How the Xi Jinping Health Rumour Travelled

In four days, an awkward moment at a summit in New Delhi became a story about a stroke, an intensive care unit and a looming civil war in China. Tracking thousands of posts across ten platforms and 17 languages shows how the rumour was built, why it crossed from one audience and platform to the next, and why the corrections struggled to catch it.

A real moment, and no explanation

On 12 September, at the opening session of the BRICS summit in New Delhi, Indian Prime Minister Narendra Modi paused his speech, turned toward Xi Jinping’s interpreter and asked, twice, “Is it OK?” That evening Xi skipped the gala dinner Modi hosted for visiting leaders. Indian broadcasters and wire services reported both moments within hours. None of those reports claimed Xi was unwell.

Rumours tend to form where a question is both important and unanswered, and this one was both. The health of China’s leader matters to markets, to Taiwan and to the Trump–Xi state visit scheduled for Washington eleven days later. Beijing discloses almost nothing about its leaders’ health, and it offered no explanation for the pause on stage or the empty seat at dinner. People supplied their own.

The subject had been quiet. The two weeks before the summit produced just 2.5% of the month’s posts about Xi’s health. The single day of 17 September produced 29%.

A seed built to travel

The first post in our data asserting that Xi was ill came from a Chinese-language account on the morning of 13 September. The story took its lasting shape late that day, when Wanjun Xie, chairman of the exiled China Democracy Party, posted that Xi had reportedly fainted at the summit, been treated by his medical team and cut his visit short.

Over the next two days Xie added detail in stages. On 15 September he cited “witnesses” who said Xi had been carried off his plane in Beijing and flown by military helicopter to 301 Hospital, the military hospital that treats China’s senior leaders. Early on 16 September he posted three messages in 27 minutes: Xi had suffered a severe ischaemic stroke; he had refused treatment in India to keep his medical information from Indian hands; he was in intensive care and not out of danger; and China’s military theatre commands had gone on alert because his death could bring civil war.

Three features made these posts travel. The first is the hedge. Almost every post opened with “It is rumored” or “It is said”, which let readers pass the claim on without vouching for it. The second is specificity. A named hospital, a helicopter, a type of stroke and an ICU give a story the texture of insider reporting, and none of it can be checked from outside China. The third is stakes. Linking a medical rumour to civil war turned a curiosity about one man into a question about the stability of a nuclear power.

Xie’s two accounts, one writing in Chinese and one in English, published 26 posts on the subject. Those posts drew 19.6% of all engagement in the conversation. His 16 September stroke post alone drew 11.2%, more than any other post we collected.

Translation was the first jump

The rumour began in Chinese and gained most of its reach in English. Xie did the first translation himself: his English-language account posted its own version of most claims, usually within hours of the Chinese original. After the stroke post, the claim appeared in English within 16 minutes, in Japanese within two hours, in Portuguese within four and in Spanish within five.

Chinese-language posts made up 14.5% of the conversation but drew 29.8% of its engagement. The audience for the original was already primed: on the one platform where we mapped reposts, 49% of the reposts of health-related posts started inside a single community of Chinese dissident, diaspora and overseas Chinese-language media accounts. English-speaking audiences received the story second-hand, often through screenshots and machine translations, and many of the English versions dropped the hedge. “It is rumored that Xi fainted” became “Xi suffered a major stroke”.

Each audience rewrote the story for itself

When a rumour crosses into a new community, it adapts to what that community already cares about. This one changed at every crossing. US conservative influencers framed it around the coming Washington summit and weakness in Beijing; one with more than four million followers described “an apparent health emergency”, and his wording was copied by 24 accounts on five platforms. A Portuguese-language account added that China’s leadership suspected Xi had been poisoned in India, and posts blaming Indian food, some joking and some not, followed. Others folded in South African President Cyril Ramaphosa, whose office did confirm he had fallen ill after the summit, and concluded that several BRICS leaders had been struck down. Japanese commentators asked what an incapacitated Xi would mean for a crisis over Taiwan.

Through all of this the core claim held steady. The stroke and 301 Hospital story appeared in 45.8% of posts and drew 55.7% of engagement. The civil-war and succession framing appeared in 7.0% of posts and drew 11.1% of engagement. The frame around the claim changed with each audience, so each new audience received a version that answered a question it already had.

The core claim stayed fixed. The frame around it changed with each audience it reached.

Each kind of platform played a different part

The stroke claim reached all ten platforms in our data within 21 hours of Xie’s post. It was on a short-video app in under three hours and on the largest social network in just over three. Messaging channels, alternative social networks and anonymous forums had it within seven hours, and the remaining platforms by the end of the day. Across languages the jump was faster still.

The main social media platform worked as the switchboard. It carried 62.9% of posts and 70.4% of engagement, and the stroke claim surfaced there before it appeared anywhere else. Visual platforms did something different: they circulated the evidence. On one photo- and video-sharing platform, which carried 3.5% of posts but 11.2% of engagement, nearly half the interactions went to broadcasters’ clips of Modi’s pause. Viewers could watch the moment and supply the explanation themselves. On the leading short-video app, which carried 3.8% of posts and 7.6% of engagement, most of the reach went to videos asserting the illness outright.

Alternative social networks gave the rumour volume without reach. Three of them together carried 9.6% of posts but 0.3% of engagement. Much of this was cross-posting: about a quarter of the text on two of those networks, and nearly two-thirds of the text in messaging channels, also appeared word for word on the main microblogging platform. These copies matter less for spreading a rumour than for keeping it alive, because they remain in circulation after the original has been corrected. On anonymous forums the story became a running joke, which carried it to readers who would never have shared it straight.

Why the corrections struggled

By the standards of official communications, the corrections came quickly. A crowd-sourced context note was attached to Xie’s stroke post on the afternoon of 16 September. India’s Ministry of External Affairs issued a “fake alert” in the early hours of 17 September, calling the claims about sick leaders malicious. The same day, Russian state television aired footage suggesting that Modi’s question was prompted by Xi’s interpretation earpiece failing, and Chinese state media continued to publish letters and speeches in Xi’s name.

By the standards of social media, they came after the peak. Posts rebutting the claims made up 19.1% of the conversation but drew only 12.4% of its engagement. On average, a post asserting the illness drew 1.6 times the interactions of a post rebutting it. Corrections carry a built-in handicap: to rebut a rumour they must repeat it, and the answer they offer is duller than the question.

A large share of the correcting was done by software. An AI chatbot built into one of the major platforms wrote 175 posts on the topic, 8.4% of the entire conversation. Of those, 134 were rebuttals, posted as replies to users who had tagged the bot to ask whether the claims were true. The chatbot’s 175 posts drew barely 400 interactions between them. Automated fact-checking now sits inside the rumour’s own threads, where few people see it.

A rumour built to survive evidence

The hardest test came when Xi flew to Washington on 23 September for a three-day state visit. A man recovering from a severe stroke in intensive care should not be able to walk down aircraft stairs and stand through a military flyover. The rumour adjusted instead. On 17 September Xie described the rebuttals as a propaganda push by Beijing’s paid online commentators. On 21 September he claimed Xi faced a “33% chance” of another stroke if he travelled. When Xi descended the stairs at Joint Base Andrews holding his wife’s hand, Xie presented that as proof: Peng Liyuan’s support, his English account wrote, “confirms that Xi suffered a stroke in India”.

Posts treating the visit as a test of Xi’s health peaked on 24 September, the day after he landed, and nearly half of them were in Chinese. The pattern also predates the summit. On 30 August, when Xi arrived in Bishkek, Xie’s English account described him as appearing “severely swollen, pale, and unsteady on his feet”. A rumour that reads every new image as confirmation cannot be ended by images alone.

The rumour reopened the succession question

Every scare about Xi’s health is also a question about who would replace him, and this one brought that question to the surface. We found 450 posts in the month discussing who might succeed Xi or how long he will rule. Half were written in Chinese and 40% in English. The conversation moved through three distinct phases.

Before the summit, succession was a slow-burning debate about Xi staying. The two weeks from 28 August to 11 September produced 28% of the month’s succession posts. The dominant theme was continuity: Xi securing a fourth term at the 21st Party Congress in 2027, what that would mean for Taiwan, and whether the removal of senior generals, including Central Military Commission vice-chairman Zhang Youxia, in late August pointed to a purge or a power struggle. Health came up in a quarter of these posts. Chinese-language dissident and diaspora media and anti-Communist Party commentators wrote two-thirds of them.

The health rumour turned it into an emergency. Between 12 and 22 September the daily pace of succession posts more than doubled (up 107%). More than half of these posts (57%) invoked Xi’s health, and the phase drew 70% of all the engagement succession talk received in the month. The most-shared framing came from Xie himself: if Xi died or could no longer govern, China faced civil war. A veteran US national-security correspondent relayed the claim that rival factions were mobilising, and partisan US pages turned that into headlines. Posts naming a caretaker or successor appeared, and posts explaining that China has no designated successor and no settled mechanism for choosing one rose by about 460% on a daily basis compared with the weeks before the summit. Individual accounts, most of them writing in English, became the largest group of participants, with 40% of posts in the phase.

Xi’s visit to Washington ended the emergency, but not the conversation. From 23 to 27 September the daily pace rose by a further 56%, to more than three times the pre-summit rate and the busiest stretch of the month, yet engagement fell below a tenth of the previous phase. The themes returned to palace politics: claims that party elders would force Xi out at the Central Committee plenum due on 26–29 October, that his daughter was being positioned as an heir, and renewed arguments about a fourth term. Health still featured in 37% of posts, now recast as lasting frailty after a stroke.

The participants were a narrow group. Chinese-language anti-Communist Party commentators and dissident and diaspora media wrote about half of all succession posts. Xie wrote 2.2% of them but drew 29.2% of the engagement; a handful of analysts and journalists wrote 1.8% and drew 26.4%. Mainstream news outlets wrote 3.8% of succession posts and drew 0.6% of the engagement. Accounts affiliated with the Chinese state were almost silent, accounting for 0.4% of succession posts. Succession talk lives among China-watchers and the diaspora, and it reaches a wider public only when a health scare makes it urgent. The plenum at the end of October is the next obvious trigger.

Organic spread, with small pockets of coordination

A rumour that travels this far invites the suspicion that someone organised it. We tested for the usual signatures of coordinated inauthentic behaviour: identical posts from different accounts within minutes, clusters of accounts reposting one another, and concentrations of auto-generated handles. We found three small clusters of accounts posting identical text seconds or minutes apart, on both sides of the story. One cross-posted Chinese dissident-media programmes that carried health claims. Another, a set of four pro-Beijing pages, declared the rumours dead within seconds of each other. None drew meaningful engagement.

The accounts that reposted Xie showed no sign of automation. They did not repost one another, and auto-generated handles were no more common among them than among the audiences of comparable accounts. The reach came from real users and real influencers: ten accounts produced 45% of all engagement. For anyone responsible for responding, this is the harder finding. There was no network to expose or take down. The rumour ran on attention.