Why Your PageSpeed Score Doesn’t Match Real User Experience

A 95 Lighthouse score means almost nothing if your real users are on cheap phones with weak signal. Understanding the lab vs field split will save you months of confusion.

SP

SAT Team

6 min read

Picture a common situation. A site owner has a Lighthouse score of 95, green across the board, and is genuinely proud of it. Then they open Search Console and it says the same URLs fail Core Web Vitals. "Which one is lying?"

Neither. They're measuring completely different things, and once you understand the split, a lot of PageSpeed confusion evaporates.

Lab data: a simulation of one visit

When you run Lighthouse (or the bottom half of a PageSpeed Insights report), a machine loads your page once, under fixed conditions: a simulated mid-range device, a throttled connection, no cache, no logged-in state, nobody actually clicking anything. It's a controlled experiment. Same inputs, roughly same outputs.

That makes lab data brilliant for debugging. Change something, re-run, compare. It's your before-and-after tool.

But it's one synthetic visit. It knows nothing about your actual audience.

Field data: what your real visitors experienced

The top section of a PageSpeed Insights report comes from the Chrome User Experience Report, usually called CrUX. Chrome collects anonymized performance timings from real users who opted into syncing, aggregates 28 days of them, and reports the 75th percentile.

Read that again: 28 days, 75th percentile, real people. If your audience skews toward budget Android phones on mobile data in areas with patchy coverage, your field numbers will reflect that no matter how fast your page runs on a MacBook over fiber.

And here's the part that matters for SEO: Google's ranking systems use field data. The Lighthouse performance score, that satisfying 0-100 number, is not a ranking factor. It never was. Search Console's Core Web Vitals report is built entirely on CrUX.

Why the two disagree so often

A few patterns explain most mismatches:

Your users are slower than the lab. The lab simulates a decent 4G connection. Plenty of real visits happen on worse. If half your traffic comes from older phones, your field INP and LCP will run high while lab numbers look fine.

The lab never interacts. INP needs real clicks and taps to measure. A default lab run can't produce it, which is why the lab section shows Total Blocking Time instead. Your page can score 95 and still have a terrible INP because that laggy mega-menu only reveals itself when a human uses it.

Caching and CDNs help repeat visitors, not the lab. Lab runs are always cold loads. If most of your audience are returning visitors hitting warm caches, your field data may actually be better than the lab suggests. It cuts both ways.

The 28-day window lags. You shipped a fix on Tuesday; Search Console still shows failures three weeks later. Normal. The window has to roll over with new data. This lag causes more unnecessary panic than any other quirk of the system.

Different pages, different visitors. Field data at the origin level blends your whole site. Your homepage might be fast while a template used by thousands of tag pages drags the aggregate down.

So which should you optimize for?

Both, but with different jobs:

  • Field data is the scoreboard. It's what users feel and what Google counts. Your goals live here: LCP under 2.5s, INP under 200ms, CLS under 0.1, all at the 75th percentile.
  • Lab data is the workshop. It's where you diagnose, experiment, and verify changes instantly instead of waiting a month.

The workflow that actually works: read the field data to pick your battle (say, mobile LCP is at 3.4s). Reproduce the problem in the lab with realistic throttling. Fix it, confirm the lab improvement, ship. Then watch field data trend down over the following weeks and hit "Validate fix" in Search Console when it crosses the threshold.

A note on chasing 100

There's a small industry of tricks for inflating Lighthouse scores: delaying scripts until interaction so the lab never sees them, hiding content from the crawler, that sort of thing. Some of these genuinely help users. Others just game a simulation while real visitors get the same slow, janky experience, and your field data (the part that ranks) doesn't move an inch.

A perfect 100 makes a nice screenshot. A field LCP that dropped from 4.1s to 2.2s makes you money. If you ever have to choose where to spend a week, choose the second one.

Questions that come up every time

Why does my page have no field data at all? Not enough real Chrome traffic on that URL over the past 28 days. CrUX needs a minimum sample before it publishes numbers, and it doesn't disclose the exact threshold. Low-traffic pages fall back to origin-wide data, which blends every page on your site. If even the origin has no data, your site is simply too small for field measurement yet, and lab data plus your own real-user monitoring are all you have. That's not a problem to fix; it's just how the system works.

My score changes every time I run Lighthouse. Is the tool broken? No, variance is baked in. Network conditions, server load, CPU contention on the testing machine, third-party scripts behaving differently between runs: all of it moves the number a few points. Run three tests and look at the median. Treat any single-digit score difference as noise, because it is.

Can I have good field data and a bad lab score? Absolutely, and it's more common than you'd think. If your audience is largely returning visitors on fast connections hitting warm caches, real experiences can be great while the cold, throttled lab run looks mediocre. When the two disagree, believe the field data. It's measuring reality; the lab is measuring a worst-case rehearsal.

How long until fixes show in Search Console? The full 28-day window has to roll over, so meaningful movement takes two to four weeks and full stabilization about a month. Ship, verify in the lab, then be patient on purpose.

Quick reference

When someone sends you a PageSpeed report, check these in order: First, does the page have field data at all? Low-traffic pages often don't, and then origin-level data is shown instead. Second, are the field Core Web Vitals green? That's the pass/fail that matters. Third, only then look at lab diagnostics for clues about why something fails.

Get comfortable with that reading order and you'll never again waste a weekend chasing five Lighthouse points that no user, and no ranking system, would ever notice.

Check your own site in 30 seconds

Run a free SEO audit and see exactly where your pages stand on the issues covered in this article.

Run Free SEO Audit

Keep reading