Skip to content
search-engine-optimizedby Netsolution

Speed and loading experience

Three metrics describe a page's loading experience: how long the largest element takes to show, how long interactions take to respond, and how visually stable the page stays. They are published with their thresholds. There is, however, no single page experience signal, and content relevance comes first.

What are you measuring when you measure speed?

Not one duration. Three distinct things, describing three different moments in the relationship between a page and the person opening it.

Three metrics are published as core web vitals: the render of the largest visible content, the latency of interactions, and cumulative layout shift.

officialSource : web.dev, “Web Vitals”

The first moment is the wait: how long before the essential thing shows. The second is the reaction: how long between a gesture and its visible consequence. The third is stability: does what is displayed stay put, or does the page move under the finger.

Those three questions have no reason to share an answer. A page can appear very fast and react to nothing for several seconds; another can be slow to display and then perfectly stable. That is why no single figure sums them up, and it is already a warning about overall scores.

A Swiss mechanical stopwatch resting on a stone, its dial scratched, its graduations fine, both hands stopped.
A dial that gives a duration, very finely, and nothing else. It does not say whether the wait felt long, nor what the person was looking at meanwhile — and that is where measuring parts company with understanding.R. Henrik Nilsson — CC BY 4.0 · Wikimedia Commons, relevé le

Why are the thresholds not read on one measurement?

The recommended thresholds are assessed at the 75th percentile of page loads, and not on a single measurement.

officialSource : web.dev, “Web Vitals”

That detail completely changes how you work. A test run from your machine produces one measurement, under conditions you chose without meaning to: your connection, your device, your cache. The threshold bears on the distribution of real loads, and it keeps the 75th percentile — in other words, it demands that a large majority of visits be good, not that one isolated test be.

The consequence is nearly always unpleasant to discover: a flattering lab score coexists very comfortably with a quarter of visitors having a painful time, typically on mobile and on a slow network. The two figures do not contradict each other, they measure different populations.

For interaction latency, the bounds are published:

For interaction latency, the threshold for the “good” band is two hundred milliseconds or less, and the “needs improvement” band runs up to five hundred milliseconds.

officialSource : web.dev, “Interaction to Next Paint (INP)”

Those values come from the browser vendor and are quoted with the date they were read, because they have already changed — including in the choice of which metrics count as core. A page citing them without a date will go out of date without warning.

What place does speed actually occupy?

Here are the two sentences that put the whole subject the right way up, and that are missing from most technical audits.

There is no single page experience signal: the ranking systems look at a variety of signals that align with the overall experience.

officialSource : Google Search Central, “Understanding page experience in Google Search results”

The engine still aims to show the most relevant content, even if the page experience is poor.

officialSource : Google Search Central, “Understanding page experience in Google Search results”

The second is remarkably blunt. It says that a page which is unpleasant to use but relevant will come ahead of a page that is pleasant and off topic. It therefore forbids selling speed as a standalone lever for position, which is exactly what half the audit market does with it.

What remains, on the other hand, is solid:

Where several pieces of useful content exist for the same query, a good page experience can contribute to success in search.

officialSource : Google Search Central, “Understanding page experience in Google Search results”

In other words: a differentiator between comparable pages, never a substitute for the content. It is a modest and real place, and it is enough to justify the work — all the more so since the main benefit of a fast page was never the ranking, but the fact that people read it.

Fixing what genuinely gets in the way of reading before optimising a score

hypothesis

Effort
moderate
Time to effect
weeks
Risk
none
Durability
lasting

On this site, measured on 2026-08-16, the fonts are the heaviest fixed item on every page — heavier than the document itself — including on pages with no illustration.

testedProtocol : Download the document, then every resource it declares, and add up the bytes received by kind. Two traps to avoid, both of them hit: count only one version per srcset, and do not count the page itself among its resources. Five pages measured on 2026-08-16; full record in section C.3 of the public register.

Where do you actually start?

With what produces the widest gap between the lab and the field: the weight of third-party scripts, images served without declared dimensions — the first cause of layout jumps — and fonts loaded without a readable fallback. Those are three structural fixes, made once, that survive rebuilds.

Then comes the question of rendering. A page whose essential content depends on scripts running stacks up two problems: the visitor’s wait and the engine-side render lag, handled in the guide on JavaScript. Both are fixed in the same place, which is good news.

What comes only afterwards, deliberately: hunting score points. Moving from an already decent score to a perfect one consumes a considerable amount of time for a gain the thresholds, measured at the 75th percentile on real visits, will probably not register.

What does not work, or no longer works

Push your PageSpeed score to a hundred out of a hundred to improve your ranking: speed has been an official ranking criterion for years.

The documentation states the opposite on both counts: there is no single page experience signal, and relevance comes first even over a poor experience. As for the score, it comes from a single test while the thresholds are assessed on the distribution of real visits.

One more discreet trap remains: optimising for the measuring tool. Delaying the load of everything not immediately visible improves a score and can degrade interaction latency, which is precisely one of the three metrics. A score is not an experience; it is an approximation of one, useful for as long as you do not mistake it for the goal.

Sources

  1. Understanding page experience in Google Search results — Google Search Centralaccessed on
  2. Web Vitals — web.dev (Chrome team)accessed on
  3. Interaction to Next Paint (INP) — web.dev (Chrome team)accessed on
  4. Understand the JavaScript SEO basics — Google Search Centralaccessed on

Block revalidated on

Frequently asked questions

Is speed a ranking criterion?
The official documentation says there is no single page experience signal, and that content relevance comes first even where the experience is poor. Speed therefore counts as a differentiator between comparable pages, never as a standalone lever for position.
Why is my tool score good when my users complain?
Because a lab test simulates a load under conditions you chose, while the thresholds are read at the 75th percentile of real loads. A quarter of your visitors may be having a markedly worse experience than a single test displays.
What threshold should I aim for on interaction latency?
The documentation places the “good” band at two hundred milliseconds or less, and the middle band up to five hundred milliseconds, always at the 75th percentile. Those bounds come from the browser vendor, and they have already changed in the past.
Should you optimise speed before content?
Rarely. A fast page that answers beside the question stays off topic, whereas a slow and relevant page can come up perfectly well. The reasonable order is to fix what genuinely gets in the way of reading, then go back to the substance.