User experience (UX)

User experience is the whole of how a person perceives and operates a site: how quickly it becomes useful, whether it responds when touched, whether it stays still while being read, and whether the thing they came for is findable once it does. It is a description of the experience, not a metric, which is why the useful work is always in the specific measures underneath it.

What is actually measured

Google's page experience documentation is explicit that this is not one number: "There is no single signal. Our core ranking systems look at a variety of signals that align with overall page experience." It offers a self-assessment instead — does the page have good Core Web Vitals, is it served securely, does it display well on mobile, does it avoid an excessive amount of ads that interfere with the main content, does it avoid intrusive interstitials, and can a visitor easily distinguish the main content from everything else.

Only the first of those is instrumented. Core Web Vitals define three field metrics with published thresholds: Largest Contentful Paint within 2.5 seconds of when the page starts loading, Interaction to Next Paint at 200 milliseconds or less, and Cumulative Layout Shift at 0.1 or less. Each is assessed at the 75th percentile of page loads, segmented across mobile and desktop — a deliberate choice, because a mean would let a fast median hide a slow quarter.

The rest of the self-assessment list is qualitative on purpose. "Distinguish the main content from other content" has no threshold because it cannot have one; it is a judgement about layout that a reader makes in a second and a measurement tool cannot make at all.

The three metrics name three different failures

They are frequently treated as one score. They are not, and the fixes do not overlap.

• LCP is a delivery problem. The largest element took too long to paint — usually a hero image without dimensions, a render-blocking stylesheet, or a slow server response. See page speed.
• INP is a main-thread problem. The page received an interaction and could not paint a response because script was busy. It is not about load; a fully-loaded page can fail INP badly.
• CLS is a reservation problem. Something arrived late and pushed content that was already being read. Images without width and height, injected banners, and fonts that swap metrics are the usual causes.

Diagnosing all three as "the site is slow" is the most common way teams spend a quarter optimising the wrong thing.

Why this matters for being read by a machine

An agent fetching your page is a user with unusually harsh constraints. It does not scroll, it does not wait for a lazy-loaded section to enter the viewport, it does not dismiss an interstitial, and it does not click "show more". It requests a URL, takes what the response contains, and moves on, often under a timeout.

That collapses a large part of UX into one question: is the content in the document, or is it assembled afterwards? A page whose substance appears only after client-side hydration can be a perfectly good human experience and an empty one for a retriever. A page that hides its answer behind a tab, an accordion whose contents are injected on click, or a consent wall that gates the body, is doing the same thing more deliberately.

The structural half of UX carries over more directly. Headings that state what a section claims, one subject per section, alt text that describes the image, and link text that names its destination are accessibility practice — and they are also the signals an extractor uses to decide which passage answers a question. See mobile-first indexing for the related constraint that the mobile rendering is the one that counts.

Failure modes

1. Optimising the lab score, not the field. Synthetic tools measure one load on one machine. The thresholds above are defined on real page loads at the 75th percentile; a good lab number and a failing field number are entirely compatible.
2. Treating Core Web Vitals as the whole of page experience. Google's own documentation says there is more to it, and that relevance can still win over experience. A fast page with nothing to say does not rank on speed.
3. Interaction-gated content. Anything that requires a click to exist in the DOM is invisible to a fetcher and often to a keyboard user.
4. Late-arriving layout. A banner injected after paint fails CLS and, separately, pushes the first thing a reader was looking at off-screen.
5. Divs all the way down. A page with no semantic structure gives a screen reader no landmarks and a parser no passage boundaries — the same defect, two audiences.

Frequently asked questions

Is UX a ranking factor?

Not as one factor. Google says there is no single signal, and names Core Web Vitals as among the signals its ranking systems use.

What thresholds should I hold?

LCP within 2.5 seconds, INP at or below 200 milliseconds, CLS at or below 0.1, at the 75th percentile.

Does it change whether an AI assistant can quote me?

Through one channel: whether the content is in the fetched document rather than assembled after interaction.

Is accessibility separate?

It is a distinct discipline with overlapping mechanics — the structure that helps a screen reader is the structure that helps an extractor.