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.
| Tab | Main display | Decision question |
|---|---|---|
| Summary | Visualization / Top Issues / Page Details | Start here to decide where to investigate. Treat top issues as hypotheses and confirm in detailed tabs. |
| Performance | FCP / SI / LCP / TBT / CLS | Use lab measurements to identify slow display, blocking, or layout shifts. Browser Timings are supplementary. |
| Structure | Lighthouse audits / GTmetrix audits | Inspect which implementation candidates affect metrics. Confirm the relevant resource and effect rather than simply seeking a score of 100. |
| CrUX | LCP / INP / CLS / TTFB / FCP | Review real-user field data. Origin-level data may appear when URL data is insufficient; GTmetrix aggregates across devices and other categories. |
| Waterfall | request / status / timing / initiator | Isolate slow requests, redirects, third-party scripts, cache, connections, and backend waiting over time. |
| Video / History | Playback / trend graph | Review 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
Choose one symptom
Choose one metric to improve, such as LCP, TBT, CLS, TTFB, transfers, or request size. Do not target Grade alone.
- 2
Identify the element or request
Expand Performance and Structure details and inspect DOM elements, URLs, potential savings, and main-thread tasks.
- 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
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
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
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.


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