A working notebook on what Google's own performance research says about load time and behavior, read carefully and written down in plain language.
Read a real Lighthouse report
Reading the score, not chasing it
Core metricLCP
The research
Google has published a fair amount of its own thinking on this over the years, mostly through web.dev, the Chrome team's engineering blog, and the public Chrome User Experience Report. None of it reads like marketing. It reads like engineers trying to explain why a page that loads in three seconds behaves so differently from one that loads in one.
The pattern that keeps showing up isn't a straight line. Bounce probability doesn't creep upward evenly as load time increases. It behaves more like a threshold effect: people tolerate a little friction, then their patience drops off quickly once a page crosses some point that varies by device, connection, and what they came to do.
"The relationship between speed and user behavior isn't linear. Small delays are absorbed. Past a certain point, tolerance falls away fast, and that point moves depending on what the user expects to happen next."
Paraphrased from Google's publicly available web performance documentation
This is why the blog exists in the current form. We're not trying to sell anyone a fix. We're trying to read the same public documents, the same CrUX dataset, the same HTTP Archive snapshots that Google references, and write down what we notice without inflating it into a bigger claim than the data supports.
Context matters
A delay lands inside a decision that's already fragile. The shopper has a competing tab open, a card in hand, and a dozen reasons to reconsider. A stalled page doesn't just cost a second, it hands back the moment of hesitation that the whole page was designed to move past.
The reader already clicked through from a search result or a link, so there's some existing intent to arrive. A slow load here is more likely to be read as clutter or irrelevance than as a reason to leave outright, though it still raises the odds. The bounce is often about relevance mismatch first, speed second.
This is the piece that gets flattened in most speed advice: the same delay, measured the same way, doesn't carry the same weight for every kind of page. Knowing which one you're running changes which metric deserves your attention first.
No code required
Lighthouse, the free auditing tool built into Chrome, produces a report most people glance at, note the score, and close. The score is the least interesting part. The Opportunities and Diagnostics sections underneath it are where the actual story lives, written in sentences rather than jargon.
Priorities
The pattern we keep seeing: the first few fixes on a slow page tend to matter a great deal. Somewhere past that, effort and perceptible benefit stop moving together, and further work becomes more about the number on the screen than the person looking at the page.
Felt speed
A page can pass every metric in a Lighthouse report and still leave someone with the impression that it dragged. Perceived performance is the gap between how long something actually took and how long it felt like it took, and that gap is shaped by things a stopwatch doesn't capture.
A skeleton screen that appears instantly, even if the real content takes another second, tends to read as faster than a blank white page followed by an abrupt content flash. Small, well-timed motion can smooth over a wait. Abrupt layout jumps do the opposite, even on objectively quick pages, because they interrupt whatever the reader was already doing with their eyes.
This is the part of the conversation that gets skipped most often. Two pages can share an identical load time and produce different reactions, because one of them was designed with the waiting in mind and the other wasn't.
Behind the articles
Four people who read the same public research and disagree about parts of it more often than you'd expect.
Research Editor
Data Analyst
Walkthroughs Writer
Tools Curator
Start here
Screenshot by screenshot, with the jargon translated as we go. No prior knowledge of code assumed.
Open the walkthroughsOr, if you'd rather explore
The public tools this blog actually references when writing an article, described plainly, limits included.
See the toolsMethodology
Not certifications. Just the public datasets and standards behind every claim we make.
Real-user field data collected by Chrome, publicly queryable, and the closest thing to ground truth we reference.
Lab data run locally against public pages, useful for structure and diagnostics rather than absolute claims.
A historical, openly published crawl of how the web loads over time, updated on a regular public schedule.
Google's own explanations of the metrics behind Core Web Vitals, read in full rather than skimmed for headlines.