Does PageSpeed Insights work in China?
Partly. As a global tool for building a fast page it works fine — but PageSpeed Insights has a mainland-China blind spot: it does not measure from inside the mainland, so a passing score tells you almost nothing about whether your site actually works for a visitor there.
Google documents both halves itself. PSI's lab data is "a simulated load of a page on a single device and fixed set of network conditions," run from a Google datacenter it reports as "North America, Europe, or Asia" — none inside mainland China; and its field data is "powered by the Chrome User Experience Report (CrUX) dataset," which counts a user only if they "Sync their browser history," "Enable usage statistic reporting" and "Use a supported platform" — conditions few mainland visitors meet, since Chrome is not the dominant browser there and Google sync is unreliable behind the border. So the questions that actually decide a mainland presence sit outside the tool: a measurement taken from inside the mainland, and the compliance gate — an ICP filing on your public site and PIPL for any China visitor data. The table below maps each, in Google's own words and China's own law.
This is a map of what to check, not a verdict on your site — what matters is where you measure from and which obligations apply to you. Our China team can map both with you →
What actually tells you whether your site works for mainland-China users
| Fact | Primary source |
|---|---|
| PageSpeed Insights measures from Google's own servers, never from inside mainland China. Google documents that PSI's lab data is "a simulated load of a page on a single device and fixed set of network conditions," generated through Lighthouse, and that "PageSpeed will report running in one of: North America, Europe, or Asia" — a Google datacenter, not the mainland. A load simulated from offshore cannot see the cross-border path a real mainland visitor actually travels. | Google — About PageSpeed Insights (developers.google.com), retrieved 2026-10-07 |
| Its "real-user" field data comes from a slice of Chrome users that mainland China barely fills. PSI's field data is "powered by the Chrome User Experience Report (CrUX) dataset," and Google's CrUX methodology counts a user only if they "Sync their browser history," "Enable usage statistic reporting," "Not have a Sync passphrase set" and "Use a supported platform" — which explicitly excludes "Chrome on iOS" and "Other Chromium browsers." In the mainland, where Chrome is not the dominant browser and reaching Google's sync servers is unreliable, few visitors meet all four, so a CrUX reading under-represents your real China audience. | Google — Chrome User Experience Report methodology (developer.chrome.com), retrieved 2026-10-07 |
| The fix is not a higher score — it is a measurement taken from inside the mainland. Because PSI runs from offshore and its field data skips most mainland devices, the only honest way to know what a China visitor experiences is to measure from inside the mainland on a real in-country line. That is a different exercise from a PageSpeed report, and an easy one to get wrong — here is how we do it, including the mistakes that quietly flatter the result. | Google — About PageSpeed Insights (developers.google.com), retrieved 2026-10-07 |
| And the gate a speed score can never read is compliance. Whether you may lawfully serve mainland China turns on two rules no audit tool checks: a public site shown to mainland visitors needs an ICP filing tied to a licensed in-country host (State Council Order No. 292; MIIT Order No. 33), and any China visitor data you collect is personal information whose handling and cross-border transfer PIPL governs (Articles 38–43; in-country storage under Article 40 with Cybersecurity Law Article 39 (formerly Article 37)). A green PageSpeed score with neither in place is a site that looks fine and still cannot lawfully operate. | State Council Order No. 292; MIIT Order No. 33; PIPL Articles 38–43 & 40; Cybersecurity Law Article 39 (formerly Article 37) |
Sources verified by the 21YunBox compliance team on 2026-10-07.
PageSpeed Insights is a genuinely useful tool, and nothing here says stop using it. For building a fast page it does exactly what it promises — anywhere in the world except, quietly, the one place this article is about. The tool has a mainland-China blind spot built into how it measures: it runs from Google’s own servers, and its real-user data is drawn from a narrow slice of Chrome users that mainland China barely fills. So a green score is a true statement about a page measured from outside China — and almost nothing about whether that same page works for someone sitting in Shenzhen. And even a page that genuinely loads fast there can still be a site you are not cleared to operate, because the question that governs a mainland presence was never a speed score to begin with. It is compliance.
That is the half of “does PageSpeed Insights work in China” that no score answers. Below is what Google itself documents about where PSI measures from and whose experience its field data represents — in Google’s own words — and then the compliance gate a measurement tool was never built to read: an ICP filing on your public site, and PIPL for any China visitor data you touch. Read it as a map of what to check, not a verdict on your site.
PageSpeed Insights in China at a glance
| What decides it | In Google's own words — and China's own law |
|---|---|
| What PSI is | A global page-quality tool. Google: it “reports on the user experience of a page on both mobile and desktop devices” and “provides both lab and field data about a page.” Excellent for building a fast page — measured from outside China. |
| Where its lab data runs | Not the mainland. Google: the lab run is “a simulated load of a page on a single device and fixed set of network conditions,” and “PageSpeed will report running in one of: North America, Europe, or Asia.” None of those is inside mainland China. |
| Where its field data comes from | A narrow slice of Chrome users. Field data is “powered by the Chrome User Experience Report (CrUX) dataset,” which counts a user only if they “Sync their browser history,” “Enable usage statistic reporting” and “Use a supported platform” (not “Chrome on iOS,” not “Other Chromium browsers”). Few mainland visitors qualify. |
| Serving the public in China | A public site shown to mainland visitors needs an ICP filing tied to a licensed in-country host (State Council Order No. 292; MIIT Order No. 33). No audit score grants that filing, and none reads whether you hold it. |
| China visitor data | A mainland visitor's IP and identifiers are personal information. Collecting it, and moving it across the border, is governed by PIPL (Articles 38–43; in-country storage under Article 40, with Cybersecurity Law Article 39 (formerly Article 37)). |
| What actually tells you | A measurement taken from inside the mainland, on a real in-country line — plus the two compliance doors above. That is the question a PageSpeed score leaves open. |
What PageSpeed Insights measures — and from where
PageSpeed Insights reports two kinds of number, and Google is precise about where each comes from. The lab half is “a simulated load of a page on a single device and fixed set of network conditions,” generated through Lighthouse in a controlled environment. That control is exactly what makes it useful for debugging — and exactly what makes it silent about China, because the simulation is run from a Google datacenter that, in Google’s words, “will report running in one of: North America, Europe, or Asia.” Every one of those vantage points sits outside the mainland, on the wrong side of the border from your China visitors.
The field half looks more promising, because it is labelled real-user data. But Google documents its source too: it is “powered by the Chrome User Experience Report (CrUX) dataset.” And CrUX is not every real user — it is the users Google can see. Its methodology counts a person only if they “Sync their browser history,” “Enable usage statistic reporting,” “Not have a Sync passphrase set” and “Use a supported platform,” and it names the platforms it leaves out, among them “Chrome on iOS” and “Other Chromium browsers.” So both halves of PSI measure from outside the mainland’s actual conditions: one by simulating from offshore, the other by sampling a population mainland China mostly isn’t in.
Why the mainland is PSI’s blind spot
Put those two facts next to how people actually browse in China and the gap becomes structural rather than incidental. Chrome is not the dominant browser on mainland devices — a large share of traffic runs through domestic browsers and in-app webviews that never contribute to CrUX at all — and the handful of Chrome users who remain can only feed the dataset if their browser is syncing history to Google and reporting usage statistics, both of which depend on reaching Google infrastructure that is unreliable from behind the border. The result is a field sample that thins out precisely where you need it most. Meanwhile the lab result, run from North America, Europe, or Asia, measures a route your mainland visitor never takes.
This is why a passing PageSpeed score and a broken China experience can coexist without contradiction: the tool is answering a question about a page as seen from outside China, and you are asking a question about a visitor inside it. The honest way to close that gap is not to chase a greener score — it is to measure from inside the mainland, on a real in-country connection. That is a different discipline with its own traps, and we have written up how we do it, including the measurement mistakes that quietly flatter the result.
The gate a score can’t read: ICP and PIPL
Even a site that measures fast from inside the mainland still has to clear a gate no performance tool touches. Serving a public website to people in mainland China turns on an ICP filing (备案): under State Council Order No. 292 and MIIT Order No. 33, that filing attaches to a hosting resource physically inside the mainland and to a licensed in-country entity. A PageSpeed report neither grants it nor notices its absence — the score will look identical whether or not you are cleared to operate.
The second door is data. A mainland visitor’s IP address and identifiers are personal information under China’s Personal Information Protection Law, and once you collect it, how you handle and move it is governed law, not a setting. Carrying that data out of the mainland to an offshore origin is a cross-border transfer PIPL regulates through Chapter III (Articles 38–43: notice, separate consent, and one permitted transfer mechanism), and for a critical information infrastructure operator or a large-volume handler, Article 40 — with Cybersecurity Law Article 39 (formerly Article 37) — adds a duty to store it inside China. None of that shows up in a speed audit, because it was never a speed question.
This is a map of what to check, not a verdict on your site — which doors apply turns on your entity, your data and who your visitors are, and it is worth settling with counsel rather than inferring from a dashboard.
Where 21YunBox fits — a compliant overlay, not a migration
A measurement tool can tell you a page is fast; it cannot make that page lawful to serve, and it cannot show you what a real visitor in the mainland experiences. That is the work we do. Our China team measures your site from inside the mainland on real in-country lines — the vantage point PageSpeed Insights structurally cannot take — and maps the two doors a score never reads: whether your public site needs an ICP filing, and how PIPL treats the China visitor data you collect. Where an in-country presence is the answer, we add ICP-filed delivery from inside the mainland, in front of your existing origin — no rebuild, no second codebase, and no move off the stack you already run. You keep PageSpeed Insights for what it is good at; we cover the half of the China question it was never built to answer.
Related reading:
