web-vitals

• Overview
• Install and load the library
  • From npm
  • From a CDN
• Usage
  • Basic usage
  • Report the value on every change
  • Report only the delta of changes
  • Send the results to an analytics endpoint
  • Send the results to Google Analytics
  • Send the results to Google Tag Manager
  • Send attribution data
  • Batch multiple reports together
• Build options
  • Which build is right for you?
• API
  • Types
  • Functions
  • Rating Thresholds
  • Attribution
• Browser Support
• Limitations
• Development
• Integrations
• License

Overview

The web-vitals library is a tiny (~2K, brotli'd), modular library for measuring all the Web Vitals metrics on real users, in a way that accurately matches how they're measured by Chrome and reported to other Google tools (e.g. Chrome User Experience Report, Page Speed Insights, Search Console's Speed Report).

The library supports all of the Core Web Vitals as well as a number of other metrics that are useful in diagnosing real-user performance issues.

Core Web Vitals

• Cumulative Layout Shift (CLS)
• Interaction to Next Paint (INP)
• Largest Contentful Paint (LCP)

Other metrics

• First Contentful Paint (FCP)
• Time to First Byte (TTFB)

Install and load the library

The web-vitals library uses the buffered flag for PerformanceObserver, allowing it to access performance entries that occurred before the library was loaded.

This means you do not need to load this library early in order to get accurate performance data. In general, this library should be deferred until after other user-impacting code has loaded.

From npm

You can install this library from npm by running:

1npm install web-vitals

[!NOTE] If you're not using npm, you can still load web-vitals via <script> tags from a CDN like unpkg.com. See the load web-vitals from a CDN usage example below for details.

There are a few different builds of the web-vitals library, and how you load the library depends on which build you want to use.

For details on the difference between the builds, see which build is right for you.

1. The "standard" build

To load the "standard" build, import modules from the web-vitals package in your application code (as you would with any npm package and node-based build tool):

1import {onLCP, onINP, onCLS} from 'web-vitals';

2. The "attribution" build

Measuring the Web Vitals scores for your real users is a great first step toward optimizing the user experience. But if your scores aren't good, the next step is to understand why they're not good and work to improve them.

The "attribution" build helps you do that by including additional diagnostic information with each metric to help you identify the root cause of poor performance as well as prioritize the most important things to fix.

The "attribution" build is slightly larger than the "standard" build (by about 1.5K, brotli'd), so while the code size is still small, it's only recommended if you're actually using these features.

To load the "attribution" build, change any import statements that reference web-vitals to web-vitals/attribution:

1import {onLCP, onINP, onCLS} from 'web-vitals';
2import {onLCP, onINP, onCLS} from 'web-vitals/attribution';

Usage for each of the imported function is identical to the standard build, but when importing from the attribution build, the metric objects will contain an additional attribution property.

See Send attribution data for usage examples, and the attribution reference for details on what values are added for each metric.

From a CDN

The recommended way to use the web-vitals package is to install it from npm and integrate it into your build process. However, if you're not using npm, it's still possible to use web-vitals by requesting it from a CDN that serves npm package files.

The following examples show how to load web-vitals from unpkg.com. It is also possible to load this from jsDelivr, and cdnjs.

Important! The unpkg.com, jsDelivr, and cdnjs CDNs are shown here for example purposes only. unpkg.com, jsDelivr, and cdnjs are not affiliated with Google, and there are no guarantees that loading the library from those CDNs will continue to work in the future. Self-hosting the built files rather than loading from the CDN is better for security, reliability, and performance reasons.

Load the "standard" build (using a module script)

1<!-- Append the `?module` param to load the module version of `web-vitals` -->
2<script type="module">
3  import {onCLS, onINP, onLCP} from 'https://unpkg.com/web-vitals@5?module';
4
5  onCLS(console.log);
6  onINP(console.log);
7  onLCP(console.log);
8</script>

Note: When the web-vitals code is isolated from the application code in this way, there is less need to depend on dynamic imports so this code uses a regular import line.

Load the "standard" build (using a classic script)

1<script>
2  (function () {
3    var script = document.createElement('script');
4    script.src = 'https://unpkg.com/web-vitals@5/dist/web-vitals.iife.js';
5    script.onload = function () {
6      // When loading `web-vitals` using a classic script, all the public
7      // methods can be found on the `webVitals` global namespace.
8      webVitals.onCLS(console.log);
9      webVitals.onINP(console.log);
10      webVitals.onLCP(console.log);
11    };
12    document.head.appendChild(script);
13  })();
14</script>

Load the "attribution" build (using a module script)

1<!-- Append the `?module` param to load the module version of `web-vitals` -->
2<script type="module">
3  import {
4    onCLS,
5    onINP,
6    onLCP,
7  } from 'https://unpkg.com/web-vitals@5/dist/web-vitals.attribution.js?module';
8
9  onCLS(console.log);
10  onINP(console.log);
11  onLCP(console.log);
12</script>

Load the "attribution" build (using a classic script)

1<script>
2  (function () {
3    var script = document.createElement('script');
4    script.src =
5      'https://unpkg.com/web-vitals@5/dist/web-vitals.attribution.iife.js';
6    script.onload = function () {
7      // When loading `web-vitals` using a classic script, all the public
8      // methods can be found on the `webVitals` global namespace.
9      webVitals.onCLS(console.log);
10      webVitals.onINP(console.log);
11      webVitals.onLCP(console.log);
12    };
13    document.head.appendChild(script);
14  })();
15</script>

Usage

Basic usage

Each of the Web Vitals metrics is exposed as a single function that takes a callback function that will be called any time the metric value is available and ready to be reported.

The following example measures each of the Core Web Vitals metrics and logs the result to the console once its value is ready to report.

(The examples below import the "standard" build, but they will work with the "attribution" build as well.)

1import {onCLS, onINP, onLCP} from 'web-vitals';
2
3onCLS(console.log);
4onINP(console.log);
5onLCP(console.log);

Note that some of these metrics will not report until the user has interacted with the page, switched tabs, or the page starts to unload. If you don't see the values logged to the console immediately, try reloading the page (with preserve log enabled) or switching tabs and then switching back.

Also, in some cases a metric callback may never be called:

• INP is not reported if the user never interacts with the page.
• CLS, FCP, and LCP are not reported if the page was loaded in the background.

In other cases, a metric callback may be called more than once:

• CLS and INP should be reported any time the page's visibilityState changes to hidden.
• All metrics are reported again (with the above exceptions) after a page is restored from the back/forward cache.

[!WARNING] Do not call any of the Web Vitals functions (e.g. onCLS(), onINP(), onLCP()) more than once per page load. Each of these functions creates a PerformanceObserver instance and registers event listeners for the lifetime of the page. While the overhead of calling these functions once is negligible, calling them repeatedly on the same page may eventually result in a memory leak.

Report the value on every change

In most cases, you only want the callback function to be called when the metric is ready to be reported. However, it is possible to report every change (e.g. each larger layout shift as it happens) by setting reportAllChanges to true in the optional, configuration object (second parameter).

[!IMPORTANT] > reportAllChanges only reports when the metric changes, not for each input to the metric. For example, a new layout shift that does not increase the CLS metric will not be reported even with reportAllChanges set to true because the CLS metric has not changed. Similarly, for INP, each interaction is not reported even with reportAllChanges set to true—just when an interaction causes an increase to INP.

This can be useful when debugging, but in general using reportAllChanges is not needed (or recommended) for measuring these metrics in production.

1import {onCLS} from 'web-vitals';
2
3// Logs CLS as the value changes.
4onCLS(console.log, {reportAllChanges: true});

Report only the delta of changes

Some analytics providers allow you to update the value of a metric, even after you've already sent it to their servers (overwriting the previously-sent value with the same id).

Other analytics providers, however, do not allow this, so instead of reporting the new value, you need to report only the delta (the difference between the current value and the last-reported value). You can then compute the total value by summing all metric deltas sent with the same ID.

The following example shows how to use the id and delta properties:

1import {onCLS, onINP, onLCP} from 'web-vitals';
2
3function logDelta({name, id, delta}) {
4  console.log(`${name} matching ID ${id} changed by ${delta}`);
5}
6
7onCLS(logDelta);
8onINP(logDelta);
9onLCP(logDelta);

[!NOTE] The first time the callback function is called, its value and delta properties will be the same.

In addition to using the id field to group multiple deltas for the same metric, it can also be used to differentiate different metrics reported on the same page. For example, after a back/forward cache restore, a new metric object is created with a new id (since back/forward cache restores are considered separate page visits).

Send the results to an analytics endpoint

The following example measures each of the Core Web Vitals metrics and reports them to a hypothetical /analytics endpoint, as soon as each is ready to be sent.

The sendToAnalytics() function uses the navigator.sendBeacon() method, which is widely available across browsers, and supports sending data as the page is being unloaded.

1import {onCLS, onINP, onLCP} from 'web-vitals';
2
3function sendToAnalytics(metric) {
4  const body = JSON.stringify({
5    name: metric.name,
6    value: metric.value,
7    id: metric.id,
8
9    // Include additional data as needed...
10  });
11
12  // Use `navigator.sendBeacon()` to send the data, which supports
13  // sending while the page is unloading.
14  navigator.sendBeacon('/analytics', body);
15}
16
17onCLS(sendToAnalytics);
18onINP(sendToAnalytics);
19onLCP(sendToAnalytics);

Send the results to Google Analytics

Google Analytics does not support reporting metric distributions in any of its built-in reports; however, if you set a unique event parameter value (in this case, the metric_id, as shown in the example below) on every metric instance that you send to Google Analytics, you can create a report yourself by first getting the data via the Google Analytics Data API or via BigQuery export and then visualizing it any charting library you choose.

Google Analytics 4 introduces a new Event model allowing custom parameters instead of a fixed category, action, and label. It also supports non-integer values, making it easier to measure Web Vitals metrics compared to previous versions.

1import {onCLS, onINP, onLCP} from 'web-vitals';
2
3function sendToGoogleAnalytics({name, delta, value, id}) {
4  // Assumes the global `gtag()` function exists, see:
5  // https://developers.google.com/analytics/devguides/collection/ga4
6  gtag('event', name, {
7    // Built-in params:
8    value: delta, // Use `delta` so the value can be summed.
9    // Custom params:
10    metric_id: id, // Needed to aggregate events.
11    metric_value: value, // Optional.
12    metric_delta: delta, // Optional.
13
14    // OPTIONAL: any additional params or debug info here.
15    // See: https://web.dev/articles/debug-performance-in-the-field
16    // metric_rating: 'good' | 'needs-improvement' | 'poor',
17    // debug_info: '...',
18    // ...
19  });
20}
21
22onCLS(sendToGoogleAnalytics);
23onINP(sendToGoogleAnalytics);
24onLCP(sendToGoogleAnalytics);

For details on how to query this data in BigQuery, or visualise it in Looker Studio, see Measure and debug performance with Google Analytics 4 and BigQuery.

Send the results to Google Tag Manager

While web-vitals can be called directly from Google Tag Manager, using a pre-defined custom template makes this considerably easier. Some recommended templates include:

• Core Web Vitals by Simo Ahava. See Track Core Web Vitals in GA4 with Google Tag Manager for usage and installation instructions.
• Web Vitals Template for Google Tag Manager by The Google Marketing Solutions team. See the README for usage and installation instructions.

Send attribution data

When using the attribution build, you can send additional data to help you debug why the metric values are they way they are.

This example sends an additional debug_target param to Google Analytics, corresponding to the element most associated with each metric.

1import {onCLS, onINP, onLCP} from 'web-vitals/attribution';
2
3function sendToGoogleAnalytics({name, delta, value, id, attribution}) {
4  const eventParams = {
5    // Built-in params:
6    value: delta, // Use `delta` so the value can be summed.
7    // Custom params:
8    metric_id: id, // Needed to aggregate events.
9    metric_value: value, // Optional.
10    metric_delta: delta, // Optional.
11  };
12
13  switch (name) {
14    case 'CLS':
15      eventParams.debug_target = attribution.largestShiftTarget;
16      break;
17    case 'INP':
18      eventParams.debug_target = attribution.interactionTarget;
19      break;
20    case 'LCP':
21      eventParams.debug_target = attribution.element;
22      break;
23  }
24
25  // Assumes the global `gtag()` function exists, see:
26  // https://developers.google.com/analytics/devguides/collection/ga4
27  gtag('event', name, eventParams);
28}
29
30onCLS(sendToGoogleAnalytics);
31onINP(sendToGoogleAnalytics);
32onLCP(sendToGoogleAnalytics);

[!NOTE] This example relies on custom event parameters in Google Analytics 4.

See Debug performance in the field for more information and examples.

Batch multiple reports together

Rather than reporting each individual Web Vitals metric separately, you can minimize your network usage by batching multiple metric reports together in a single network request.

However, since not all Web Vitals metrics become available at the same time, and since not all metrics are reported on every page, you cannot simply defer reporting until all metrics are available.

Instead, you should keep a queue of all metrics that were reported and flush the queue whenever the page is backgrounded or unloaded:

1import {onCLS, onINP, onLCP} from 'web-vitals';
2
3const queue = new Set();
4function addToQueue(metric) {
5  queue.add(metric);
6}
7
8function flushQueue() {
9  if (queue.size > 0) {
10    // Replace with whatever serialization method you prefer.
11    // Note: JSON.stringify will likely include more data than you need.
12    const body = JSON.stringify([...queue]);
13
14    // Use `navigator.sendBeacon()` to send the data, which supports
15    // sending while the page is unloading.
16    navigator.sendBeacon('/analytics', body);
17
18    queue.clear();
19  }
20}
21
22onCLS(addToQueue);
23onINP(addToQueue);
24onLCP(addToQueue);
25
26// Report all available metrics whenever the page is backgrounded or unloaded.
27addEventListener('visibilitychange', () => {
28  if (document.visibilityState === 'hidden') {
29    flushQueue();
30  }
31});

[!NOTE] See the Page Lifecycle guide for an explanation of why visibilitychange is recommended over events like beforeunload and unload.

Build options

The web-vitals package includes both "standard" and "attribution" builds, as well as different formats of each to allow developers to choose the format that best meets their needs or integrates with their architecture.

The following table lists all the builds distributed with the web-vitals package on npm.

Filename (all within dist/*)	Export	Description
web-vitals.js	pkg.module	An ES module bundle of all metric functions, without any attribution features. This is the "standard" build and is the simplest way to consume this library out of the box.
web-vitals.umd.cjs	pkg.main	A UMD version of the web-vitals.js bundle (exposed on the self.webVitals.* namespace).
web-vitals.iife.js	--	An IIFE version of the web-vitals.js bundle (exposed on the self.webVitals.* namespace).
web-vitals.attribution.js	--	An ES module version of all metric functions that includes attribution features.
web-vitals.attribution.umd.cjs	--	A UMD version of the web-vitals.attribution.js build (exposed on the self.webVitals.* namespace).
web-vitals.attribution.iife.js	--	An IIFE version of the web-vitals.attribution.js build (exposed on the self.webVitals.* namespace).

Which build is right for you?

Most developers will generally want to use "standard" build (via either the ES module or UMD version, depending on your bundler/build system), as it's the easiest to use out of the box and integrate into existing tools.

However, if you'd lke to collect additional debug information to help you diagnose performance bottlenecks based on real-user issues, use the "attribution" build.

For guidance on how to collect and use real-user data to debug performance issues, see Debug performance in the field.

API

Types:

Metric

All metrics types inherit from the following base interface:

1interface Metric {
2  /**
3   * The name of the metric (in acronym form).
4   */
5  name: 'CLS' | 'FCP' | 'INP' | 'LCP' | 'TTFB';
6
7  /**
8   * The current value of the metric.
9   */
10  value: number;
11
12  /**
13   * The rating as to whether the metric value is within the "good",
14   * "needs improvement", or "poor" thresholds of the metric.
15   */
16  rating: 'good' | 'needs-improvement' | 'poor';
17
18  /**
19   * The delta between the current value and the last-reported value.
20   * On the first report, `delta` and `value` will always be the same.
21   */
22  delta: number;
23
24  /**
25   * A unique ID representing this particular metric instance. This ID can
26   * be used by an analytics tool to dedupe multiple values sent for the same
27   * metric instance, or to group multiple deltas together and calculate a
28   * total. It can also be used to differentiate multiple different metric
29   * instances sent from the same page, which can happen if the page is
30   * restored from the back/forward cache (in that case new metrics object
31   * get created).
32   */
33  id: string;
34
35  /**
36   * Any performance entries relevant to the metric value calculation.
37   * The array may also be empty if the metric value was not based on any
38   * entries (e.g. a CLS value of 0 given no layout shifts).
39   */
40  entries: PerformanceEntry[];
41
42  /**
43   * The type of navigation.
44   *
45   * This will be the value returned by the Navigation Timing API (or
46   * `undefined` if the browser doesn't support that API), with the following
47   * exceptions:
48   * - 'back-forward-cache': for pages that are restored from the bfcache.
49   * - 'back_forward' is renamed to 'back-forward' for consistency.
50   * - 'prerender': for pages that were prerendered.
51   * - 'restore': for pages that were discarded by the browser and then
52   * restored by the user.
53   */
54  navigationType:
55    | 'navigate'
56    | 'reload'
57    | 'back-forward'
58    | 'back-forward-cache'
59    | 'prerender'
60    | 'restore';
61}

Metric-specific subclasses:

CLSMetric

1interface CLSMetric extends Metric {
2  name: 'CLS';
3  entries: LayoutShift[];
4}

FCPMetric

1interface FCPMetric extends Metric {
2  name: 'FCP';
3  entries: PerformancePaintTiming[];
4}

INPMetric

1interface INPMetric extends Metric {
2  name: 'INP';
3  entries: PerformanceEventTiming[];
4}

LCPMetric

1interface LCPMetric extends Metric {
2  name: 'LCP';
3  entries: LargestContentfulPaint[];
4}

TTFBMetric

1interface TTFBMetric extends Metric {
2  name: 'TTFB';
3  entries: PerformanceNavigationTiming[];
4}

MetricRatingThresholds

The thresholds of metric's "good", "needs improvement", and "poor" ratings.

• Metric values up to and including [0] are rated "good"
• Metric values up to and including [1] are rated "needs improvement"
• Metric values above [1] are "poor"

Metric value	Rating
≦ [0]	"good"
> [0] and ≦ [1]	"needs improvement"
> [1]	"poor"

1type MetricRatingThresholds = [number, number];

See also Rating Thresholds.

ReportOpts

1interface ReportOpts {
2  reportAllChanges?: boolean;
3}

Metric-specific subclasses:

INPReportOpts

1interface INPReportOpts extends ReportOpts {
2  durationThreshold?: number;
3}

AttributionReportOpts

A subclass of ReportOpts used for each metric function exported in the attribution build.

1interface AttributionReportOpts extends ReportOpts {
2  generateTarget?: (el: Node | null) => string | null | undefined;
3}

Metric-specific subclasses:

INPAttributionReportOpts

1interface INPAttributionReportOpts extends AttributionReportOpts {
2  durationThreshold?: number;
3}

LoadState

The LoadState type is used in several of the metric attribution objects.

1/**
2 * The loading state of the document. Note: this value is similar to
3 * `document.readyState` but it subdivides the "interactive" state into the
4 * time before and after the DOMContentLoaded event fires.
5 *
6 * State descriptions:
7 * - `loading`: the initial document response has not yet been fully downloaded
8 *   and parsed. This is equivalent to the corresponding `readyState` value.
9 * - `dom-interactive`: the document has been fully loaded and parsed, but
10 *   scripts may not have yet finished loading and executing.
11 * - `dom-content-loaded`: the document is fully loaded and parsed, and all
12 *   scripts (except `async` scripts) have loaded and finished executing.
13 * - `complete`: the document and all of its sub-resources have finished
14 *   loading. This is equivalent to the corresponding `readyState` value.
15 */
16type LoadState =
17  | 'loading'
18  | 'dom-interactive'
19  | 'dom-content-loaded'
20  | 'complete';

Functions:

onCLS()

1function onCLS(callback: (metric: CLSMetric) => void, opts?: ReportOpts): void;

Calculates the CLS value for the current page and calls the callback function once the value is ready to be reported, along with all layout-shift performance entries that were used in the metric value calculation. The reported value is a double (corresponding to a layout shift score).

[!IMPORTANT] CLS should be continually monitored for changes throughout the entire lifespan of a page—including if the user returns to the page after it's been hidden/backgrounded. However, since browsers often will not fire additional callbacks once the user has backgrounded a page, callback is always called when the page's visibility state changes to hidden. As a result, the callback function might be called multiple times during the same page load (see Reporting only the delta of changes for how to manage this).

If the reportAllChanges configuration option is set to true, the callback function will be called as soon as the value is initially determined as well as any time the value changes throughout the page lifespan (though not necessarily for every layout shift). Note that regardless of whether reportAllChanges is used, the final reported value will be the same.

onFCP()

1function onFCP(callback: (metric: FCPMetric) => void, opts?: ReportOpts): void;

Calculates the FCP value for the current page and calls the callback function once the value is ready, along with the relevant paint performance entry used to determine the value. The reported value is a DOMHighResTimeStamp.

onINP()

1function onINP(
2  callback: (metric: INPMetric) => void,
3  opts?: INPReportOpts,
4): void;

Calculates the INP value for the current page and calls the callback function once the value is ready, along with the event performance entries reported for that interaction. The reported value is a DOMHighResTimeStamp.

[!IMPORTANT] INP should be continually monitored for changes throughout the entire lifespan of a page—including if the user returns to the page after it's been hidden/backgrounded. However, since browsers often will not fire additional callbacks once the user has backgrounded a page, callback is always called when the page's visibility state changes to hidden. As a result, the callback function might be called multiple times during the same page load (see Reporting only the delta of changes for how to manage this).

A custom durationThreshold configuration option can optionally be passed to control the minimum duration filter for event-timing. Events which are faster than this threshold are not reported. Note that the first-input entry is always observed, regardless of duration, to ensure you always have some INP score. The default threshold, after the library is initialized, is 40 milliseconds (the event-timing default of 104 milliseconds applies to all events emitted before the library is initialised). This default threshold of 40 is chosen to strike a balance between usefulness and performance. Running this callback for any interaction that spans just one or two frames is likely not worth the insight that could be gained.

If the reportAllChanges configuration option is set to true, the callback function will be called as soon as the value is initially determined as well as any time the value changes throughout the page lifespan (though not necessarily for every interaction). Note that regardless of whether reportAllChanges is used, the final reported value will be the same.

onLCP()

1function onLCP(callback: (metric: LCPMetric) => void, opts?: ReportOpts): void;

Calculates the LCP value for the current page and calls the callback function once the value is ready (along with the relevant largest-contentful-paint performance entry used to determine the value). The reported value is a DOMHighResTimeStamp.

If the reportAllChanges configuration option is set to true, the callback function will be called any time a new largest-contentful-paint performance entry is dispatched, or once the final value of the metric has been determined. Note that regardless of whether reportAllChanges is used, the final reported value will be the same.

onTTFB()

1function onTTFB(
2  callback: (metric: TTFBMetric) => void,
3  opts?: ReportOpts,
4): void;

Calculates the TTFB value for the current page and calls the callback function once the page has loaded, along with the relevant navigation performance entry used to determine the value. The reported value is a DOMHighResTimeStamp.

Note, this function waits until after the page is loaded to call callback in order to ensure all properties of the navigation entry are populated. This is useful if you want to report on other metrics exposed by the Navigation Timing API.

For example, the TTFB metric starts from the page's time origin, which means it includes time spent on DNS lookup, connection negotiation, network latency, and server processing time.

1import {onTTFB} from 'web-vitals';
2
3onTTFB((metric) => {
4  // Calculate the request time by subtracting from TTFB
5  // everything that happened prior to the request starting.
6  const requestTime = metric.value - metric.entries[0].requestStart;
7  console.log('Request time:', requestTime);
8});

[!NOTE] Browsers that do not support navigation entries will fall back to using performance.timing (with the timestamps converted from epoch time to DOMHighResTimeStamp). This ensures code referencing these values (like in the example above) will work the same in all browsers.

Rating Thresholds:

The thresholds of each metric's "good", "needs improvement", and "poor" ratings are available as MetricRatingThresholds.

Example:

1import {CLSThresholds, INPThresholds, LCPThresholds} from 'web-vitals';
2
3console.log(CLSThresholds); // [ 0.1, 0.25 ]
4console.log(INPThresholds); // [ 200, 500 ]
5console.log(LCPThresholds); // [ 2500, 4000 ]

[!NOTE] It's typically not necessary (or recommended) to manually calculate metric value ratings using these thresholds. Use the Metric['rating'] instead.

Attribution:

In the attribution build each of the metric functions has two primary differences from their standard build counterparts:

1. Their callback is invoked with a MetricWithAttribution objects instead of a Metric object. Each MetricWithAttribution extends the Metric object and adds an additional attribution object, which contains potentially-helpful debugging information that can be sent along with the metric values for the current page visit in order to help identify issues happening to real-users in the field.

2. They accept an AttributionReportOpts objects instead of a ReportOpts object. The AttributionReportOpts object supports an additional, optional, generateTarget() function that lets developers customize how DOM elements are stringified for reporting purposes. When passed, the return value of the generateTarget() function will be used for any "target" properties in the following attribution objects: CLSAttribution, INPAttribution, and LCPAttribution. If null or undefined is returned by the generateTarget() function, or no function is given, then the default selector function will be used.

   1interface AttributionReportOpts extends ReportOpts {
   2  generateTarget?: (el: Node | null) => string | null | undefined;
   3}

   For example, if a web page has unique data-name attibute on many elements, you may prefer to use those over the built-in selector-style strings that are generated by default.

   1function customGenerateTarget(el) {
   2  if (el.dataset.name) {
   3    return el.dataset.name;
   4  }
   5
   6  // Otherwise use default selector function
   7}
   8
   9onLCP(sendToAnalytics, {generateTarget: customGenerateTarget});

The next sections document the shape of the attribution object for each of the metrics:

CLSAttribution

1interface CLSAttribution {
2  /**
3   * By default, a selector identifying the first element (in document order)
4   * that shifted when the single largest layout shift that contributed to the
5   * page's CLS score occurred. If the `generateTarget` configuration option
6   * was passed, then this will instead be the return value of that function,
7   * falling back to the default if that returns null or undefined.
8   */
9  largestShiftTarget?: string;
10  /**
11   * The time when the single largest layout shift contributing to the page's
12   * CLS score occurred.
13   */
14  largestShiftTime?: DOMHighResTimeStamp;
15  /**
16   * The layout shift score of the single largest layout shift contributing to
17   * the page's CLS score.
18   */
19  largestShiftValue?: number;
20  /**
21   * The `LayoutShiftEntry` representing the single largest layout shift
22   * contributing to the page's CLS score. (Useful when you need more than just
23   * `largestShiftTarget`, `largestShiftTime`, and `largestShiftValue`).
24   */
25  largestShiftEntry?: LayoutShift;
26  /**
27   * The first element source (in document order) among the `sources` list
28   * of the `largestShiftEntry` object. (Also useful when you need more than
29   * just `largestShiftTarget`, `largestShiftTime`, and `largestShiftValue`).
30   */
31  largestShiftSource?: LayoutShiftAttribution;
32  /**
33   * The loading state of the document at the time when the largest layout
34   * shift contribution to the page's CLS score occurred (see `LoadState`
35   * for details).
36   */
37  loadState?: LoadState;
38}

FCPAttribution

1interface FCPAttribution {
2  /**
3   * The time from when the user initiates loading the page until when the
4   * browser receives the first byte of the response (a.k.a. TTFB).
5   */
6  timeToFirstByte: number;
7  /**
8   * The delta between TTFB and the first contentful paint (FCP).
9   */
10  firstByteToFCP: number;
11  /**
12   * The loading state of the document at the time when FCP `occurred (see
13   * `LoadState` for details). Ideally, documents can paint before they finish
14   * loading (e.g. the `loading` or `dom-interactive` phases).
15   */
16  loadState: LoadState;
17  /**
18   * The `PerformancePaintTiming` entry corresponding to FCP.
19   */
20  fcpEntry?: PerformancePaintTiming;
21  /**
22   * The `navigation` entry of the current page, which is useful for diagnosing
23   * general page load issues. This can be used to access `serverTiming` for example:
24   * navigationEntry?.serverTiming
25   */
26  navigationEntry?: PerformanceNavigationTiming;
27}

INPAttribution

1interface INPAttribution {
2  /**
3   * By default, a selector identifying the element that the user first
4   * interacted with as part of the frame where the INP candidate interaction
5   * occurred. If this value is an empty string, that generally means the
6   * element was removed from the DOM after the interaction. If the
7   * `generateTarget` configuration option was passed, then this will instead
8   * be the return value of that function, falling back to the default if that
9   * returns null or undefined.
10   */
11  interactionTarget: string;
12  /**
13   * The time when the user first interacted during the frame where the INP
14   * candidate interaction occurred (if more than one interaction occurred
15   * within the frame, only the first time is reported).
16   */
17  interactionTime: DOMHighResTimeStamp;
18  /**
19   * The type of interaction, based on the event type of the `event` entry
20   * that corresponds to the interaction (i.e. the first `event` entry
21   * containing an `interactionId` dispatched in a given animation frame).
22   * For "pointerdown", "pointerup", or "click" events this will be "pointer",
23   * and for "keydown" or "keyup" events this will be "keyboard".
24   */
25  interactionType: 'pointer' | 'keyboard';
26  /**
27   * The best-guess timestamp of the next paint after the interaction.
28   * In general, this timestamp is the same as the `startTime + duration` of
29   * the event timing entry. However, since duration values are rounded to the
30   * nearest 8ms (and can be rounded down), this value is clamped to always be
31   * reported after the processing times.
32   */
33  nextPaintTime: DOMHighResTimeStamp;
34  /**
35   * An array of Event Timing entries that were processed within the same
36   * animation frame as the INP candidate interaction.
37   */
38  processedEventEntries: PerformanceEventTiming[];
39  /**
40   * The time from when the user interacted with the page until when the
41   * browser was first able to start processing event listeners for that
42   * interaction. This time captures the delay before event processing can
43   * begin due to the main thread being busy with other work.
44   */
45  inputDelay: number;
46  /**
47   * The time from when the first event listener started running in response to
48   * the user interaction until when all event listener processing has finished.
49   */
50  processingDuration: number;
51  /**
52   * The time from when the browser finished processing all event listeners for
53   * the user interaction until the next frame is presented on the screen and
54   * visible to the user. This time includes work on the main thread (such as
55   * `requestAnimationFrame()` callbacks, `ResizeObserver` and
56   * `IntersectionObserver` callbacks, and style/layout calculation) as well
57   * as off-main-thread work (such as compositor, GPU, and raster work).
58   */
59  presentationDelay: number;
60  /**
61   * The loading state of the document at the time when the interaction
62   * corresponding to INP occurred (see `LoadState` for details). If the
63   * interaction occurred while the document was loading and executing script
64   * (e.g. usually in the `dom-interactive` phase) it can result in long delays.
65   */
66  loadState: LoadState;
67  /**
68   * If the browser supports the Long Animation Frame API, this array will
69   * include any `long-animation-frame` entries that intersect with the INP
70   * candidate interaction's `startTime` and the `processingEnd` time of the
71   * last event processed within that animation frame. If the browser does not
72   * support the Long Animation Frame API or no `long-animation-frame` entries
73   * are detected, this array will be empty.
74   */
75  longAnimationFrameEntries: PerformanceLongAnimationFrameTiming[];
76  /**
77   * Summary information about the longest script entry intersecting the INP
78   * duration. Note, only script entries above 5 milliseconds are reported by
79   * the Long Animation Frame API.
80   */
81  longestScript?: INPLongestScriptSummary;
82  /**
83   * The total duration of Long Animation Frame scripts that intersect the INP
84   * duration excluding any forced style and layout (that is included in
85   * totalStyleAndLayout). Note, this is limited to scripts > 5 milliseconds.
86   */
87  totalScriptDuration?: number;
88  /**
89   * The total style and layout duration from any Long Animation Frames
90   * intersecting the INP interaction. This includes any end-of-frame style and
91   * layout duration + any forced style and layout duration.
92   */
93  totalStyleAndLayoutDuration?: number;
94  /**
95   * The off main-thread presentation delay from the end of the last Long
96   * Animation Frame (where available) until the INP end point.
97   */
98  totalPaintDuration?: number;
99  /**
100   * The total unattributed time not included in any of the previous totals.
101   * This includes scripts < 5 milliseconds and other timings not attributed
102   * by Long Animation Frame (including when a frame is < 50ms and so has no
103   * Long Animation Frame).
104   * When no Long Animation Frames are present this will be undefined, rather
105   * than everything being unattributed to make it clearer when it's expected
106   * to be small.
107   */
108  totalUnattributedDuration?: number;
109}

INPLongestScriptSummary

1interface INPLongestScriptSummary {
2  /**
3   * The longest Long Animation Frame script entry that intersects the INP
4   * interaction.
5   */
6  entry: PerformanceScriptTiming;
7  /**
8   * The INP subpart where the longest script ran.
9   */
10  subpart: 'input-delay' | 'processing-duration' | 'presentation-delay';
11  /**
12   * The amount of time the longest script intersected the INP duration.
13   */
14  intersectingDuration: number;
15}

LCPAttribution

1interface LCPAttribution {
2  /**
3   * By default, a selector identifying the element corresponding to the
4   * largest contentful paint for the page. If the `generateTarget`
5   * configuration option was passed, then this will instead be the return
6   * value of that function, falling back to the default if that returns null
7   * or undefined.
8   */
9  target?: string;
10  /**
11   * The URL (if applicable) of the LCP image resource. If the LCP element
12   * is a text node, this value will not be set.
13   */
14  url?: string;
15  /**
16   * The time from when the user initiates loading the page until when the
17   * browser receives the first byte of the response (a.k.a. TTFB). See
18   * [Optimize LCP](https://web.dev/articles/optimize-lcp) for details.
19   */
20  timeToFirstByte: number;
21  /**
22   * The delta between TTFB and when the browser starts loading the LCP
23   * resource (if there is one, otherwise 0). See [Optimize
24   * LCP](https://web.dev/articles/optimize-lcp) for details.
25   */
26  resourceLoadDelay: number;
27  /**
28   * The total time it takes to load the LCP resource itself (if there is one,
29   * otherwise 0). See [Optimize LCP](https://web.dev/articles/optimize-lcp) for
30   * details.
31   */
32  resourceLoadDuration: number;
33  /**
34   * The delta between when the LCP resource finishes loading until the LCP
35   * element is fully rendered. See [Optimize
36   * LCP](https://web.dev/articles/optimize-lcp) for details.
37   */
38  elementRenderDelay: number;
39  /**
40   * The `navigation` entry of the current page, which is useful for diagnosing
41   * general page load issues. This can be used to access `serverTiming` for example:
42   * navigationEntry?.serverTiming
43   */
44  navigationEntry?: PerformanceNavigationTiming;
45  /**
46   * The `resource` entry for the LCP resource (if applicable), which is useful
47   * for diagnosing resource load issues.
48   */
49  lcpResourceEntry?: PerformanceResourceTiming;
50  /**
51   * The `LargestContentfulPaint` entry corresponding to LCP.
52   */
53  lcpEntry?: LargestContentfulPaint;
54}

TTFBAttribution

1interface TTFBAttribution {
2  /**
3   * The total time from when the user initiates loading the page to when the
4   * page starts to handle the request. Large values here are typically due
5   * to HTTP redirects, though other browser processing contributes to this
6   * duration as well (so even without redirect it's generally not zero).
7   */
8  waitingDuration: number;
9  /**
10   * The total time spent checking the HTTP cache for a match. For navigations
11   * handled via service worker, this duration usually includes service worker
12   * start-up time as well as time processing `fetch` event listeners, with
13   * some exceptions, see: https://github.com/w3c/navigation-timing/issues/199
14   */
15  cacheDuration: number;
16  /**
17   * The total time to resolve the DNS for the requested domain.
18   */
19  dnsDuration: number;
20  /**
21   * The total time to create the connection to the requested domain.
22   */
23  connectionDuration: number;
24  /**
25   * The total time from when the request was sent until the first byte of the
26   * response was received. This includes network time as well as server
27   * processing time.
28   */
29  requestDuration: number;
30  /**
31   * The `navigation` entry of the current page, which is useful for diagnosing
32   * general page load issues. This can be used to access `serverTiming` for
33   * example: navigationEntry?.serverTiming
34   */
35  navigationEntry?: PerformanceNavigationTiming;
36}

Browser Support

The web-vitals code is tested in Chrome, Firefox, and Safari. In addition, all JavaScript features used in the code are part of (Baseline Widely Available), and thus should run without error in all versions of these browsers released within the last 30 months.

However, some of the APIs required to capture these metrics are currently only available in Chromium-based browsers (e.g. Chrome, Edge, Opera, Samsung Internet), which means in some browsers those metrics will not be reported.

Browser support for each function is as follows:

• onCLS(): Chromium
• onFCP(): Chromium, Firefox, Safari
• onINP(): Chromium
• onLCP(): Chromium, Firefox
• onTTFB(): Chromium, Firefox, Safari

Limitations

The web-vitals library is primarily a wrapper around the Web APIs that measure the Web Vitals metrics, which means the limitations of those APIs will mostly apply to this library as well. More details on these limitations is available in this blog post.

The primary limitation of these APIs is they have no visibility into <iframe> content (not even same-origin iframes), which means pages that make use of iframes will likely see a difference between the data measured by this library and the data available in the Chrome User Experience Report (which does include iframe content).

For same-origin iframes, it's possible to use the web-vitals library to measure metrics, but it's tricky because it requires the developer to add the library to every frame and postMessage() the results to the parent frame for aggregation.

[!NOTE] Given the lack of iframe support, the onCLS() function technically measures DCLS (Document Cumulative Layout Shift) rather than CLS, if the page includes iframes).

Development

Building the code

The web-vitals source code is written in TypeScript. To transpile the code and build the production bundles, run the following command.

1npm run build

To build the code and watch for changes, run:

1npm run watch

Running the tests

The web-vitals code is tested in real browsers using webdriver.io. Use the following command to run the tests:

1npm test

To test any of the APIs manually, you can start the test server

1npm run test:server

Then navigate to http://localhost:9090/test/<view>, where <view> is the basename of one the templates under /test/views/.

You'll likely want to combine this with npm run watch to ensure any changes you make are transpiled and rebuilt.

Integrations

• Web Vitals Connector: Data Studio connector to create dashboards from Web Vitals data captured in BiqQuery.
• Core Web Vitals Custom Tag template: Custom GTM template tag to add measurement handlers for all Core Web Vitals metrics.
• web-vitals-reporter: JavaScript library to batch callback functions and send data with a single request.

License

Apache 2.0