Web performance measurement guide

How to use GTmetrix: measurement conditions, metrics, and reports

A GTmetrix score is not an absolute property of a page. It is a lab measurement affected by location, device, connection, viewport, time, and tool version. Fix conditions, investigate from Summary through Performance, Structure, CrUX, and Waterfall, and retest under the same conditions.

Published 2024.12.10 Updated 2026.11.05 Written by Finite Field Editorial Team
GTmetrix Summary tab showing Speed Visualization, Top Issues, and Page Details
An official 2026 GTmetrix Summary tab image. Screens and metrics change; consult the official interface when using it.

This article was edited using official 2025–2026 GTmetrix materials and screens checked as of July 25, 2026. The interface, free allowance, test environment, Lighthouse version, and score calculation can change. Improved scores, rankings, and sales are not guaranteed.

Decide first

Before looking at scores, decide whose conditions to reproduce

If the goal is better experience for mobile users in Tokyo, judging only Seattle, Desktop Chrome, and Unthrottled does not match them. Use analytics to choose key regions, devices, and connections and hold them constant during comparison.

Comparison principle

Measure repeatedly with the same URL, location, device, connection, viewport, and settings.

What this article helps you do

Use GTmetrix for conditional diagnosis and change validation, not as a one-time grading machine.

Fix conditions

Record Location, Device/Browser, Connection Speed, URL, time, and options.

Read tabs by purpose

Start with Summary, then use Performance, Structure, CrUX, and Waterfall for different questions.

Interpret metrics

Interpret LCP, TBT, CLS, and field INP by role as well as Grade.

Verify changes

Change one item at a time and compare medians and variability from repeated runs under identical conditions.

Six stages of using GTmetrix

Fix measurement conditions

A standard GTmetrix report is a lab, or synthetic, measurement running Chrome under selected conditions. The same URL varies with conditions. Maintain a baseline set for before/after comparisons and a separate set representing key users.

Location

Distance between the test server and origin affects latency. Choose a location near key users and keep it fixed during comparisons. Check the official interface for free regions and PRO scope.

Record example:Tokyo equivalent / Seattle fixed

Device / Browser

Desktop and Simulated Mobile differ in viewport, CPU, delivered images, and responsive branches. Base the choice on common devices in analytics.

Record example:Chrome Desktop 1366×768

Connection Speed

Default Unthrottled is very fast and may not represent typical users. Choose conditions such as 4G, LTE, or Broadband that resemble usage and record the named speed.

Record example:LTE / 4G Slow / Broadband

Exact URL

HTTP versus HTTPS, trailing slashes, query strings, and login state affect results. Record both the entered URL and redirect chain, not just the final URL.

Record example:https://example.com/page/

Additional options

Adblock, Video, Cookie, and HTTP Authentication change the measurement. Record enabled and disabled options and do not switch them mid-comparison.

Record example:Adblock off / Video on

Time and count

Shared infrastructure, CDN cache, third-party scripts, and load cause variability. In practice, run at least three tests under identical conditions and record median and range rather than a single value.

Record example:Median and minimum/maximum of three to five runs

Measure using the interface documented at the verification date

At the verification date, GTmetrix required login to a verified account to view reports. The older article's claim that registration was unnecessary no longer applied. Free allowances and limits can change; use the official account screen on the day of work.

1. Select targets

Select representative URLs by page type

Choose representative URLs for different templates such as articles, products, and forms, not just the home page. Define the target and success criteria first.

2. Enter a URL

Enter URL to Analyze

Enter a public URL. For authenticated pages, avoid casually sharing credentials or cookies; use a dedicated test environment and official options.

3. Choose conditions

Location / Device / Connection

Check the three fields and Analysis Options before Analyze. Record a preset's contents too; do not infer conditions from its name alone.

4. Run repeated tests

Three to five runs under the same conditions

Do not use only the first run. Measure repeatedly with the same settings; treat warm and cold cache comparisons as separate series.

5. Save the report URL

Date, conditions, report link

Save measurement date, conditions, key metrics, report URL, and changes together rather than copying only Grade.

6. Establish a baseline

median + variation

Use the median as a baseline and investigate high variability. Compare after changes using the same count and conditions.

Defaults may not represent your users

Official material updated in May 2026 gave defaults of Seattle, Chrome Desktop, and Unthrottled. Do not judge mobile users solely on Unthrottled; add conditions matching your analytics.

Read report tabs by purpose

Do not decide changes from Summary alone; inspect detailed tabs for the symptom. GTmetrix Grade is an entry point, not a cause.

Questions answered by each tab
TabMain displayDecision question
SummaryVisualization / Top Issues / Page DetailsStart here to decide where to investigate. Treat top issues as hypotheses and confirm in detailed tabs.
PerformanceFCP / SI / LCP / TBT / CLSUse lab measurements to identify slow display, blocking, or layout shifts. Browser Timings are supplementary.
StructureLighthouse audits / GTmetrix auditsInspect which implementation candidates affect metrics. Confirm the relevant resource and effect rather than simply seeking a score of 100.
CrUXLCP / INP / CLS / TTFB / FCPReview real-user field data. Origin-level data may appear when URL data is insufficient; GTmetrix aggregates across devices and other categories.
Waterfallrequest / status / timing / initiatorIsolate slow requests, redirects, third-party scripts, cache, connections, and backend waiting over time.
Video / HistoryPlayback / trend graphReview visual progress and long-term trends. Video must be enabled beforehand; compare History series under the same conditions.

Do not mix lab and field

Regular tab values are lab measurements under selected conditions; CrUX aggregates real users over the previous 28 days. They are not the same moment or population, so differing values are not necessarily errors.

INP and TBT are different metrics

INP measures responsiveness to real interactions in the field; TBT measures main-thread blocking during lab load. GTmetrix shows TBT as a comparison proxy, not a replacement for INP.

Do not compare directly with PageSpeed Insights

Even Lighthouse-based tools differ in hardware, location, network, throttling, and version. Prefer consistent trends within each tool over cross-tool score differences.

Interface names change

Do not rely on the old PageSpeed/YSlow interface. Refer to Summary, Performance, Structure, CrUX, and Waterfall and check official release notes.

Turn metrics into symptoms and possible causes

A low score alone does not identify a change. Use metrics to identify symptoms, audits and Waterfall to narrow down resources, and inspect implementation causes.

Combinations to check first

  • Slow LCP: inspect the LCP element, image fetch, server response, and render blocking.
  • High TBT: inspect long tasks, JavaScript evaluation, third-party scripts, and the main thread.
  • High CLS: inspect media without dimensions, fonts, slow requests, and inserted iframes.
  • Slow TTFB: inspect redirects, connection, backend, cache, and regional differences in Waterfall.
  • Transfers or failures: inspect status, Location, request initiators, and dependencies.

Avoid immediate conclusions

  • Do not choose a new host solely because the Grade is low.
  • Do not treat a Structure Score of 100 as a real-user performance guarantee.
  • Do not conclude that field INP is good merely because TBT is good.
  • Do not report a single measurement difference as an improvement.
  • Do not combine scores from different locations, devices, connections, or Lighthouse versions into one series.

Check resources and impact before fixing every high-impact issue

Top Issues impact helps prioritize, but consider site goals, implementation cost, side effects, and field data. If multiple audits point to one resource, fixing one cause may change several values.

A practical workflow for isolating causes

Connect candidate changes to observed symptoms and specific resources. Choose countermeasures from evidence before deciding to add a WordPress plugin or move hosts.

  1. 1

    Choose one symptom

    Choose one metric to improve, such as LCP, TBT, CLS, TTFB, transfers, or request size. Do not target Grade alone.

  2. 2

    Identify the element or request

    Expand Performance and Structure details and inspect DOM elements, URLs, potential savings, and main-thread tasks.

  3. 3

    Inspect sequence in Waterfall

    Inspect dependencies from document to LCP resource, DNS, connection, SSL, backend, redirects, and third-party request start times and waits.

  4. 4

    Choose one change

    Choose a change that directly addresses the observed cause, such as image format or dimensions, preload, cache, deferred scripts, fewer third parties, or server processing.

  5. 5

    Plan side effects and rollback

    Check for damage to function, display, analytics, ads, and forms; save prior values and rollback steps. Avoid duplicate cache plugins.

  6. 6

    Retest under the same conditions

    Run the same number of tests with the same URL, location, device, connection, and options as before. Compare medians, variability, target metrics, and effects on other metrics.

Establish causation before changing servers

If TTFB remains slow across regions and runs, isolate application processing, cache, database, external APIs, and delivery before assessing environment limits and migration effects. Do not recommend a move from one GTmetrix result.

Retest under consistent conditions and check long-term trends

Immediate lab verification and long-term field checks operate on different timescales. Check regressions in the lab immediately after release, then assess user impact with CrUX or real-user monitoring after enough time has passed.

Attach a conditions table

Attach URL, Location, Device/Browser, Connection, viewport, date, Preset, Video/Adblock, run count, and tool version to before and after results.

Review median and range

Do not take the best single run. Review median and minimum/maximum; withhold conclusions if variability exceeds the effect.

Check target metrics and regressions

Check CLS, TBT, request count, page size, function, and appearance as well as the targeted LCP or other metric. Do not hide deterioration elsewhere.

Separate History series

Treat different Analysis Options as separate series. Note major Lighthouse or GTmetrix update dates and do not directly compare scores across them.

Allow time for CrUX

CrUX aggregates the latest 28 days of field data; some URLs lack sufficient page-level data. Do not assess immediate release effects using CrUX alone.

Use real-user metrics too

Use RUM, Search Console Core Web Vitals, business events, and error rates to understand effects on users and functions.

Release-stop and rollback conditions

Roll back for broken layouts or forms, missing analytics, increased errors, worse CLS, or worse performance on key devices. Do not accept a score-only improvement that harms function or field experience.

Locate controls in official 2025–2026 screenshots

Rather than memorizing screen names, check Analysis Options before starting and the boundary between lab and field data in the report. The images below come from official GTmetrix articles and retain their source dates.

GTmetrix dashboard URL entry and the Location, Device/Browser, and Connection Speed measurement options
Analysis Options in GTmetrix's Basic Analysis With GTmetrix, updated May 14, 2026. Check Location, Device/Browser, and Connection Speed after entering the URL and before Analyze. Captured and checked July 25, 2026.
GTmetrix CrUX tab showing Core Web Vitals assessment and LCP, INP, and CLS field data
Official GTmetrix CrUX tab image published January 2025. Read the 28-day real-user aggregate separately from lab values. The interface may change. Captured and checked July 25, 2026.

Sources

GTmetrix updates its interface, free allowances, conditions, and score calculation. Check the latest official sources when working.

GTmetrix frequently asked questions

Can I use GTmetrix without registering?

As described at the verification date, viewing generated reports requires logging into a verified account. Free quotas, retention, and available regions or devices can change, so check the official account screen on the day of use.

Does an A Grade mean Core Web Vitals pass too?

No. Grade combines GTmetrix Performance and Structure. Check Core Web Vitals with CrUX field data for LCP, INP, and CLS. If page-level data is insufficient, origin-level data may be shown instead.

Why do GTmetrix and PageSpeed Insights scores differ?

Location, hardware, network, throttling, Lighthouse version, and Analysis Options differ. Do not force the scores to match; track changes within each tool under consistent conditions.

Is improving LCP alone enough?

No. LCP concerns loading; check responsiveness with INP or lab proxy TBT and visual stability with CLS. Also assess functionality, errors, and real-user data.

How many measurements should I run?

In practice, measure at least three times under the same conditions, or five or more if variability is high, then compare median and range. More runs do not make differing conditions comparable.

Should I switch hosts if TTFB is slow?

Do not decide from one value. Separate redirect, connection, and backend in Waterfall, reproduce across regions and runs, and isolate cache, application, database, and external API before assessing hosting limits.

Related articles

GTmetrix tests hypotheses under consistent conditions

Choose conditions matching key users, run repeated tests with the same settings, and investigate from Summary into detailed tabs. Change one item at a time and review medians, variability, functional regressions, and field data. The aim is to safely reduce observed user-experience problems, not merely raise a score.

Written and reviewed by

Finite Field Editorial Team

We review official GTmetrix updates and interface images alongside Lighthouse and CrUX measurement conditions, emphasizing reproducible comparison and regression testing after changes.

Web performance advice

Organize measurement, improvements, and retesting

When multiple templates, third-party scripts, cache, CDN, and application processing interact, combine RUM, lab results, Waterfall, and server logs to prioritize causes.

Discuss performance diagnosis

We do not guarantee improvements in GTmetrix Grade, Core Web Vitals, rankings, traffic, or sales. We determine scope after reviewing conditions and change risks.