Methodology

PageSpeed Insights measures. We diagnose and fix.

The pipeline behind every report: Google's own data collected, every flag it raises put on trial, the Silent Bottleneck hunted in your field data, fixes written in your platform's vocabulary — and every prediction held to account. Delivered in less than 24 hours — the full method, end to end, nothing skipped — instead of an agency's 2–4 weeks.

  1. Data collection — both instruments, both device classes

    Your URL is measured through Google's PageSpeed Insights infrastructure for mobile and desktop: full Lighthouse lab runs plus the Chrome UX Report — what real visitors experienced over the last 28 days, including LCP, CLS and INP at the 75th percentile. Lab and field are different instruments answering different questions; the report never conflates them. The input itself isn't exempt from scrutiny: when the measurement warrants it, an integrity gate — the same user-agent-flip test as our free cloaking test, same thresholds — checks that Lighthouse was served the page your users actually get; a run that fails it is re-based, never trusted.

  2. Evidence grounding — your real page, not theory

    The auditor fetches your page and the resources behind every hypothesis, and ties each finding to a verbatim artifact: the actual image tag, the actual response header, the actual byte count — with a documented attempt to falsify the finding before it is asserted, and a confidence grade on the record. No artifact, no verdict.

  3. Adjudication — every flag put on trial

    Every warning Google raises leaves with a verdict — Real, Wrong, Trivial or Locked — with the evidence quoted, and the report states its Noise Rate: the share of warnings owed no engineering time, each dismissal carrying its proof. Then the Silent Bottleneck hunt: decomposing your field metrics to find the constraint that produced zero flags — the number-one problem often lives exactly there.

  4. Root causes and platform-native fixes

    Real flags — plus up to five deep findings PSI can't produce — get causal chains with the culprit named, and fixes written in your stack's own mechanism: the theme.liquid snippet on Shopify, the cart-fragments rule on WooCommerce, the Varnish-and-production-mode checklist on Magento, the next.config.js change on Next.js, the exact plugin setting on WordPress, the corrected <head> block verbatim — deep commerce modules for Shopify, WooCommerce, BigCommerce, Magento / Adobe Commerce, PrestaShop and Salesforce Commerce Cloud, each fix aimed at the lever the merchant controls on that platform — plus an edge apply pack when you're on Cloudflare, Akamai, Fastly, Bunny, Azure Front Door, AWS or Google Cloud. Several findings come from probes PSI never runs — your CDN's live configuration, per-asset compression and cache headers, your platform's plugin and app census, your font files' own metrics, your DNS and TLS connection setup, the compilation era of your JavaScript bundles — so they appear in no export of the free report. Ranked by impact per effort, with impact stated as ranges and the arithmetic shown — never fake point precision. What arrives as a wall of warnings leaves as a short list of fixes worth shipping, and documented permission to ignore the rest.

  5. Three deliverables

    • The report — self-contained HTML, printable, every verdict and every fix with quoted evidence.
    • The executive summary — one page, plain language, impact sized honestly.
    • The JSON — every finding machine-readable via the REST API.

  6. Verification — predict, prove, confirm

    Every fix ships with a predicted range. Your purchase includes two re-runs of the same URL within 30 days of the first audit: ship the fixes, re-run, and hold the report to its own predictions — while your real-user data confirms the movement on its own 28-day clock. In our public calibration pilot, realized savings ran 1.0–1.9× the prediction.

That is the pipeline. The canonical statement of the method — the evidence rules, the Four Verdicts, the worked example and the verification ladder — is our Noise & Silence framework.

Try the free tools, then get the full audit.

Explore the free tools Get your audit