Key claims
Rumors exist along a spectrum—not all are problematic. Before the internet, rumors were a normal part of crisis response as people tried to make sense of incomplete, rapidly changing information (Is there one shooter or two? Is the dam breached?). Social media has changed the scale, speed, and distance at which rumors spread, but rumoring itself is a natural human sense-making process during uncertainty.
Three types of rumors emerge during crisis events:
- Sensemaking rumors — people trying to understand what's happening (Boston Marathon: speculation about suspects, geography)
- Clickbait rumors — attention-seeking false claims (the shark in hurricane floodwater, false reports of deaths in the bombing)
- Conspiracy narratives — coordinated false explanations claiming events are staged, staged by government, or part of a hidden agenda (false flag, crisis actors, inside-job theories)
False flag and crisis actor narratives are not marginal. In the Boston Marathon data, ~20,000 tweets promoted false-flag conspiracy theories—the claim that the 2013 bombing was perpetrated by the U.S. government, not the Tsarnaev brothers. This pattern repeated across 2016 shooting events (Orlando, San Bernardino), with identical messaging (same actors in Sandy Hook and Boston, gun takeaway agenda). The same conspiratorial tropes resurface across years and events, suggesting organized or at least coordinated narrative templates.
Conspiracy ecosystems mix multiple ideologies and concerns. Content analysis of conspiracy-promoting sites revealed content on GMOs, chemtrails, flat earth, 9/11 trutherism, anti-vaccine narratives, and anti-globalism—often framed as appealing to "critical thinking." These aren't random: sites appear to act as gateways where people entering through one concern (GMO worry) get exposed to others (9/11 trutherism, flat earth). Starbird confirms this through interviews: a person she spoke with entered through GMO concerns on InfoWars, moved to 9/11 trutherism, then flat earth, and ended up buying nutritional supplements on conspiracy-theory sites.
State-sponsored media plays a role in these ecosystems. Starbird's domain-network analysis identified Press TV (Iran), RT, Sputnik, and pro-Putin think-tank content embedded in conspiracy-theory sharing networks from 2016. These weren't the most-cited sites, but their presence in the network suggests information-operation coordination around alternative narratives and anti-globalism themes.
Computational propaganda amplifies conspiracy narratives through bots. Domain analysis identified a bot network (~700 accounts) promoting "realest strategy.com" (7,000 tweets in the dataset showing mechanical, hourly posting patterns unlike organic discourse). A separate bot network amplified InfoWars, despite InfoWars not being heavily linked by organic conspiracy theorists—suggesting bought amplification. This coordination is invisible to individual users scrolling Twitter but visible in aggregate temporal and network analysis.
The "alternative media" frame masks a coordinated ecosystem. Sites term themselves "alternative" or "independent" media while explicitly attacking mainstream media as lying. The framing "We are free and critical thinkers; the mainstream media is censored" resonates with genuine concerns about media consolidation and attracts people who believe in democratic information ideals. But the ecosystem functions as coordinated disinformation infrastructure mixing fringe, credible-sounding, and state-operated sources.
Anti-globalism is the ideological convergence point. Starbird's 2016 data showed left-leaning anti-globalism (against exploitative economic policies) and right-leaning anti-globalism (anti-immigration, nationalism) converging on a single set of alternative-media sources and narratives. Anti-Semitic tropes (Rothschild Zionist media conspiracy) appeared in the data, linking old conspiracy narratives to contemporary political messaging.
Mechanisms discussed
How false narratives spread during crisis:
- Information vacuums and emotional intensity (fear, anger, confusion) drive rapid sharing
- Conspiracy narratives provide simple explanations for complex, frightening events
- Questioning language ("Was it a false flag?") uses uncertainty rhetorically to plant doubt
- Repetition across events (connecting Sandy Hook, Boston, Orlando) builds a false pattern
Domain-network methodology:
- Maps which domains users cite together across tweets
- Shows structural relationships in information ecosystems
- Reveals both organic and bot-amplified content
- Temporal signatures distinguish human sharing (spikes and decay) from bot patterns (regular intervals, minimal decay)
Ecosystem architecture:
- Red sites: explicitly promote conspiracy narratives
- Blue sites: debunk conspiracy theories
- Yellow sites: contain factual content (e.g., articles mentioning suspects' backgrounds) that conspiracy theorists repurpose as "evidence"
- Aggregators (Before It's News, Veterans Today, Activist Post): republish and connect conspiratorial content
- State sources: positioned as "alternative" voices to Western mainstream media
Notes
This talk is a crucial bridge between Starbird's early crisis-informatics work (studying prosocial social-media use during disasters) and her pivot to studying information warfare and conspiracy ecosystems. Her opening caveat—that she used to give "happy talks" but now gives darker ones—marks the moment she recognized that the same platforms enabling disaster response were being weaponized for coordinated disinformation.
The Boston Marathon bombing case study is particularly valuable because it isolates rumoring behavior in a bounded event: the same Twitter stream contains sensemaking rumors (trying to identify the actual perpetrator), attention-seeking rumors (false deaths), and conspiracy narratives (false-flag claims) simultaneously. Starbird's tweet-by-tweet analysis shows the types of language and framing used to propagate conspiracy theories: leading questions, speculation dressed as evidence, appeals to "critical thinking," and repetition connecting unrelated events into a false pattern.
The domain-network analysis was methodologically novel at the time (2017). Rather than classifying individual sites as "misinformation" or "reliable," Starbird mapped relationships—which domains shared audiences, which appeared together in user timelines—revealing an interconnected ecosystem. This approach prefigured later work on platform affordances, algorithmic filtering, and the structure of information disorder at scale.
The observation about state-sponsored media's role is understated in this talk but significant in retrospect (post-2016). Starbird identifies RT, Sputnik, and Iranian outlets in conspiracy-theory networks but doesn't overstate their influence (they're small, not central). Yet the presence is notable: foreign governments were opportunistically inserting themselves into U.S. conspiracy ecosystems, not creating them from scratch. This prefigured later revelations about Russian troll-farm operations and state information warfare.
The "gateway" model of conspiracy theories—entering through GMOs, graduating to 9/11 trutherism, flat earth—challenges the popular theory that conspiracy beliefs are psychologically unified (one personality type believes all conspiracies). Starbird's evidence suggests information-ecosystem design makes the journey: sites strategically cross-link and recommend related content to keep audiences engaged across the conspiracy spectrum. This is less about individual vulnerability and more about ecosystem architecture.
Starbird's distinction between "alternative media" (genuine critique of consolidation) and "conspiracy media" (using that framing as cover for disinformation) is crucial: it explains why conspiracy ecosystems are resilient to criticism. Attacking them as "misinformation" plays into their narrative that "mainstream facts-checkers" are suppressing truth. Solutions require institutional rebuilding (funding independent local journalism, platform redesign) not just individual media literacy.