EXC: The Architect of ‘Misinformation’ Censorship Is Now Targeting ‘Anti-Immigrant’ Posts

For nearly a decade, Kate Starbird has been one of the most influential academic voices shaping how “misinformation” is defined, studied, and ultimately moderated online.

Her research helped supply the intellectual framework used by major platforms, government agencies, and NGOs to justify sweeping censorship of election-related speech after 2016 and 2020. Claims about voting systems, ballot integrity, and election administration were no longer treated as political arguments. They were reframed as harmful narratives subject to intervention, throttling, or removal.

Now, Starbird is applying that same framework to a new domain: immigration.

A recently published paper in the Proceedings of the Association for Computing Machinery, co-authored by Starbird, makes a clear pivot. Titled Data Visualizations as Propaganda: Tracing Lineages, Provenance, and Political Framings in Online Anti-Immigrant Discourse, the study does not investigate falsehoods. It investigates political dissent. Specifically, how Americans use government data to criticize immigration policy.

The paper’s central move is subtle but profound. It abandons the question of whether immigration claims are true or false and replaces it with a different test: whether those claims support what the authors label “anti-immigrant frames.”

Once speech is defined by its framing rather than its accuracy, censorship becomes a design problem.

From Election Integrity to Immigration Narratives

Starbird’s earlier work on elections focused on “rumors,” “sensemaking,” and “misinformation dynamics.” Those concepts did not remain academic. They were incorporated into platform policies, trust-and-safety playbooks, and federal “disinformation” initiatives.

In this new paper, the same logic reappears but the subject has changed.

Instead of election skepticism, the authors analyze immigration charts built from official U.S. government data, including Customs and Border Protection, Census, CDC, and Pew Research datasets. These charts are not accused of fabricating numbers. They are accused of supporting the wrong political conclusions.

The paper explicitly categorizes four “anti-immigrant frames” expressed through data visualization.

First are demographic arguments: charts showing population change over time, birth rates, or shifts in racial composition. The authors argue that even accurate Census or natality data can invoke what they describe as “Great Replacement” subframes when immigrant and non-immigrant populations are compared. The act of comparison itself becomes suspect.

Second are visualizations assigning blame to government policy, particularly those linking immigration trends to the Biden-Harris administration. Charts juxtaposing border encounters across presidential terms are described as misleading not because the data is wrong, but because the framing assigns culpability.

Third are crime-related visualizations. Charts that connect immigration to criminality — even when using government statistics — are treated as dangerous rhetoric. The authors acknowledge inconsistent sourcing in some examples, but the core objection is broader: crime should not be visualized in connection with immigration at all.

Fourth are fiscal arguments. Maps and infographics showing the cost of immigration services, welfare usage, or strain on public resources are labeled misleading because budgets vary across jurisdictions and are difficult to verify. Again, the issue is not falsity, but narrative impact.

Across all four categories, the same rule applies: accuracy is secondary to effect.

The China Case and National Security Speech

The paper then extends the framework further, analyzing viral posts about Chinese migrants crossing the southern border in 2024. Users filtered a publicly available CBP dashboard by nationality to show an increase in Chinese nationals.

The authors describe this as evidence of a subframe suggesting that “a large, dangerous group of Chinese military-aged men were coming to the United States as part of a pseudo-military action.”

The data itself is not disputed. It comes directly from a federal dashboard. What is condemned is the use of that data to make a national-security argument.

The paper reframes this activity as participation in “framing contests,” a term that recasts political debate as a narrative battle requiring oversight. Ordinary citizens using government tools to argue policy become “motivated individuals” generating suspect evidence.

This is not analysis for understanding. It is analysis for control.

Where the Paper Is Headed: Moderation

The authors’ intent becomes unmistakable in the later sections.

In a chapter titled “Defining Harms,” the paper claims that immigration data visualizations “have the potential to cause physical and material harm,” linking them broadly to hate crimes, deportations, and enforcement actions. The authors admit they struggled to specify concrete harms throughout the paper, citing a lack of shared definitions. Rather than narrowing their claims, they expand the concept of harm to include stereotyping, dehumanization, and disparagement.

From there, the paper openly turns to moderation.

The authors argue that immigration data visualizations should be evaluated alongside other content types already subject to platform rules and that it is “imperative” to understand how these visuals fit into “schemas of online harm types” and toward “potential moderation.”

They call for research into how such visualizations might be governed by “content moderation and community guidelines” and explicitly suggest that immigration charts should be treated like other moderated visual objects.

They also raise the prospect of AI-generated data visualizations, arguing that future systems should include “built-in safeguards” against producing these kinds of immigration charts and questioning who should be held accountable when such content is generated or amplified by AI.

The Pattern Is Familiar

This is the same progression seen with elections.

First, political arguments are reframed as harmful narratives.

Then harm is expanded beyond falsity to include perception and impact.

Then speech is sorted into categories.

Then moderation is presented as mitigation.

Starbird’s name sits at the center of both arcs.

The difference now is the target. What was once election discourse is now immigration policy, demographic change, crime statistics, welfare costs, and even discussion of Chinese migration trends.

The common denominator is not inaccuracy. It is disagreement.

The Bottom Line

This paper does not call for censorship explicitly. It does not have to.

By redefining immigration arguments as harm, by treating government data as dangerous in civilian hands, and by explicitly situating immigration visualizations within moderation frameworks, it lays the intellectual groundwork for the next phase of speech control.

The same researcher who helped shape the academic case for policing election speech is now doing the same for immigration.

Different issue. Same architecture. Check out some of the research paper’s graphics below.