VIDEOS of hundreds of people protesting in Shanghai started to appear on the WeChat application last Saturday night.
Showing chants about removing Covid-19 restrictions and demanding freedom, the videos would stay up for only a few minutes before being censored.
Elliot Wang, a 26-year-old in Beijing, was amazed.
"I started refreshing constantly, saving videos and taking screenshots of what I could before they got censored," said Wang, who only agreed to be quoted using his English name, in fear of government retaliation.
"A lot of my friends were sharing videos of protests in Shanghai. I shared them, too, but they would get taken down quickly."
That Wang was able to glimpse the extraordinary outpouring of grievances highlights the cat-and-mouse game that goes on between millions of Chinese Internet users and the country's gargantuan censorship machine.
Chinese authorities maintain a tight grip on the country's Internet via a complex, multi-layered censorship operation that blocks access to almost all foreign news and social media, and blocks topics and keywords considered politically sensitive or detrimental to the Chinese Communist Party's rule.
Videos of or calls to protest are usually deleted immediately.
But images of protests began to spread on WeChat, a ubiquitous Chinese social networking platform used by more than one billion people, in the wake of a deadly fire on Nov 24 in the northwestern city of Urumqi.
Many suspected that lockdown measures prevented residents from escaping the flames, something the government denies.
The sheer number of unhappy Chinese users who took to the Internet to express their frustration, together with the methods they used to evade censors, led to a brief period in which censors were overwhelmed, according to Han Rongbin, an associate professor at the University of Georgia's International Affairs department.
"It takes censors some time to study what is happening and to add that to their portfolio in terms of censorship, so it's a learning process for the government on how to conduct censorship effectively," Han said.
In 2020, the death from Covid-19 of Li Wenliang, a doctor who was arrested for allegedly spreading rumours following an attempt to alert others to a "SARS-like" virus, sparked widespread outrage and an outpouring of anger against the Chinese censorship system.
Users posted criticism for hours before censors deleted posts.
As censors took down posts on the fire, Chinese Internet users often used humour and metaphor to spread critical messages.
Chinese users started posting images of blank sheets of white paper, said Liu, in a silent reminder of words they weren't allowed to post.
Others posted sarcastic messages like "Good good good sure sure sure right right right yes yes yes", or used Chinese homonyms to evoke calls for President Xi Jinping to resign, such as "shrimp moss", which sounds like the words for "step down", and "banana peel", which has the same initials as Xi's name.
But within days, censors moved to contain images of white paper.
They would have used a range of tools, said Chauncey Jung, a policy analyst who previously worked for several Chinese Internet companies based in Beijing.
Most content censorship is not done by the state, Jung said, but outsourced to content moderation operations at private social media platforms, who use a mix of humans and artificial intelligence.
Some censored posts are not deleted, but may be made visible only to the author or removed from search results. In some cases, posts with sensitive key phrases may be published after review.
A search on the Weibo platform on Thursday for the term "white paper" mostly turned up posts that were critical of the protests, with no images of a single sheet of blank paper, or of people holding white papers at protests.
It's possible to access the global Internet from China by using virtual private networks that disguise Internet traffic, but these systems are illegal and many Chinese Internet users access only the domestic Internet.
Wang does not use a VPN.
"I think I can say for all the mainlanders in my generation that we are really excited. But we're also really disappointed because we can't do anything… They just keep censoring, keep deleting, and even releasing fake accounts to praise the cops," he said.
But the system works well enough to stop many users from ever seeing them.
The writer is from the Associated Press news agency