Contextualizing critical disinformation during the 2023 Voice referendum on WeChat: Manipulating knowledge gaps and whitewashing Indigenous rights

---

Implications

This study reveals how subtle forms of disinformation were spread on WeChat to Australian audiences that speak Chinese during Australia’s 2023 referendum on a constitutional change to create an Indigenous Voice advisory body in Parliament. Our work is situated within Critical Disinformation Studies (Kuo & Marwick, 2021), which moves beyond defining disinformation in terms of fact, intent, and harm. While these components remain important considerations, they risk narrowing the scope of analysis in ways that leave other forms of disinformation unexamined and thus outside regulatory scrutiny and enforcement, where they may cause greater harm. Aligning with the critical disinformation perspective, we also frame critical disinformation as a set of media strategies that blend facts and falsehoods to reinforce colonialism, manifested in existing racial hierarchies and underpinning the very ideological architecture of Australian institutions and national media in this research context. This critical and reflective stance, as opposed to claiming objective truth, allows our work to contextualize how the media discourse on the Constitutional referendum was weaponized to mobilize one group of marginal communities (Chinese-speaking Australians on WeChat) to harm another (Aboriginal and Torres Strait Islander peoples).

The October 14, 2023, Australian Indigenous Voice referendum proposed adding a purely advisory body for Aboriginal and Torres Strait Islanders, empowered to comment on matters before the Parliament and the executive branch of government. The constitutional change required a double majority: support from most Australian voters and from majorities in at least four of six states¹Australia comprises six federated states—New South Wales, Queensland, South Australia, Tasmania, Victoria, and Western Australia—and ten federated territories. Of the three internal territories, the Australian Capital Territory (ACT) and the Northern Territory (NT) have their own legislatures, while the Jervis Bay Territory does not. In the Voice referendum, votes from the ACT and NT counted toward the national majority but not the required majority of states, creating unequal voting power across jurisdictions (Vivian, 2023). (Parliamentary Education Office, n.d.). The proposal failed on both counts. While advocates emphasized the Voice as a modest institutional reform to enable Aboriginal and Torres Strait Islander peoples’ input into policy, opponents mobilized doubts through discourses of uncertainty, cost, ineffectiveness, and alleged constitutional risk.

Rather than relying solely on overt falsehoods, campaigns opposing the Voice referendum on WeChat effectively used culturally relevant messaging, coded language aimed primarily at their target community, and strategic ambiguity to discredit a policy that started out with a 65% approval rating (Graham, 2024). Of note, WeChat’s Chinese-language Australian news translations, ads, and comments are largely out of the regulatory frameworks that govern English-language media in Australia. Existing Australian government regulations primarily focus on major English-language publishers—such as broadcast media and U.S.-owned major digital platforms that offer a limited degree of transparency to regulators and researchers—and their influence on electoral or referendum processes. Content producers on WeChat are effectively exempt from these rules due to the platform’s digital and non-English nature (Yang et al., 2024). On WeChat, content that does not touch on Beijing’s core interests generally escapes strict enforcement of platform governance—whether through manual reporting or algorithmic censorship—even when it contains problematic or misleading information—content that, paradoxically, drives much of the platform’s economy. Moderation of such content largely relies on a self-reporting mechanism by users, through which platform moderators evaluate reports before deciding whether to remove the content. The critical disinformation environment examined in this paper largely thrives within the regulatory gap between WeChat’s platform governance enforcement and Australia’s media regulatory environment.

Suffice it to say, making domestic disinformation in WeChat visible here demonstrates how second-language media are not given the same considerations as English media in Australia at present. This disparity remains critical in this nation where, between 2016 and 2021, 5.8 million people (22.8%) reported speaking a language other than English at home—a figure projected to rise in the upcoming census as immigration continues to outpace natural population growth. In 2021, nearly 3.9% of Australians spoke Mandarin or Cantonese at home, representing the largest non-English language group. Of these, 25.9% of Mandarin speakers and 23.7% of Cantonese speakers reported low English proficiency (Australian Bureau of Statistics, 2022), suggesting that Chinese-language media likely serves as their primary source of information.

The follow-on key implication is that traditional approaches to fact-checking and disinformation regulation, typically centered around platforms accessible to English-speaking majorities, premised on binaries of truth and falsehood, and centered on intention and harm, are insufficient for addressing critical disinformation within culturally, linguistically, and infrastructurally distinct and sophisticated media ecosystems. Our analysis demonstrates that, in the Australian context, existing regulatory frameworks are not focused on information sources for those that gather what they need to know in languages other than English due to perceived inaccessibility—whether stemming from platform infrastructure or language barriers for English-speaking regulators and researchers. Yet, there is no easy fix.

Designing disinformation policy is hard—it runs up against freedom of speech protections and the competing interests of regulators, technology companies, and political actors, including parties, interest groups, and social movements across a diverse range of socio-historical and common law traditions. These conflicts were exemplified in Australia’s Combatting Online Misinformation and Disinformation Bill that was to take effect for the 2025 federal election and measure disinformation based on truth and falsehood, intention, and severity of harm, but was withdrawn in November 2024. However, even if enacted, the reach of this bill would likely fall short in addressing platforms that are technologically and linguistically inaccessible to English-speakers—such as those used by marginalized communities, or those operating through encrypted channels or within the dark and grey web. Consequently, misleading content remains largely unchecked, particularly when delivered in culturally nuanced ways that evade straightforward classifications of misinformation or disinformation.

It is important to note that our claims of disinformation build from what Kuo and Marwick (2021) suggest is critical disinformation, which situates the production and normalization of disinformation within broader histories, cultures, and politics, and centers questions of social (in)justice/(in)equality in relation to historical power dynamics. Specific to the Australian context, we show the difficulty of policing and countering disinformation from Australia’s colonial histories and the socio-political positioning of Indigenous peoples and migrants of color, compared to the predominant white Australian society. Along the lines of critical disinformation studies, we define disinformation in our research context as media strategies that mix truths and falsehoods to reinforce racial hierarchies, perpetuate colonial narratives, and sometimes advance the interests of institutions and national media rooted within a colonial ideological architecture. Our data show local campaigns strategically deploy these types of critical disinformation in attempts to persuade and cement (colonial) power structures. Further, in some instances, the amplification of these messages might not be organic, based on volume and sentiment of engagement data that is outside historical norms for media accounts on WeChat. Nevertheless, this speech escapes regulation in part because platforms like WeChat are by law subject to legal oversight; they havefunctionally avoided classification by regulators due to their enclosed nature, lack of interfacing with Australian regulators, and limited algorithmic transparency that American counterparts may provide, and the content resists standard fact-checking frameworks.

The implication of framing disinformation in this critical way is allowing policymakers to look beyond external national security concerns and consider how domestic narratives also shape public understanding and political outcomes. As Kuo and Marwick (2021) argue, selective focus on national interest confines disinformation to an issue of democracies being threatened by authoritarian regimes and associates the phenomenon with digital technologies, echoing Seo and Faris’s (2021) empirical review of U.S.-centrism in disinformation studies. Further research emphasizes the complexity of disinformation, which operates across ethical, legal, and commercial dimensions, rendering legislation or technology alone insufficient to address the problem (Ruiz, 2023; Soliman & Rinta-Kahila, 2024; Walker et al., 2019).

In our data collection and analysis, we ground the Voice referendum disinformation campaigns in Australian history, culture, and politics, and center our questions on power and inequality, focusing on Australia’s structural and social racism against Aboriginal and Torres Strait Islanders on the continent. At stake are power relations where Aboriginal and Torres Strait Islanders might gain greater symbolic or institutional representation and are thus central to the referendum-disinformation that surrounded the official referendum campaigns.

Further implications include showing how the scope of prevailing definition of disinformation and its regulation in the Australian institutional context fail to account for the role of dominant English-language media and government institutions in spreading misleading content in relation to race, gender, epidemics, finance, climate change, and geopolitics (Cheng, 2020; Hunter, 2024; Kreiss et al., 2020; Peck, 2019; Zigler, 2015).

While disinformation produced and circulated in major social media services has been well-studied (including for Australia’s  Voice referendum, see Brancatisano, 2023; Carson et al., 2024; Graham, 2024), this paper examines the influence of disinformation (as critically defined) in marginalized media ecologies that influence outcomes for Aboriginal and Torres Strait Islander people, as well as all Australians.

Our findings below carry substantial consideration for future work at the policy and research nexus. Firstly, they highlight the urgent need for regulatory bodies, policymakers, and electoral authorities to extend oversight mechanisms beyond mainstream English-language platforms. Policymakers should ensure that transparency requirements and truth-in-political-advertising laws explicitly cover multilingual media environments to close regulatory gaps.

Secondly, there is a clear need for culturally sensitive media literacy programs tailored specifically to diasporic communities—what we call “fit-for-audience-and-purpose” civic education. Educators and community leaders should co-develop resources that help audiences critically assess subtle disinformation tactics, such as emotionally charged rhetoric used for the profit motives of (diasporic) media ecosystems. For instance, co-designed media literacy programs can educate communities on the current digital media ecology and emerging risks associated with AI-generated content based on academic research. They can also address the political economy of national media—highlighting how ownership structures and shifting funding models influence editorial orientations and public discourse—and explain how algorithmic systems shape content visibility. Such ongoing, culturally attuned media training empowers users to navigate their media environments more critically and supports the development of inclusive, community-informed moderation practices.

Thirdly, the study suggests a need for infrastructural interventions by platform providers. Tech companies doing business in Australia should conduct regular algorithmic audits, publicly disclosing how their content recommendation systems influence the reach and engagement of political messages across different language communities. Enhanced transparency from platforms would improve accountability and help mitigate the unintended amplification of misleading content. However, at this stage, Australia’s technology-neutral policy stance may present challenges to advancing this proposal.

Finally, cross-sector collaboration between regulators, technology companies, community organizations, and academics is essential for effectivity in rapidly identifying emerging critical disinformation trends and collaboratively develop responsive interventions. Such coalitions are especially valuable for anticipating and countering the evolving tactics of modern disinformation campaigns. They also strengthen democracy in diverse, multicultural societies by building diasporic engagement in civic discourse to establish public trust. However, such collaborations must remain cautious of centralized funding structures, which risk allowing the most financially invested stakeholders—often the big tech companies—to define the scope and criteria of disinformation, undermining the proposal’s intent to prioritize collaborative decision-making led by communities, journalism, and academic voices.

Findings

Finding 1: WeChat coverage of the Voice referendum peaked ahead of the election, featuring pro-Yes ads and negative-sensational posts, despite platform restrictions on political advertising.

Throughout 2023, coverage of the Voice referendum on WeChat peaked between September and October with the intensified political campaigning ahead of the referendum date on October 14. While political ads are formally not allowed on WeChat, a large number of political ad content about the referendum proliferated (24% of all referendum-related content posted). Of 262 WeChat Official Account (WOA) posts, 52 explicitly supported “No,” while sensational yet ostensibly neutral articles predominated coverage of the referendum, averaging 2,676 views each. Sixty-five campaigning ads were included in the collected posts: 11 (17%) campaigned for Yes, though most ads featured less overt Yes / No messaging. The presence of political ads on WeChat for Australian WOAs is surprising, considering the platform’s regulations, but this is explained by previous findings on its distributed platform capitalism, which prioritizes monetizing user attention through advertising (Yang et al., 2024). Pro-Yes banner ads (n = 58) appeared across multiple accounts, receiving up to 32,600 views.

On WOAs, ads from the Australian Electoral Commission (AEC) and politician Keith Wolahan were authorized as required by the AEC. However, as we will discuss in the following section, campaigning materials circulated as short videos via WeChat’s Channel function were either unauthorized or not properly authorized. This contravenes Section 321D of the Commonwealth Electoral Act 1918 that electoral communications include an authorization statement specifying the person or entity responsible for the communication and their address (Australian Electoral Commission, 2025). Our ongoing monitoring leading up to the 2025 Australian federal election revealed an increasing number of such cases, highlighting an evolving regulatory gap driven by tech deployment (Yang et al., 2025).

Finding 2: Short videos emerged as an engaging format for political communication, driving disproportionate engagement with pro-No disinformation.

Short videos are a new format for WeChat with a higher volume of engagement via comments (n = 2,582) compared to WOA posts—Voice referendum-oriented or otherwise—and prove a fertile ground for critical disinformation campaigns. For these short videos, 91.7% of comments favored “No,” versus 6.0% for “Yes,” with the rest choosing “uncertain” or “irrelevant.” Pro-No videos averaged 979 likes and 4,447 shares—outperforming pro-Yes content (110 likes, 359 shares) by orders of magnitude. These spreads of sentiment diverge drastically from voter and Chinese diaspora sentiment on the referendum, where the ethnic Chinese vote leaned to “Yes” (see Fu et al., 2025). Table 1 shows the disparity in user engagement between pro-Yes and pro-No content and highlights the importance of conceptualizing the anatomy of critical disinformation. On WeChat, we can identify elements that contribute to the effectiveness of disinformation campaigns and, by extension, understand the factors that hinder anti-disinformation efforts within the platform economy and governance.

	Total posts (including campaign ads)	Posts without comments	Post comments coded “no”	Post comments coded “yes”	Post comments coded as “ambiguous”
WOA posts	262	129	76	10	47
Channel short videos	20	0	2368	157	56

Finding 3: Pro-No disinformation is framed through colonial tropes, anti-Labor sentiment, and racial or economic anxieties specific to Chinese-Australian contexts, skirting traditional media regulatory scrutiny.

While WOA posts generally remained ambiguous regarding their stance on the Voice, likely for advertising and user engagement considerations, comments on WOA posts showed patterns of critical disinformation: They adhered to identified disinformation narratives below and served to reinforce existing racial hierarchies in Australia against Indigenous populations, yet their content would be easily missed by regulators looking for a disinformation campaign. WOAs post news stories translated from English sources or opinion pieces of the WOA businesses themselves. The comments on these stories were largely sympathetic to the “No” campaign (n = 76), considering the small number (n = 10) of comments with pro-Yes views. Coding the “No” comments unearthed political opinion that echoed colonialist tropes, inflicting a celebratory tone about “civilization” and “British saviorism” in Australia to discredit a “Yes” vote.

Table 2 expands on the quantitative sentiment analysis of WOAs’ posts to highlight how the Yes ads and ambiguous Voice referendum content flowed into the public discussion, resulting in a higher volume of No-leaning discourse in online comments. Negative sentiment towards the referendum was not, for the most part, grounded in the issue of the referendum itself. Instead, concerns and anxious messaging particularly relevant to middle-class suburban Chinese-Australian communities were circulated alongside referendum discourse, triggering negative audience responses, even when referendum-related content was presented without explicit sentiment or opinions from media outlets. These concerns included anti-Labor government views (n = 24), concerns over racial inequality disadvantaging Chinese communities (n = 24), fears of increased taxation (n = 11), colonial primacy expressed through the embrace of British colonialism (n = 9), perceived irrelevance to Chinese Australians (n = 5), and unspecific reasons (n = 3). See the Methods section for examples of these themes in both Chinese and English.

Theme	Frequency (n = 76)
Anti-Labor	24
Racial inequality	24
Tax increase	10
Colonial primacy	9
Lack of relevance	5
Uncodable	3

Finding 4: Pro-No campaigners manipulate knowledge gaps and whitewash Indigenous rights for communal attention and trust, while leveraging colonial norms.

We further contextualize our quantitative findings with a case study on a short video, enabling an in-depth analysis of how disinformation narratives were deployed in the WeChat discourse. While WOAs rely on minimal curation to target subscribers, WeChat’s short videos employ algorithms that factor in user preferences, social networks, location, and demographics to extend reach beyond subscribers and amplify public engagement.

On September 27, 2023, a WeChat short-video account YamiChew published a video advocating a No vote. YamiChew was later identified as David Chu, a Liberal Party member and referendum volunteer (Cheng & Liu, 2023). Chu’s video reached over 10,000 reposts, 1,800 likes, and 300 comments within the first 24 hours of its publication, prompting further circulation and user engagement—consistent with platform algorithms that reward high-attention content. We selected YamiChew’s video for interpretative analysis not only due to its high user engagement, but also because of its distinct anti-sentimental, rational appeal. The video’s deliberately rational delivery appeared to obey platform logics, which typically reward emotional expression with greater content visibility, user engagement, and advertising revenue. What made YamiChew’s case particularly notable was that the popularity of his video stemmed from his rational tone—contrasting with typical campaigning videos that still relied on persuasive and emotionally charged messaging.

In the video, filmed at eye level, Chu stood on a park path wearing a black shirt with a “Vote No” badge, addressing the camera directly. He framed his No stance around two key claims: the potential erosion of Australia’s Constitutional integrity and opposition to perceived racial division through Indigenous recognition. His measured tone and detailed elaborations lent an appearance of rational argumentation to his “No” stance.

Chu supported his arguments with historical and legal references, though their validity warrants scrutiny. He claimed that his decision to vote No was informed by the AEC’s official referendum booklet, which presented eight reasons for a Yes vote and ten for a No vote. Two more reasons for a No vote. He further asserted that the Voice lacked legal comparability to established Western democracies such as the United States, the United Kingdom, and France. According to Chu, no equivalent “fourth constitutional body” exists in these systems, making the Voice in Parliament an unprecedented and legally unsound modification to Australia’s constitutional framework.

Chu criticized the term “Voice” as informal, arguing that the absence of Aboriginal recognition in Australian law since 1967 foregrounded the current inclusive, multicultural, and open society. He framed First Nations recognition as conferring race-based privileges that would impede societal progress and disadvantage all Australians, including Indigenous peoples and Chinese migrants. Chu claimed the recognition of Indigenous peoples would position Chinese migrants as a “third nation” below Indigenous peoples (the First Nation) and white people (the Second Nation) in a racial hierarchy. To support his position, Chu drew parallels to the Qing dynasty, suggesting that Manchurian elite privileges contributed to its decline. He further alleged that the Voice could be exploited by individuals with Indigenous heritage who live outside Indigenous communities, potentially undermining Indigenous interests.

Chu’s articulation is not only intentionally misleading but also serves to deny structural racism from society. For Chu, multiculturalism is achieved when race is entirely absent from public and legal discussion. He erroneously characterized the Voice as a “fourth constitutional power” despite its proposed advisory function without legislative or executive authority—a claim previously invalidated by AAP FactCheck (Williams, 2023). His interpretation of the 1967 referendum’s amendment to Section 51(xxvi) demonstrates a significant misreading of constitutional history. This amendment centralized race-based legislative authority under federal jurisdiction rather than eliminating it, thereby maintaining the settler-colonial governance while continuing the systematic exclusion of Aboriginal peoples from decision-making processes regarding their tangible and intangible resources on their lands and water.

Chu further manipulated knowledge gaps regarding Indigenous rights and political literacy around the Voice referendum among the Chinese diaspora, while also exploiting uncertainty about future implications. Chu achieved this by presenting a flawed comparison between colonized Indigenous peoples and Manchurian elites who were the governing power in the Qing dynasty. His arguments introduced or amplified fears and anxieties among Chinese migrants by suggesting the Voice would undermine democratic principles, reinforce discrimination against Chinese migrants, and contribute to societal decay. Despite factual inconsistencies, Chu’s eloquence, community status, and relatable historical analogies contrasted with pro-Voice information videos by social and political elites, frequently in English, rendering them less accessible to diaspora communities.

Methods

This paper addresses the questions: How was the 2023 Constitutional Referendum in Australia covered on WeChat among Chinese-speaking Australians? What disinformation narratives emerged when examined through the lens of Critical Disinformation Studies? And how did WeChat users engage with and respond to these narratives? In what ways did the producers and distributors of disinformation succeed in attracting online attention while simultaneously evading both WeChat’s regulatory mechanisms and conventional definitions of disinformation? And finally, does this indicate that our current disinformation regulation may be fighting the wrong battle? To answer these questions, we offer a mixed methods approach of quantitative content analysis and qualitative interpretive analysis (Wester et al., 2004) of WeChat Official Accounts and Channel short videos that is grounded in Kuo and Marwick’s (2021) Critical Disinformation Studies. Using the self-developed “share-capture” tool (Forydce, 2023), a bilingual researcher monitored 135 Australia-based accounts daily, exporting 262 relevant articles (including 65 campaigning ads) and 133 comments via Telegram for automated scraping; 20 short videos and 2,582 comments. Relevant public posts, along with their URLs, were exported from WeChat to Telegram’s group chat for researcher access and the bot’s data scraping. During the export process, the researcher categorized major themes of public posts and copied user comments anonymously into a group chat. A Telegram bot scraped the exported data for analysis. The tool worked well for WOAs, but faced challenges with short-video content due to the absence of unique URLs, requiring videos to be manually saved for further analysis. We offer a subset of coding work in Table 3, showing the Chinese text, researcher translation, and indicative code as per analysis above.

Codes	Translation	Original
Anti-labor	The Labor Party has taken power games to the extreme.	工黨將權力遊戲玩兒到極致了
Tax increase	If this Voice passes, then be prepared to pay more for everything. There are only downsides and no benefits.	这个voice如果通过了， 那就等着什 么都得掏更多的费用。 只有坏处没有好处
Colonial primacy	Without these immigrants from outside, who were predominantly British, Australia would have remained a pre-modern society, let alone become a developed country as it is today.	没有外面这些以英国人为大多数的移民进入，澳大利亚肯定还停留在前现代社会，更不用说目前的发达国家了
Racial inequality	You need to understand this first: The referendum is about granting Indigenous people more privileges, not about whether they can participate in the Parliament! The proportion of Indigenous people in Parliament already far exceeds their share of the population	你先弄弄明白， 公投是要给予土著人更多的特权， 而不是能否参加议会！ 土著人在议会的比例已经大大超过他们的人口占比！
Lack of perceived relevance to Chinese communities	Are you a robber yourself or an immigrant brought by robbers? I am neither so why should I bear the responsibility for others’ mistakes?	你是自己强盗还是强盗带来的移民？ 我两者都不是， 为什么要我承担别人犯错的责任？

We used quantitative content analysis (de Sola Pool, 1959) to count frequencies of referendum stance (yes/no/uncertain) and theme occurrences, and we also collected engagement metrics (views, likes, shares). This quantitative content analysis was completed by interpreting posts and comments as informed by Critical Disinformation Studies to code their meanings to Australian and diasporic political contexts and examine how disinformation was framed vis-à-vis social inequalities being expressed, reinforced, and justified through language, topic, and narrative structures. Following the line of critical disinformation, we have sought to code data points as disinformation to the extent they are situated in structures of power, grievance, and racism in Australia. Our approach is not meant to make a distinction between opinion and information (see Zaller, 1991) and instead focuses on dimensions of power.

The study builds on our ongoing research of WeChat’s “translational news ecology” (Yang et al., 2024) in Australia, where content is produced primarily through translation of English media into Chinese by media entrepreneurs, making the methods appropriate to explore and link our RQs to the WOA ecosystem that is the focus of this inquiry. The method can be also valuable for analyzing short videos on other platforms, particularly those depicting users’ everyday lives and conceptual metaphors that quantitative analysis alone fails to capture.

Framing disinformation through a critical lens (Kuo & Marwick, 2021) brings into relief the challenges that Western democracies face in post-truth and transnational communication era. We see a critical approach to disinformation emphasizing understanding intentionally misleading information within historical, social, and structural contexts. This means our work is as much about gathering data through empirical methods as it is about translating how the history of Australia, and its evolving migrant communities contextualize disinformation into specific contexts and forms, including here in the Voice referendum. We remain concerned with how pre-existing social hierarchies and identity-based prejudices (including racism) are fundamentally related to power in contemporary institutional knowledge production (Kuo & Marwick, 2021).