Over 500ms |
| Cumulative Layout Shift (CLS) | How much the page jumps around while loading | 0.1 or less | Over 0.25 |
| First Contentful Paint (FCP) | How long until anything shows | 1.8s or less | Over 3s |
| Time to First Byte (TTFB) | How long the server takes to start answering | 0.8s or less | Over 1.8s |
Anything between good and poor is rated Needs work. These are Google's thresholds.
Core Web Vitals is the worst rating of LCP, INP and CLS. A session passes only when all three are good.
The typical load
Each measure shows its p75: 75% of page loads were this fast or faster. It's the figure Google uses, and it's taken over page loads, not sessions. Next to it is the share of loads rated good, needs work and poor.
What slowness costs
Metrickle puts each session in the bucket of the worst rating it had: good, needs work or poor. Then it compares how each bucket converted.
Lost is how many conversions the slower sessions fell short of the fast sessions' rate. If your goals carry revenue, it's shown as money too. Site-wide, the figure uses Core Web Vitals.
The gap is marked clear when slow sessions convert less beyond what chance explains, with at least 10 sessions on each side.
Pages
Pages lists your most measured pages, with the most conversions lost first. Each page has its p75 for every measure, the share of good, needs-work and poor loads, and what slow loads cost on that page.
Start with the page at the top. Fix what makes it slow, then watch its number.
Who it hits
Speed is split by device type and by accessibility setting, such as zoom or keyboard navigation. For each, you see the p75, the share of poor loads, and how fast and slow sessions converted.
The settings come from the same page load that was measured, so a setting someone only turned on later in the visit doesn't count.
What it doesn't tell you
Slow loads and lost conversions go together, but not only because of speed. People on older phones or weak connections get slower pages, and they may convert less for other reasons too. Slow devices and networks are part of any gap. Treat the number as where to look first, fix the slowest page, and check whether conversions follow.
Slow loads also appear on Converters, next to every other trait, where you can ask people about them. See what sets converters apart.
Ask an AI assistant
Connected AI assistants can read the same results with get_page_speed. See the MCP server.