Real user monitoring, usually shortened to RUM, is the practice of measuring how a website or application performs for the people actually using it, from inside their own browsers, and sending those measurements back for analysis. It exists because a page that loads in one second on a developer's laptop can take six on a mid-range phone over a congested network, and only visitors experience the second number. A scripted test from a data centre describes one device on one connection; RUM describes the whole distribution. Google's web.dev documentation treats "field data" and RUM as interchangeable terms, which is how most of the industry now uses them.

How the measurement works

A deployment starts with a small JavaScript file on every page, which reads timing data the browser already records. The Navigation Timing interface exposes timestamps for each phase of loading the main document: redirects, DNS lookup, connection, time to first byte (TTFB) and the moment loading completes. Resource Timing does the same for every image, stylesheet and script. A PerformanceObserver lets the script subscribe to newer entries as they occur, including largest contentful paint (LCP), layout shifts and the timing of user interactions.

The script holds those values until the page is hidden or closed, then transmits them in a single request known as a beacon. A beacon typically carries metric values alongside context: URL, device class, browser, connection type, country and sometimes a session identifier.

Aggregation is where interpretation begins. A single visit produces one LCP value; a site produces a distribution. Tools reporting Core Web Vitals summarise that distribution at the 75th percentile, the point three quarters of visits reach or beat. As of September 2026, Google classifies LCP as good at 2.5 seconds or less, interaction to next paint (INP) at 200 milliseconds or less, and cumulative layout shift (CLS) at 0.1 or less. Joost de Valk's website specification, published in May 2026, lists those targets as required, measured at the 75th percentile of real users.

A worked example from web.dev shows why the percentile matters. Where 88% of visits reach LCP within 2.5 seconds, 8% land between 2.5 and 4 seconds and 4% exceed 4 seconds, the page passes, even though one visit in eight missed the threshold. A 95th percentile view would fail the same page.

Sampling is the other configuration choice. High-traffic sites often beacon only a fraction of sessions to control cost, and web.dev warns that a sample too small to represent the audience skews every metric derived from it.

Who operates it

Three groups run RUM. Site owners deploy it through open-source libraries such as Boomerang, now maintained by Akamai, or through commercial platforms. Datadog, for example, announced its RUM product in July 2019. Infrastructure providers run it from the network edge: Cloudflare made RUM free for all accounts when Observatory entered open beta on September 26, 2025, with automatic deployment only for free zones outside the European Union and United Kingdom.

The third operator is the browser. The Chrome User Experience Report (CrUX) is RUM collected by Chrome from users who have opted in, published as a public dataset over a rolling 28-day window. It feeds the field data shown in PageSpeed Insights and Search Console. With no site-side script involved, CrUX also covers competitors.

On the sell side, publishers use RUM to see how ad tags, header bidding wrappers and consent platforms slow their pages. On the buy side, advertisers apply it to landing pages, where delay costs conversions rather than inventory.

From beacons to browser standards

Boomerang, a library built by Philip Tellis in Yahoo's Exceptional Performance team, recorded its first commit in May 2010. Stewardship passed to Log-Normal in 2011, to SOASTA in 2012 and to Akamai in 2017, according to the project's copyright notices. Early scripts relied on JavaScript timers blind to anything before the script loaded.

Browsers closed that gap. The W3C published Navigation Timing as a Recommendation on December 17, 2012, edited by Zhiheng Wang of Google, and High Resolution Time reached the same status that day. User Timing followed on December 12, 2013.

Google then built its own collection layer. The Chromium blog introduced CrUX on October 23, 2017, and by December 14 that year the dataset covered more than 1 million origins, with monthly releases. PageSpeed Insights moved to Lighthouse on November 12, 2018, displaying lab and field data together for the first time.

The decisive shift came in May 2020, when the Chromium team introduced Web Vitals, naming LCP, first input delay (FID) and CLS as the core set. Search attached consequences. The page experience ranking signal rolled out on mobile between mid-June and the end of August 2021 and reached desktop from February 2022.

The metrics kept moving. INP entered the programme in May 2022 and replaced FID as a Core Web Vital on March 12, 2024. Chrome then withdrew FID from PageSpeed Insights, the CrUX API and BigQuery from the 202409 dataset onward. Google removed the Page Experience report from Search Console on November 18, 2024, keeping the separate Core Web Vitals report. In December 2025, Safari 26.2, Firefox 146 and Chrome 143 reached parity on the underlying measurement APIs, ending a period in which LCP and INP could be collected reliably only from Chromium browsers.

Why marketers watch it

The commercial case rests on conversion data. Shopify analysed Core Web Vitals across its merchant base over 28 days in January and February 2026 and found conversion roughly 3.5% lower for every additional 100 milliseconds of LCP, with stores at 2.5 seconds converting about 30% below those at 1.5 seconds. Both figures are Shopify's own. The explainer on latency places that result alongside the timeouts that govern programmatic auctions.

Publishers see a direct revenue link. In March 2026, AdSense emailed publishers promoting bfcache and speculation rules, citing a 27% page view increase at Netzwelt and 9% mobile revenue growth at Yahoo! JAPAN News, figures supplied by Google. Many sites remain far from passing: a holiday 2025 benchmark found 76% of retail and wholesale sites failing Core Web Vitals assessments on mobile.

Field data also catches errors that lab scores reward. A case shared in November 2025 described lazy loading lifting a PageSpeed score from 65 to 92 while field LCP worsened from 1.8 to 4.2 seconds, followed by a 20% traffic decline. Behavioural analytics now borrow the same signals: Microsoft Clarity allows quick-back sessions to be filtered by LCP, INP or CLS.

Where it breaks down

Coverage comes first. CrUX includes only Chrome users, excludes Chrome on iOS because that build runs on WebKit, and requires enough traffic for a statistically significant sample. First-party RUM sees only visitors whose browsers load the script, so blocked tags and refused consent leave gaps. CrUX's 28-day window can hide yesterday's regression for weeks.

Consent is the second problem. RUM scripts read device data, and European rules generally require permission for that. In December 2024, Microsoft Clarity began requiring an explicit call to its Consent API before enabling session recordings and funnel tracking. In the United States, analytics and session replay code have drawn wiretapping claims under the California Invasion of Privacy Act (CIPA). The Ace Hardware complaint of April 2026 alleged that analytics code kept firing after users rejected non-essential cookies. California's SB 690, passed by state lawmakers and awaiting signature at the time of that report, would narrow one CIPA provision while leaving the Section 631 wiretapping clause untouched.

The privacy critique predates the lawsuits. Princeton researchers Steven Englehardt, Gunes Acar and Arvind Narayanan reported in November 2017 that session replay scripts from seven vendors ran on 482 of the Alexa top 50,000 sites, and that whole-page collection could leak medical and payment details to third parties.

Measurement also adds weight, which is why Boomerang's maintainers recommend loading it through an asynchronous iframe that cannot delay the onload event.

The weight given to the metrics is disputed too. Among 131 search practitioners surveyed in September 2026, speed and Core Web Vitals ranked among the most contested signals. Google's Martin Splitt has described their ranking effect as tiny in most cases.

Advertising itself resists instrumentation. Most ads render inside cross-origin iframes, and Chrome's documentation states that no JavaScript API exists to let RUM tooling detect them. A publisher cannot measure its own ad density from the page.

Not the same as

Synthetic monitoring, often called lab testing, loads pages from scripted agents on fixed devices and networks. Results are reproducible but describe one scenario. Lighthouse is the best-known example.

CrUX is one RUM dataset rather than the practice. It is Chrome-only, opt-in, public and fixed at the 75th percentile over 28 days. A site's own RUM can cover other browsers, any percentile and any window, so the two seldom match exactly.

Session replay records clicks, scrolls and page content to reconstruct individual visits. Performance monitoring needs only timings.

Application performance monitoring (APM) instruments servers and back-end code. RUM begins where the response reaches the device.

Recent developments

Today, Chrome added four experimental advertising metrics to CrUX: Ad Count, Ad Density, Ad Weight: CPU and Ad Weight: Network, reported at the 75th percentile for pages naming at least one authorised seller in ads.txt. They sit outside Core Web Vitals and carry no thresholds. The CPU figure excludes ad scripts running in the main frame, where header bidding wrappers usually sit. For ad load, the browser now supplies what site-side RUM cannot, and the company publishing the figures also runs the largest ad server and sells the ads being measured.

Timeline

  • May 2010: First commit to Boomerang at Yahoo
  • December 17, 2012: Navigation Timing and High Resolution Time become W3C Recommendations
  • December 12, 2013: User Timing becomes a W3C Recommendation
  • October 23, 2017: Chrome User Experience Report introduced
  • November 15, 2017: Princeton researchers publish session replay findings covering 482 of the top 50,000 sites
  • December 14, 2017: CrUX expands to more than 1 million origins with monthly releases
  • November 12, 2018: PageSpeed Insights moves to Lighthouse and displays CrUX field data
  • July 17, 2019: Datadog announces Real User Monitoring
  • May 2020: Chromium team introduces Web Vitals with LCP, FID and CLS as the core set
  • June 15, 2021: Page experience ranking begins rolling out on mobile, completed by the end of August 2021
  • February 2022: Page experience ranking extends to desktop
  • May 2022: INP enters the Web Vitals programme
  • March 12, 2024: INP replaces FID as a Core Web Vital
  • September 2024: Chrome announces FID removal from CrUX tools from the 202409 dataset
  • November 18, 2024: Page Experience report removed from Search Console
  • December 2024: Microsoft Clarity requires a Consent API call for recordings and funnel tracking
  • September 26, 2025: Cloudflare launches Observatory with free RUM for all accounts
  • December 2025: Safari 26.2, Firefox 146 and Chrome 143 reach parity on Core Web Vitals measurement
  • April 27, 2026: Shopify publishes its storefront speed and conversion analysis
  • September 15, 2026: Chrome adds four experimental ad metrics to CrUX

Summary

Who: Site owners, publishers and advertisers deploy RUM through open-source libraries such as Boomerang, commercial platforms such as Datadog, and infrastructure providers such as Cloudflare. Google operates the largest public RUM dataset through Chrome, and the W3C Web Performance Working Group standardises the browser interfaces the scripts rely on.

What: A passive measurement method that collects performance timings, including LCP, INP, CLS and TTFB, from real visits in real browsers, sends them in beacons and reports them as distributions, usually at the 75th percentile.

When: Script-based collection dates to at least May 2010, browser timing standards to December 2012, CrUX to October 2017 and Core Web Vitals to May 2020. INP replaced FID on March 12, 2024, and Chrome added experimental ad metrics to CrUX today, September 15, 2026.

Where: In visitors' browsers and apps, on collection servers run by vendors and edge networks, and in Chrome's public CrUX dataset, surfaced through PageSpeed Insights, Search Console, the CrUX API and BigQuery.

Why: Lab tests describe one device on one connection, while revenue depends on every visitor. Field data ties speed to conversion and search visibility, but it carries coverage gaps, consent obligations, litigation exposure and a blind spot for ads served in cross-origin frames.