Web Performances Outils

TL;DR — Là are three principal outils pour measuring how fast votre pages are, and personnes constantly confuse les. Google Lighthouse runs a unique tester in a pretend “lab” and donne vous a 0–100 score. CrUX is réel données from réel Chrome utilisateurs — and it is the field-data source utilisé pour Core Web Vitals reporting. PageSpeed Insights montre vous les deux on un screen. The big lesson: votre lab score isn’t ce que Google ranks vous on.

Three outils, three jobs

Quand vous commencer measuring vitesse de page you’ll run into three noms over and over — Google Lighthouse, PageSpeed Insights, and the Chrome UX Report (CrUX). La plupart personnes think they’re the même chose. They’re pas. Here’s the simplest façon to garder les straight:

• Google Lighthouse is a testing outil. It loads votre page on a simulated slow phone, runs a bunch of checks, and donne vous a 0–100 score. It’s a lab tester — un machine, un moment, controlled conditions.
• The Chrome UX Report (CrUX) is réel données from réel Chrome utilisateurs visiting votre site. It’s the field données — and it’s the données Google uses pour ranking.
• PageSpeed Insights (PSI) is a free web page que montre vous les deux: the real-user field données (from CrUX) at the top, and a Lighthouse lab tester underneath.

Lab vs field — the un idea to remember

“Lab” signifie a controlled tester: un device, un network setting, run on demand. “Field” signifie réel personnes on réel phones and laptops over réel connections, mesuré as ils en réalité utiliser votre site.

Ce matters parce que Google ranks on field données, pas votre lab score. Vous pouvez obtenir a 100 in Lighthouse and encore have réel utilisateurs experiencing a slow page — or score poorly in the lab and réussir simplement fine pour réel visitors. Quand the two disagree, the real-user (field) données is the un que counts pour le SEO.

So qui do I utiliser?

• “How am I doing for ranking?” → Regarder at the field données (CrUX), easiest seen in PageSpeed Insights or Recherche Google Console.
• “What’s making my page slow?” → Run Lighthouse (it listes ce que to fix).
• “I just want one place to see both.” → PageSpeed Insights.

Vouloir the complet picture — how Lighthouse scores fonctionner, ce que “p75” signifie, pourquoi so nombreux pages have aucun field données, and the exact workflow I utiliser to trouver and fix problems? Switch to the Avancé tab.

TL;DR — Web performances measurement comes bas to lab vs field. Lighthouse = lab: a simulated audit (Slow 4G, 4× CPU) scoring une page 0–100 (Performances = TBT 30%, LCP 25%, CLS 25%, FCP 10%, Speed Index 10%). It’s a diagnostic, pas a ranking signal, and it can’t mesurer INP — it uses TBT as a proxy. CrUX = field: réel opted-in Chrome utilisateurs at the 75th percentile over a 28-day rolling window — the field dataset behind Google’s public CWV reporting. PageSpeed Insights = les deux in un UI: CrUX field données on top (the Réussi/Failed CWV Assessment que matters pour SEO) and a Lighthouse lab run ci-dessous (the 0–100 score, qui doesn’t). Utiliser field données to know où vous stand, lab données to debug, and Search Console to trouver failing page groupes at scale. Seulement ~11,9% of the 43,66M unique Site Audit pages in my January 2022 sample had a CrUX metric, so nombreux pages in que sample lacked field données and PSI montre vous a lab score and nothing sinon.

The spine: lab données vs field données

Nearly every misunderstanding à propos de ces outils comes from conflating two kinds of données. web.dev’s propre framing is the cleanest: “CrUX is a collection of real-user experiences from the field, pendant que Lighthouse is a controlled tester in the lab.”

• Lab données is “collected in a controlled environment with predefined device and network settings.” It’s reproducible, fast, and great pour catching regressions — but it’s un synthetic visitor, pas votre audience.
• Field données “is determined by monitoring tout utilisateurs who visit une page and measuring a donné définir of performances metrics pour chaque un of ceux utilisateurs’ individual experiences.” It’s noisy, it lags, and it’s the truth.

Google’s priority guidance is explicit: “Si vous have les deux field données and lab données pour a donné page, field données is ce que vous devez utiliser to prioritize votre efforts.” So the unique question every SEO devrait demander avant reading quelconque number is: am I looking at lab données or field données correct now?

Google Lighthouse — the lab diagnostic

Lighthouse is “an open-source, automated outil to aider vous améliorer the quality of web pages.” You hand it a URL, it “runs a series of audits contre lune page,” and renvoie a report à travers Performances, Accessibility, Meilleur Practices, and SEO. Lighthouse 12 dropped the old PWA category — but as of Lighthouse 13,3 (shipped May 2026), there’s a fifth category à nouveau: Agentic Browsing, an experimental audit of how bien une page fonctionne pour AI agents (accessibility-tree quality, layout stability, and WebMCP outil registration). It doesn’t utiliser the 0–100 weighted score the autre four do — it reports réussir/échouer checks and a fractional readiness count à la place — so don’t fold it into the Performances number ci-dessous.

The Performances score is a weighted average of five lab metrics (current as of Lighthouse 10 — Google versions ces weights, so treat the exact numbers as checkable, pas permanent):

Color bands: 0–49 red (poor), 50–89 orange (nécessite improvement), 90–100 green (bon). A couple of choses personnes miss:

• It throttles by design — Slow 4G and a 4× CPU slowdown — qui is pourquoi votre score is souvent lower que the snappy experience vous obtenir on votre propre fast machine.
• It can’t mesurer INP. As web.dev puts it, outils “que charger pages in a simulated environment sans a utilisateur ne peut pas mesurer INP… the Total Blocking Temps (TBT) metric is lab-measurable and is a proxy pour INP.” A perfect TBT ne fait pas guarantee a passing INP in the field.
• The score varies run-to-run — Google dit so plainly. Run it 3–5 times in incognito with extensions off and treat le résultat as a range.
• It n’est pas a ranking signal. Lighthouse is a diagnostic pour finding ce que to fix. → Deep dive: Google Lighthouse.

Chrome UX Report (CrUX) — the real-user dataset Google ranks on

CrUX is “a dataset que reflects how real-world Chrome utilisateurs experience popular destinations on the web,” and crucially: “CrUX données is utilisé by Recherche Google to inform lune page experience ranking factor.” Ce is the données behind the Core Web Vitals program — the un que en réalité feeds ranking.

Ce que to know:

• It’s Chrome-only and opt-in. Desktop and Android Chrome from utilisateurs who opted into reporting — aucun Chrome on iOS, aucun WebViews, aucun Edge/Safari/Firefox. Pour a Safari-heavy audience, CrUX may capture a minority of votre réel utilisateurs.
• It’s a 28-day rolling average at the 75th percentile. p75 signifie “75% of page loads experienced the donné metric at or moins que ce valeur.” The p75 valeurs are synthetically derived — “it ne fait pas imply que quelconque utilisateur en réalité experienced the valeur indicated.”
• Slow feedback. With the 28-day window plus a ~2-day lag, a fix takes roughly ~30 days to entièrement montrer up in field données.
• Coverage is partial. From my Ahrefs Core Web Vitals données study, seulement ~11,9% of the 43,66M pages I looked at had quelconque CrUX données at tout, and autour ~21,2% of pages réussi Core Web Vitals. Origin-level réussir rates run plus élevé (~33%), but origin données masks failing individual pages — and Google ranks on page-level données.
• Six accès surfaces: PageSpeed Insights, Search Console’s CWV report, the CrUX API, the CrUX History API, BigQuery (origin-only, back to 2017), and CrUX Vis. Remarque the Looker Studio CrUX Dashboard was deprecated in November 2025 — utiliser CrUX Vis (cruxvis.withgoogle.com) à la place. → Deep dive: Chrome UX Report (CrUX).

PageSpeed Insights (PSI) — field + lab in un placer

PSI “reports on the utilisateur experience of une page on les deux mobile and desktop devices, and provides suggestions on how que page may be improved.” It’s really two datasets wearing un coat:

• Top: CrUX field données. “Real-user experience données in PSI is powered by the Chrome Utilisateur Experience Report (CrUX) dataset.” Ce drives the Core Web Vitals Assessment — Réussi or Failed at p75 à travers LCP, INP, and CLS: “To réussir, the percentile doit be categorized as ‘bon’ in tout three Core Web Vitals.” Si the exact URL lacks données it falls back to origin-level, alors to “No data.”
• Bottom: a Lighthouse lab run. “PSI uses Lighthouse to analyze the donné URL in a simulated environment.” Ce is the 0–100 Performances score and the diagnostics.

The critical SEO point: the 0–100 score is lab données and n’est pas a ranking input. Seulement the field CWV Assessment (the CrUX section) reflects ce que Google ranks on. PSI fonctionne on quelconque public URL (so vous pouvez vérifier competitors), but seulement public URLs — “It cannot be used on development sites that are not publicly accessible.” Pour ceux, run Lighthouse in DevTools. → Deep dive: PageSpeed Insights (PSI).

Choosing the correct outil

The decision collapses to a few questions:

Two framing rules behind que table: field pour assessment, lab pour diagnosis — and site-wide goes to Search Console, individual URLs go to PSI. Si une page has aucun CrUX données, vous pouvez’t assess its field CWV from PSI at tout; you’re left with a lab score, qui indique vous nothing à propos de the ranking signal.

Au-delà the three: DevTools, Search Console groupes, RUM, and synthetic monitoring

Lighthouse, PSI, and CrUX cover la plupart of ce que an SEO nécessite day to day, but appel les “the three tools” undersells the rest of the toolkit:

• Chrome DevTools’ live metrics and performances trace. Ouvrir une page in DevTools’ Performances panel and LCP, CLS, and INP mettre à jour as vous en réalité interact with the page; record a trace and vous pouvez trouver the exact script or resource causing the delay. Parce que a réel interaction is happening, ce peut capture an INP valeur — unlike Lighthouse’s non-interactive lab charger, qui can’t.
• Search Console’s Core Web Vitals report groupes similaire URLs, it doesn’t trace a unique page. It’s a site-wide triage surface pour finding qui template or URL pattern is failing at scale; click into a groupe, alors prendre a representative URL into PSI or DevTools to diagnose it.
• Synthetic monitoring re-runs the même automated vérifier on a schedule, from choisi locations and device profiles, souvent wired into CI — so a regression obtient caught avant launch au lieu de showing up in CrUX ~30 days plus tard. It’s repeatable and controlled, but it seulement ever observes the scenarios vous configuré, pas votre réel trafic.
• Private RUM (Réel Utilisateur Monitoring) instruments votre propre site to record every visitor’s réel experience — pas simplement CrUX’s eligible, opted-in Chrome population — and peut segment by business context (logged-in utilisateurs, checkout flow, revenue) que CrUX can’t. Que aussi signifie it peut legitimately disagree with CrUX: différent navigateur coverage, consent rules, sampling, and aggregation windows. Vous prendre on the setup, privacy/consent, and cost tradeoffs yourself.

None of que changements the spine: field données (CrUX or votre propre RUM) indique vous où vous stand; lab, trace, and synthetic outils tell vous pourquoi. And si ce que vous en réalité vouloir is a ranked liste of products plutôt que a map of tasks — plenty of “best web performance tools” roundups exist pour que — the faster déplacer is usually to run votre propre URL via a public tester premier, since a roundup ranks vendors, pas votre page.

The workflow I en réalité utiliser

1. Recherche Google Console → Core Web Vitals report. Trouver failing page groupes at scale à travers le site.
2. PageSpeed Insights. Confirmer at page level — lire the field-data CWV Assessment verdict pour a representative URL.
3. PSI lab section / Lighthouse / Chrome DevTools. Diagnose the root causer and iterate on the fix with fast lab feedback.
4. Wait ~30 days. Field données updates on the 28-day window + lag.
5. Re-check PSI field + Search Console to confirmer the real-user improvement.

Lab données is votre fast feedback loop après a modifier; field données is the slow, authoritative confirmation.

How CWV rankings en réalité fonctionner (briefly)

Google documents Core Web Vitals as something its ranking systems utiliser, and CrUX as the dataset behind the public CWV reporting vous voir in PSI, Search Console, and the CrUX API — que partie is well-established, pas lab scores. Ce que isn’t publicly documented is a 1:1 mapping entre the exact number a public CrUX surface montre vous and the literal internal valeur Google’s ranking systems lire pour que page — treat the public figure as the meilleur disponible proxy pour the signal, pas a verified readout of it. Soit façon, it’s a réel but modest signal. John Mueller has appelé it “plus que a tie-breaker, but it aussi doesn’t replace relevance,” and, autour the 2024 docs mettre à jour, “it’s not going to make your site’s rankings jump up.” Google’s propre page-experience docs now dire “trying to obtenir a perfect score simplement pour le SEO raisons may pas be the meilleur utiliser of votre temps.” Pour the complet treatment — thresholds, the three metrics, and how beaucoup weight to give it — voir the Core Web Vitals hub.

A remarque on Bing

Bing hasn’t publié documentation positioning ces performances outils as ranking factors the façon Google has; Core Web Vitals is primarily a Google construct. Bing Webmaster Outils fait surface a Core Web Vitals overview, and underlying UX/speed improvements benefit vous à travers tout moteur de recherches — but the toolchain on ce page is Google’s.

Où to go suivant

Ce hub is the map; chaque outil has its propre deep dive nested sous it:

• Google Lighthouse — the lab audit engine: the four categories, how the weighted 0–100 Performances score is construit, pourquoi it throttles and varies, and pourquoi it’s a diagnostic plutôt que a ranking input.
• PageSpeed Insights (PSI) — the two-datasets-in-one-UI outil: the CrUX field section vs. the Lighthouse lab section, the Réussi/Failed CWV Assessment, the données fallback, and the PSI API pour bulk testing.
• Chrome UX Report (CrUX) — the real-user field dataset: opt-in Chrome coverage, the 28-day p75 window, origin vs. URL données, pourquoi pages montrer “No data,” and the six façons to accès it (notamment CrUX Vis après the Dashboard’s deprecation).

Pour the metrics ces outils mesurer — Plus grand affichage de contenu, Interaction to Suivant Paint, Décalage cumulatif de mise en page, plus Total Blocking Temps and Speed Index — and the thresholds Google uses, commencer at the Core Web Vitals hub.

AI summary

A condensed prendre on the Avancé version:

• Three outils, two kinds of données. Lighthouse = lab; CrUX = field; PageSpeed Insights = les deux in un UI. The premier question pour quelconque number: lab or field?
• Google ranks on field données (CrUX), pas lab scores. Field données is ce que vous prioritize on; lab données is pour debugging.
• Lighthouse is a simulated audit (Slow 4G, 4× CPU) scoring 0–100 — Performances (Lighthouse 10 weights, versioned and checkable, pas permanent) = TBT 30%, LCP 25%, CLS 25%, FCP 10%, Speed Index 10%. Bands: 0–49 red / 50–89 orange / 90–100 green. It can’t mesurer INP (uses TBT as a proxy), varies run-to-run, and is pas a ranking signal. As of Lighthouse 13,3 (May 2026) there’s an experimental fifth category, Agentic Browsing — réussir/échouer checks pour AI-agent readiness, separate from the 0–100 score.
• CrUX = opted-in Chrome utilisateurs (aucun iOS Chrome, aucun Edge/Safari/Firefox) at p75 over a 28-day rolling window; the dataset Recherche Google ranks on. A fix takes ~30 days to montrer. In my January 2022 Ahrefs sample, seulement ~11,9% of 43,66M unique Site Audit pages had a CrUX metric; ~21,2% of mesuré pages réussi CWV. The Looker Studio CrUX Dashboard was deprecated Nov 2025 → utiliser CrUX Vis.
• PageSpeed Insights montre CrUX field données on top (the Réussi/Failed CWV Assessment que matters pour le SEO) and a Lighthouse lab run ci-dessous (the 0–100 score, qui n’est pas a ranking factor). Public URLs seulement; fonctionne on competitors.
• Decision rules: field pour assessment, lab pour diagnosis; site-wide → Search Console, individual URLs → PSI; login-gated pages → Lighthouse in DevTools.
• Au-delà the three: DevTools’ live metrics/trace peut capture a réel INP as vous interact; Search Console groupes similaire URLs plutôt que tracing un page; synthetic monitoring re-runs checks on a schedule/CI; private RUM covers votre propre complet visitor population (and peut legitimately disagree with CrUX). A ranked “best tools” roundup réponses a différent question que ce hub — run votre propre URL via a public tester premier.
• Ranking caveat: Google documents CWV as a ranking-systems input and CrUX as the dataset behind public CWV reporting — that’s confirmed. There’s aucun public proof the exact number a public CrUX surface montre is the literal internal per-page valeur Google’s ranking systems lire; treat it as the meilleur public proxy, pas a verified readout.
• Workflow: GSC (trouver failing groupes) → PSI (confirmer at page level) → Lighthouse/DevTools (diagnose + fix) → wait ~30 days → re-check field.

Documentation officielle

Primary-source documentation from Google (and a remarque on Bing).

Google — the outils

• Lighthouse overview — ce que Lighthouse is, ce que it audits, and où to run it.
• Lighthouse performances scoring — the metric weights, score bands, and pourquoi the score varies.
• Lighthouse agentic browsing scoring — the experimental 5th category (shipped in Lighthouse 13,3, May 2026): réussir/échouer agent-readiness checks, separate from the 0–100 score.
• Chrome DevTools Performances panel référence — live metrics and recorded traces pour local, interaction-driven measurement (notamment INP).
• PageSpeed Insights — À propos de — PSI combines Lighthouse lab données with CrUX field données; score bands.
• En utilisant CrUX in PageSpeed Insights — how the field section fonctionne and the Réussi/Failed assessment.
• Chrome UX Report (CrUX) overview — the dataset, its role in CWV, and the accès surfaces.
• CrUX methodology — who’s eligible and qui platforms are inclus.
• CrUX API — the 28-day rolling average and the p75 definition.
• CrUX Vis — the visual trend outil que replaced the deprecated Dashboard.

Google — lab vs field & ranking

• Lab and field données differences — Philip Walton on ce que chaque is and qui to prioritize.
• Core Web Vitals outils (web.dev) — Google’s propre overview of the tooling.
• Understanding Core Web Vitals and Recherche Google results — how CWV aligns with ranking systems.
• Understanding page experience dans la recherche Google — the “perfect score… may not be the best use of your time” framing.

Bing / Microsoft

• Bing hasn’t publié guidance positioning ces performances outils as ranking factors; Bing Webmaster Outils surfaces a Core Web Vitals overview, but CWV remains primarily a Google construct. Voir Bing Webmaster Outils aider pour its general guidance.

Quotes from the source

On-the-record statements from Google. Chaque lien is a deep lien que jumps to the quoted passage on the source page.

Google — ce que chaque outil is

• “Lighthouse is an open-source, automated tool to help you improve the quality of web pages.” — Chrome pour Developers docs. Jump to quote
• “PageSpeed Insights (PSI) reports on the user experience of a page on both mobile and desktop devices, and provides suggestions on how that page may be improved.” Jump to quote
• “The Chrome User Experience Report… is a dataset that reflects how real-world Chrome users experience popular destinations on the web.” Jump to quote

Google — lab vs field, and qui Google ranks on

• “Real-user experience data in PSI is powered by the Chrome User Experience Report (CrUX) dataset.” Jump to quote
• “CrUX data is used by Google Search to inform the page experience ranking factor.” Jump to quote
• “If you have both field data and lab data for a given page, field data is what you should use to prioritize your efforts.” — Philip Walton, web.dev. Jump to quote
• “The data in the Chrome UX Report is a 28-day rolling average of aggregated metrics.” Jump to quote
• “trying to get a perfect score just for SEO reasons may not be the best use of your time.” — Recherche Google Central, page-experience docs. Jump to quote

John Mueller, Google (via secondary coverage)

• “It is a ranking factor, and it’s more than a tie-breaker, but it also doesn’t replace relevance.” On Core Web Vitals as a ranking signal. Lire the coverage
• “it’s not going to make your site’s rankings jump up.” Autour the March 2024 page-experience docs mettre à jour. Lire the coverage

Web performances outils — cheat sheet

The three outils En un coup d’œil

Fast facts

• Google ranks on field données (CrUX), pas the lab score. The PSI 0–100 number is pas a ranking input.
• Lighthouse Performances weights (Lighthouse 10, versioned — vérifier avant citing): TBT 30%, LCP 25%, CLS 25%, FCP 10%, Speed Index 10%. Bands: 0–49 red / 50–89 orange / 90–100 green. Lighthouse 13,3+ (May 2026) adds an experimental 5th category, Agentic Browsing (réussir/échouer, pas partie of the 0–100 score).
• Lighthouse can’t mesurer INP — it uses TBT as a proxy. DevTools’ live metrics screen peut observe a réel INP, since you’re generating the interaction.
• CrUX is Chrome-only, opt-in (aucun iOS Chrome, aucun Edge/Safari/Firefox), p75, 28-day window, ~2-day lag → a fix takes ~30 days to montrer.
• Seulement ~11,9% of 43,66M unique Site Audit pages in my January 2022 sample had a CrUX metric; ~21,2% of mesuré pages réussi CWV (my Ahrefs CWV données study).
• CrUX Dashboard deprecated Nov 2025 → utiliser CrUX Vis (cruxvis.withgoogle.com).
• Field pour assessment, lab pour diagnosis. Site-wide → Search Console; individual URLs → PSI; login-gated → Lighthouse in DevTools.

Web-performance measurement mistakes

Reporting a Lighthouse score as Google’s ranking input

Lighthouse is a lab diagnostic. Google uses CrUX field données pour the Core Web Vitals assessment, so étiquette the score pour Ce que c’est.

En utilisant field données to diagnose a line of code

CrUX indique vous que réel utilisateurs have a problem, but pas the exact implementation causer. Reproduce the affected journey in DevTools or Lighthouse après finding the weak field segment.

Comparing unlike tester conditions

A warm desktop run and a cold throttled mobile run réponse différent questions. Garder outil, emplacement, device profile, cache state, and journey consistent pour avant/après comparisons.

Optimizing the aggregate pendant que ignoring page groupes

An origin-level réussir peut hide a slow template, and an origin-level échec peut hide sain pages. Commencer broad, alors segment by URL pattern and experience.

Resolve courant outil disagreements

PageSpeed Insights montre field données pour the origin, pas l’URL

Probable causer: l’URL lacks suffisant eligible CrUX données. Fix: lire the scope étiquette and utiliser origin données seulement as context, alors ajouter RUM or Search Console groupes pour page-level decisions. Confirmer: the report clearly identifies qui scope chaque valeur represents.

Lighthouse changements every temps it runs

Probable causer: lab results vary with network, CPU, cache, server réponse, and third-party activity. Fix: standardize conditions and comparer several runs plutôt que un score. Confirmer: the même opportunity persists à travers the run définir.

Search Console is poor pendant que a local tester is bon

Probable causer: the local lab ne fait pas represent réel devices, locations, or plus long journeys. Fix: utiliser the field report to identifier the affected groupe, alors reproduce que scenario in a diagnostic outil. Confirmer: the lab trace explique the field segment’s bottleneck.

A fix is visible in Lighthouse but pas CrUX

Probable causer: Lighthouse is immediate pendant que CrUX uses a rolling field window. Fix: validate with controlled lab runs and votre propre RUM premier, alors autoriser the public field window to mettre à jour. Confirmer: RUM improves avant the CrUX trend follows.

Testez vos connaissances: web performances outils

Five rapide questions on the three outils and lab vs field données. Pick an réponse pour chaque, alors vérifier.

Journal des modifications