How do you test and speed up your website in China?

A site can feel instant in your office and take a long time in Shanghai. Most advice skips straight to buying a CDN, usually from the company giving the advice.

Measure first. The result tells you which fix is worth your money. This guide is part of our work on putting your website in China. It assumes your site loads at all.

If you aren’t sure, check whether it’s accessible or find out if it’s blocked or just slow.

How do you test your website’s speed from China?

Test from inside the mainland, from several cities and carriers, at more than one time of day, and repeat each test. One run from one node tells you very little.

Here’s a method you can reuse, with or without a developer.

1. Pick 3 to 5 key pages. Use the pages Chinese visitors will land on, not only your home page.
2. Test on more than one carrier. Filter by China Telecom, China Mobile and China Unicom where the tool allows it.⁷
3. Test in the morning and the evening. Slowdowns follow a daily pattern.¹ Dotcom-Monitor names 8 to 10 a.m. and 5 to 10 p.m. in Shanghai as peak hours,⁴ and Tencent Cloud names 8 to 11 p.m. Beijing time.²⁸ Both are vendor statements, so cover both windows.
4. Run each test at least 3 times. Keep the median, and write down the worst run.
5. Run the same test from your home market. That’s your baseline for reading the gap.
6. Save the waterfall. It shows which request cost the time, and a score doesn’t.

Which tools can test speed from China?

No study compares these tools, and none replaces a real visitor. Choose by what you need to see. Our accessibility guide covers the tools for checking whether a site opens at all.

Tool	Where it tests from	What you get	Limits
WebPageTest	Beijing and Shanghai in a November 2025 Catchpoint list 8	Waterfall 29, TTFB 2	We couldn’t confirm these agents are live today, so check its location list first
17CE	Not stated on the page we read	Analysis by region, ISP and test point 6	Metrics and free limits aren’t listed there 6
BOCE	Filter by carrier, and by data center or home node 7	Average response speed 7	One node by default, daily guest quota 7
ITDog	Not verified	HTTP test, TCPing, traceroute, per a third-party article 5	Described secondhand 5
Dotcom-Tools	Shanghai 4	Free, no sign-up. Its published test shows first and repeat visits with waterfalls 4	Per-element analysis needs the paid platform 4
Speed in China, WordPress plugin	Beijing, Shenzhen, Guangzhou, Shanghai, Qingdao 10	DNS, TCP, TLS, TTFB, total load time 10	One free test every 30 days, sends your URL to its API and Globalping 10
Chinafy Global Speed Test	18 servers, including Beijing, Shanghai and Guangzhou 11	Load time comparison by location 11	The vendor sells the fix

How do you read the results?

Split the load into stages, then compare each stage with your home market. The stage that grows most is where to look first. A single score hides that.

Website speed tests usually split a request into DNS, TCP connection, TLS handshake, redirects and server response.⁵

The article we cite gives no thresholds, and we found none we’d trust. The reading below is our own inference from what each stage does.

Stage that stands out	What it suggests	Where to look first
DNS lookup	Slow resolution, or too many hosts	Your DNS provider, the number of outside domains
TCP connect and TLS	A long or lossy route	Server location
Redirects	Extra round trips across the border	Chains such as http to https to www
Time to first byte, China only	The route	Server location or edge cache
Time to first byte, everywhere	Your server	Backend, page cache
One long bar in the waterfall	A resource that hangs	That domain
Many medium bars	Too much page	Weight and number of requests

Matomo’s documentation makes a useful point here: its server time measure includes network latency, so it isn’t pure server processing.²⁶

The baseline from your home market is what separates the two.

As a target, Google rates Largest Contentful Paint as good at 2.5 seconds or less, and poor above 4 seconds, at the 75th percentile of page views.¹²

These are global thresholds, not a rule for China or for Baidu.

How slow is normal for a foreign site in China?

Slow, but no published number deserves much weight. Everything below is one project or one vendor, and methods differ.

Source	What was measured	From China	Comparison
Chinafy, 2025-26 2	Load time, labeled median or average depending on the section	15.3 s, Beijing	7.4 s Virginia, 4.7 s London
Chinafy, 2025-26 2	Time to first byte, basis not stated	1.4 s, Beijing	0.35 s Virginia, 0.31 s London
Thoughtworks, 2022 3	First Contentful Paint, one client’s site	2.27 times longer	EU
Thoughtworks, 2022 3	Page onload, one client’s site	10.07 times longer	EU
Dotcom-Monitor, date of test not stated 4	TED.com, first visit average, local peak hours	19.0 s, Shanghai	6.2 s New York

The business statistics around this topic are weaker still.

• Chinafy says 90% of non-Chinese sites take 5 seconds or more, quoting Mlytics.²⁹ Mlytics gives no source for it.¹⁸
• Chinafy cites a 7% conversion loss for a 1-second delay, crediting Kissmetrics.²⁹ Mlytics, citing Akamai, gives 7% for 100 milliseconds.¹⁸
• Both repeat the idea that visitors expect 2 seconds, and Mlytics gives no source for that either.¹⁸

Treat these as talking points. We’d rather you measure your own pages.

What actually slows a site down there?

Two things: the road into China, and what your page waits for. You can only control the second one fully.

The road is the slower-moving problem. A 2020 study found slowdowns on 79% of measured pairs, and more than 70% of pairs fell below 1 Mbps for over 5 hours a day.¹

Packet loss applied to traffic entering China.¹

Don’t blame the firewall for all of it. In a one-day side test, only 34.45% of bottleneck hops coincided with it, though the authors measured only nodes that inject resets.¹

More on that in what the Great Firewall means for your website.

The second problem is waiting. A resource that can’t load can hold the page while the browser keeps trying.

AppInChina tested a lightweight blog with Chrome DevTools on a mainland network, in an article dated April 2025. Google Analytics calls failed after 3 minutes 9 seconds, DOMContentLoaded came at 1 minute 3 seconds, and the last request finished at 5 minutes 3 seconds.¹⁷

Chinafy says a browser will often wait, sometimes for 30 seconds or more.¹⁹ Both are vendors. AppInChina’s result is one page, and Chinafy gives no method.

Free check

Is your website visible on Baidu?

Enter your website. We check Baidu, access from China and how your site is set up, then email you the report.

Your websiteCheck my site

How do you speed up your site, and in what order?

Start with what’s cheap and under your control, then move to what costs money or needs a local setup. Each step below can change whether you need the next one.

1. 1Measure
2. 2Clear the waits
3. 3Lighten the page
4. 4Move closer without an ICP
5. 5Mainland hosting or CDN with an ICP

Retest after each step.

Fix	What it solves	ICP needed?	Evidence
Clear the waits	Pages stuck on outside resources	No	One vendor test, plus web.dev guidance 1517
Lighten the page	Bytes and requests on a slow route	No	Reasoning, no China-specific test found
Host in Hong Kong	Fewer hours of slowdown	No 31, but check the public security filing	2020 study, nothing recent 1
Edge cache outside the mainland	Distance, in part	No, on Alibaba Cloud’s and Tencent Cloud’s non-mainland zones 2032	No independent test found
Mainland hosting or CDN	Distance and the border	Yes 213132	Vendors, who disagree 222

We found no independent before-and-after test for any of these. The gain figures in circulation come from companies selling the fix, so we don’t repeat them.

1. Clear the waits

List every outside domain your page loads, using the method in our guide to Google services in China. Then test what happens when each one fails.

• To see what breaks: in Chrome DevTools, right-click a request in the Network panel and choose Block request.¹³
• To see what a slow resource costs: use Throttle request in the same menu.¹³
• To see what a hang costs: use WebPageTest’s Hosts to fail setting, known as SPOF. It reroutes the hosts you list to a server that silently drops all requests.¹⁴

The last one matters most. It’s the closest thing to a resource that hangs instead of failing cleanly.

Then change how those resources load.

• Scripts: synchronous scripts delay DOM construction and rendering, so use async or defer unless a script must run first.¹⁵
• Your own copies: serve fonts and libraries from your domain. The tradeoff is that self-hosted scripts can go out of date.¹⁵
• Fonts: font-display: swap gives a font an extremely small block period, so visitors see fallback text instead of nothing.¹⁶ Our guide to fonts and embeds has the rest.
• Anything optional: load it after the main content, or remove it. Removal is web.dev’s first option for a script that slows your page.¹⁵

2. Lighten the page

We found no serious China-specific study on images, CSS or JavaScript. This step rests on reasoning, not on a test.

The reasoning is simple. If packet loss hits traffic entering China,¹ every byte you send travels the troubled direction, so fewer bytes and fewer requests should help more there than at home.

• Compress images, and size them for the slot they fill.
• Replace autoplay video with an image and a link.
• Cut scripts and styles nobody uses on that template.
• Set long cache headers on static files.

The last point has some evidence behind it. Dotcom-Monitor’s repeat visits from Shanghai averaged 5.1 seconds, against 19.0 for first visits.⁴

Caching helps returning visitors, though, not someone arriving from Baidu for the first time.

3. Move closer without an ICP

Hong Kong is the one location the 2020 study singles out. Hong Kong averaged about 3 hours of daily slowdown into the mainland, while most mainland receivers averaged 5 to 17 hours across all senders.¹ We found no recent independent measure to update that.

See Hong Kong or mainland hosting.

Some CDNs offer a middle option. Alibaba Cloud’s “Global (Excluding the Chinese Mainland)” zone needs no ICP, and it serves mainland users from Japan, Singapore and Hong Kong.²⁰ Tencent Cloud’s “Outside Chinese mainland” region needs no ICP either.³²

None of this fixes your dependencies. Chinafy’s line on hosting is fair: it “only solves for proximity.”¹⁹

4. Go onshore, if the numbers justify it

Serving from the mainland needs an ICP, and a foreign company needs a local subsidiary or a licensed local partner to get one.³¹ Our guide to the ICP license explains who qualifies, and a local adviser should confirm your case.

The requirements can be substantial.

Cloudflare’s China Network needs an Enterprise plan, a separate subscription, an ICP for each apex domain, a content review by JD Cloud and your ICP number in the footer.²¹

Compare the options in hosting in China and CDNs and Cloudflare in China.

Does speed affect your Baidu rankings?

Possibly, but we can’t show you how much. We read no official Baidu page that sets a current speed threshold or says what weight speed carries.

Third-party sites reported a 2017 Baidu algorithm called Lightning, aimed at mobile home pages, with time bands at 2 and 3 seconds.²³

We haven’t read Baidu’s original, and we wouldn’t treat those numbers as a current rule.

The safer argument is simpler: a page that doesn’t finish loading is a poor page for a buyer and for a crawler. See Baidu ranking factors for the rest of the picture.

How do you keep measuring real visitors?

Collect timings from real Chinese visitors with a tool you host yourself. Test nodes tell you what’s possible, and visitor data tells you what happens on real phones and networks.

Skip Google Analytics for this, since its data from China is unreliable.¹⁷ Two self-hosted options:

• web-vitals: a small Google library, about 3K compressed, that measures LCP, INP, CLS, FCP and TTFB on real users.²⁵ Its docs recommend installing it through npm or self-hosting it, and include an example that posts to your own endpoint.²⁵
• Matomo: its page performance report separates network, server, transfer and DOM processing time.²⁶ Matomo says a copy on your own servers can track Chinese visitors, because your Matomo URL “won’t be affected by the Great Firewall.”²⁷ That’s its general claim, so test your own address from the mainland.

Before you collect anything from Chinese visitors, read whether China’s privacy law applies to your website and check with a local adviser.

To put this alongside your other launch checks, use the China-ready website checklist.

The bottom line

Measure from the mainland, several times, and read the waterfall before you spend anything. A page that hangs on one blocked script won’t be saved by a better server.

Clear the waits, lighten the page, move closer, and go onshore only when your Chinese results justify a local setup and an ICP. Then keep measuring, because no single fix lasts.

How Index in China can help

We localize your site into Simplified Chinese, with a person reviewing every page, and index it on Chinese search engines. We don’t guarantee traffic.

To go through your site, your market and what getting indexed would take, book a 30-minute call.

Book a 30-minute call

Monthly report BaiduIndexed Bing ChinaIndexed SogouIndexed