Society & surveillance
False stories travelled farther than true ones—and bots did not explain the difference
A historical Twitter study found a striking gap in diffusion. The next question is what its measures tell us about human sharing—and what they leave out.
Following the branches of a rumour
The 2018 study by Soroush Vosoughi, Deb Roy and Sinan Aral examined roughly 126,000 rumour cascades on Twitter from 2006 to 2017. Its dataset linked claims to judgments from six fact-checking organizations. False claims travelled farther, faster, deeper and more broadly than true claims within that sample.
Those are different measures. A cascade can reach many people through one popular account, or travel through a long chain of successive retweets. Comparing breadth, depth and speed helps distinguish these shapes. The paper's result is therefore more informative than saying that false stories received a large total number of clicks.
Bots were not the whole explanation
The researchers examined automated accounts and found that their inclusion did not explain the difference between true and false diffusion. False stories retained their advantage when likely bots were removed. They also investigated novelty and emotional responses as possible contributors.
That does not make automation irrelevant. A factor can increase overall circulation without explaining why one class of story outperforms another. Imagine two runners receiving the same assistance: the assistance matters to both performances, but it does not by itself explain their finishing order. The bot finding concerns the observed difference, not a universal declaration that bots have no influence.
The sample has an entrance requirement
These were fact-checked rumours, not a random sample of every statement published online. A claim must attract attention and be checkable enough to receive a verdict. That selection can affect which topics and diffusion patterns appear in the dataset.
A second boundary concerns behaviour. Retweeting a claim is observable; privately believing it is not directly measured by that action. Someone may share to endorse, mock, criticize or question. The study deliberately separated veracity from claims about the sharer's intent. Readers should preserve that distinction when using its findings to discuss deception. A false statement and an intentional lie are different categories.
A result is not a permanent platform law
Twitter during 2006–2017 had particular users, features and incentives. Its historical name and dates belong in the conclusion. An inference about a later platform, another language or a different recommendation system would require fresh evidence.
The practical value of the paper is a more demanding way to ask why information spreads. Examine the structure of sharing, the selection of stories, the role of accounts and the content's novelty before choosing one explanation. For an archive reader, popularity should not be confused with independent verification. A claim repeated through thousands of branches may still trace back to one unsupported assertion. The paper measures how that circulation can differ; it does not relieve a reader of checking the original evidence.
Sources and further reading
- The spread of true and false news online ↗
Science 359, 1146–1151; abstract, definition of cascades, Methods and bot/novelty analyses