usdigitalmarketingindex.com SEO
Why can field Core Web Vitals measurements disagree with lab testing results?
Updated November 15, 2026
Because they measure different things. A lab test is one page load, on one simulated device and network, from one location, with an empty cache and no real user. Field data is the experience of everyone who visited, on their own devices and connections, over a period of weeks, including what happened after the page loaded. Google describes Core Web Vitals as metrics of “real-world user experience” (Google’s documentation on Core Web Vitals), and field data is the basis for that assessment. Neither measurement is wrong. Each answers its own question, and a disagreement between them is information about what the lab run did not represent.
Real users versus simulated sessions
| Lab | Field | |
|---|---|---|
| Who | A testing tool | Real visitors |
| Device | One, for example an emulated mid-range phone | The full range your audience uses |
| Network | One throttled profile | Every connection your visitors have |
| Location | The test server's | Wherever visitors are |
| Cache | Empty, unless the test is configured otherwise | A mix of first and repeat visits |
| Page state | Logged out, no consent given, default content | Logged in or out, personalized, with consent dialogs and ads |
| Interaction | None, unless scripted | Real taps, scrolls, and typing |
| Result | One number per run | A distribution, summarized at a percentile |
web.dev’s guidance on this subject draws the same line between tools that report data collected in a controlled environment and tools that report data from real users (web.dev on why lab and field data differ).
Metric collection windows
The two also differ in when and for how long they measure.
- A lab run measures the load and stops. Field data covers the whole visit. Layout shifts that happen as a visitor scrolls, and interactions made minutes later, are in the field figure and absent from the lab one.
- Interaction to Next Paint needs interactions. A lab load without scripted input has none, which is why lab tools report a proxy such as total blocking time instead.
- Field data is aggregated over a rolling period of several weeks. A fix deployed today changes the lab result at once and the field figure gradually, as old visits drop out of the window.
- Field figures are reported at a percentile. web.dev’s threshold guidance for layout shift, for example, refers to at least 75 percent of page visits. A lab run gives a single visit.
- Field data from Google’s public dataset exists only for pages and origins with enough traffic. A page with few visits may have lab data only.
Device and network distributions
The lab’s single profile sits somewhere in your audience’s range, and where it sits decides the direction of the disagreement.
An illustrative comparison. The datasets are invented to show how the same page produces different figures.
| Measurement | LCP | Notes |
|---|---|---|
| Lab, emulated mid-range phone, throttled connection | 3.4 s | One cold load |
| Field, 75th percentile, all phones | 2.3 s | Audience mostly on fast phones and connections, many repeat visits |
| Field, 75th percentile, visitors in a region far from the server | 4.1 s | Longer network paths |
| Lab, desktop profile, fast connection | 1.1 s | One cold load |
| Field, 75th percentile, desktop | 1.6 s | Includes older machines |
In the invented data the mobile lab result is worse than the field figure, because the audience is better equipped than the emulated device and benefits from caching. One regional segment is worse than the lab. Both readings describe the same page.
Disagreement runs in either direction.
| Pattern | Likely explanation |
|---|---|
| Lab good, field poor | Real visitors have slower devices or networks than the lab profile; ads, consent tools, or personalization load only in real sessions; shifts or slow interactions occur after load |
| Lab poor, field good | The audience is on fast hardware; repeat visits are cached; the lab throttling is harsher than reality |
| Lab LCP element differs from field | Different viewport sizes, or a dialog or banner that appears only for real users |
Diagnosing conflicting results
Work from the field figure toward a cause.
- Confirm what the field number covers: the single URL or a group of similar pages, the device class, and the period.
- Segment field data if you collect your own: by device, connection type, country, page template, and new against returning visitors. A poor overall figure can come from one segment.
- Adjust the lab test to resemble the poor segment: a slower device profile, a slower network, a location nearer those visitors.
- Reproduce the real page state: accept or leave the consent dialog, log in, allow ads to load.
- For layout shift and responsiveness, script or perform scrolling and interaction during the lab recording.
- Compare the element identified as LCP in each source.
Use each for what it does well.
| Task | Use |
|---|---|
| Deciding whether there is a problem for users | Field |
| Finding the cause | Lab |
| Testing a fix before release | Lab |
| Confirming the fix worked for users | Field |
| Catching regressions in development | Lab, in the build pipeline |
| Pages with too little traffic for field data | Lab, with caution |
Reporting rule: always state which kind of data a number is, the device class, and the date range or test conditions. A sentence that says LCP is 2.3 seconds without those is uninterpretable, and two people quoting different sources will appear to contradict each other.
Collecting your own field data, with a real-user monitoring script that records the metrics and their attribution, closes much of the gap. It gives field measurements you can segment and connect to specific elements, which the public aggregate does not.
Once the discrepancy is understood, the specific bottleneck still has to be found. The diagnostic methods for two of the metrics are set out in the guides to diagnosing poor Largest Contentful Paint on an article template and the bottlenecks behind high Interaction to Next Paint.
The summary: the lab tells you why, the field tells you whether, and a gap between them tells you what your test left out.
Add your comment to this answer