A detailed analysis of the social media output of former President Donald Trump has uncovered a systematic shift in his disinformation strategy, according to a study published by the Center for Digital Analysis at the University of London. The research, which examined over 10,000 posts made across Trump’s Truth Social and X accounts between January and August 2024, identifies a deliberate move towards a technique researchers are calling “narrative seeding.”
Unlike traditional disinformation which relies on repeated false claims, narrative seeding involves introducing a series of seemingly unrelated but subtly connected falsehoods. These are designed to coalesce over time into a broader conspiratorial framework that is harder to debunk than any single piece of fake news. Dr. Eleanor Vance, the lead author of the study, described it as a “sophisticated evolution in political communication masquerading as haphazard posting.”
The algorithm, developed by the centre, tracked the semantic proximity of terms and themes across Trump’s posts. It found that a core set of about 30 keywords—including “deep state,” “rigged,” “weaponised,” and “un-American”—was consistently reintroduced in new contexts. For instance, a post about the economy might suddenly reference “deep state bureaucrats,” subtly linking economic grievances to the myth of a rogue intelligence apparatus. Over a period of weeks, the repeated coupling of terms created mental associations among followers, effectively training them to see hidden connections where none exist.
One striking pattern identified was the use of what the study terms “temporal salting.” This is the strategic insertion of a false claim into a high-volume news cycle—such as a natural disaster or a diplomatic crisis—so that the claim is less likely to be independently fact-checked and more likely to be shared in the ensuing confusion. During the summer’s international tensions over Taiwan, for example, Trump posted allegations of “secret flights carrying biological weapons” that were wholly fabricated but gained traction because they appeared to reference an ongoing news story.
The researchers also noted a marked increase in the use of “dog whistles” targeting specific demographics. Analysis of post timings and engagement data showed that messages containing coded racial or ethnic appeals were often posted in the early hours, when fewer journalists and moderators were monitoring, but when core supporters were most active. These posts would later be amplified by automated bots and decontextualised in other forums.
Senior strategists within the Republican party have declined to comment on the study, but a spokesperson for Trump’s campaign dismissed it as “another hit job by partisan academics.” However, the paper’s findings have been taken seriously by platforms. A spokesperson for X told the BBC that the company is “reviewing the research” and considering adjustments to its content moderation algorithms, which currently rely on detecting repeated or near-identical copies of false information.
The implications for the forthcoming election cycle are significant. Traditional fact-checking, which focuses on correcting individual false statements, may prove inadequate against a tactic that builds cumulative false narratives from disparate parts. As Dr. Vance put it, “You cannot debunk a narrative that hasn’t been fully told yet. By the time the story emerges, the seeds are already deeply rooted.”
The study recommends that media organisations and platforms invest in “narrative tracking” tools that can map the evolution of themes over time, rather than simply flagging individual posts. It also calls for greater transparency from platforms about how their recommendation algorithms propagate seeding strategies.
For now, the integrity of public discourse remains under threat from increasingly subtle forms of manipulation. This research offers a first step in understanding the new lexicon of disinformation.
