Using Social Media Data to Understand the Impact of Information Threats

Information threats can involve various forms of systematic deception and manipulation with the intent to negatively affect the information environment (Tursunaliyevich 2025). This article examines how researchers can use social media data to understand the impact of information threats using illustrative examples, and some of the key challenges encountered. It also considers how changes to platform affordances in terms of their specific culture, content, and data availability, can cause issues for researchers exploring deceptive and manipulative digital behaviour patterns over time. Importantly, whilst digital data can offer an indicator of behaviour, it needs to be contextualised using multiple data sources, social theories, knowledge of the media ecosystem, and awareness of its limits.

Whilst digital data can offer an indicator of behaviour, it needs to be contextualised...

Social Media Data

Engagement metrics (behaviours such as liking, sharing and commenting) data can be useful in showing the basic ‘reach’ of narratives. However, the large-scale presence of bots on many social media platforms may lead to overstating the true impact of information threats if using purely engagement metrics. Qualitative or topical analysis using multiple data streams is needed for greater accuracy. Digital data also offers the opportunity for longitudinal analysis, comparing engagement metrics and topic modelling over time. The following example illustrates how language analysis can enrich the insights that engagement metrics provide. It also illuminates the increased risk of information threats posed by the development of AI and new technology. Finally, the tracking of this story over time and across different platforms enables examination of behaviour patterns that provide a more robust indicator of impact and underlines the importance of longitudinal analysis. 

In April 2025, an AI-generated video of Prime Minister Sir Keir Starmer and Labour peer Lord Alli supposedly “kissing” circulated on social media (originating on TikTok and later spreading to X/Twitter). This was targeted at the government and Labour party to play on homophobic sentiments and suggest impropriety and corruption in the Labour party. This was later debunked however it continued to be widely shared despite fact checkers’ corrections. The originator of this video was linked to a TikTok account which frequently posted AI generated videos and conspiracy content. The video garnered over 4 million views and was shared widely (France 24 2025). 

An interaction network (replies, retweets, quotes, mentions around accounts which shared the TikTok video) was constructed between 18th-23th April 2025 with the intention to gauge audience perception and exposure to the content. A total of 42 user accounts were identified within the immediate interaction network (see figure below). These accounts were manually qualitatively classified into two distinct groups: sceptics (N=17, represented in green) and believers (N=25, represented in red). Believers used statements like “the moment everyone's been waiting for, The video is finally here” whilst sceptics used phrases like “Starmer needs to go but this is clearly fake and won’t stand up”. This shows the importance of qualitative coding and text analysis to discern impact beyond engagement and reach metrics by showing what degree of individuals profess to believe the content they are sharing. It also highlights that the video may be being shared for different reasons and that using simplistic engagement metrics may flatten this type of nuance.

Cross Platform Effects

When using social media data to understand the impact of information threats, it is crucial to note how information threats travel across platforms and the different ways that narratives and topics are discussed based on the platform or medium. The Breakout Scale (Nimmo 2020) charts movement across platforms and offline to assess impact. It categorises impact into six levels from category one: one platform with no breakout (stays within a single community) to category six: a policy response or call for violence. This is intended for practical use by operational researchers to enable real-time prioritisation. The below figure shows how the Breakout Scale can be understood when studying information threats and suggests additional considerations.

First it begins by noting that the AI generated video originated on TikTok but played on pre-seeded narratives around Keir Starmer’s sexuality rooted in homophobia. There have been previous allegations of inappropriate relationships between Starmer and Lord Alli, and baseless accusations involving the relationship between Starmer and perpetrators of the arson attacks against him in 2025 (Heale 2024; EurAsia Daily 2025). False news has been found to spread faster than truth on social media (Vosoughi et al. 2018). This explains why in the Breakout Scale information threats progress rapidly to category three across multiple platforms and communities, especially when considering fringe and less moderated social media platforms. Rumours regarding Starmer’s sexuality spread quickly across multiple platforms and were discussed immediately by conspiracy groups and far-right and far-left groups which were critical of Starmer. Modified or  sanitised versions of the narrative were reported on mainstream media, and more salacious or overtly homophobic and critical accounts were published on alternative media sites. Because of the disparity in the way that the narrative was being discussed, there were accusations of censorship and cover up by mainstream media. This often occurs when the threshold of category four information threat is reached. Information threats can quickly progress to category five through the actions of influencers, personalities, and commentators with large followings, who report unverified and inflammatory content. Information threats are often centred around controversial events in the policy sphere and thus can easily transition into calls for policy change and in this case incite homophobic hate speech. There were also calls to examine mainstream media censorship. 

Despite this rapid acceleration through the categories in the Breakout Scale this narrative was quickly debunked and did not resonate with audiences outside of specific far right and conspiracy circles, or over a long period of time. By tracking this story across different platforms and over time, it is clear that although the Breakout Scale offers a useful way to classify information threat narratives, even information threats with relatively small impact can quickly reach category six, and that the categories are not necessarily linear. 

Information threats can often quickly progress to category five as there are several influencers, personalities and commentators with large followings that report unverified and inflammatory content. 

Platform Affordances

The profile and components of information threats change dependent on the platform and the affordances associated with specific operating environments. Recent years have shown that platforms can have specific cultural settings and purposes which are enabled by their affordances (Van Raemdonck and Pierson 2021; Ercegovac and Tankosic 2024). For example, algorithms that allow content to be hyper-personalised create more polarised echo chambers (Lomonaco et al. 2023). Indeed, whilst certain fringe platforms may have reduced content moderation and thus permit more extreme viewpoints, they also tend to have lower visibility which decreases their impact (Schulze et al. 2024). The example on this page illustrates this point as whilst the bulk of engagement metrics and views were received on more mainstream platforms such as X/Twitter and TikTok, the more extreme rhetoric tended to be on alternative platforms such as Telegram and Rumble. Understanding the different uses enabled by specific platforms is useful in mapping the indicators of information threat effects. 

The Breakout Scale diagram shows how technological affordances of AI generated video can influence the information environment . Platform cultures and preferences for different types of content and affordances can also change over time. For example, recent years have shown a trend toward video content, and this presents new challenges for researchers to analyse (Verma and Agrawal 2016; Velčković et al. 2025). There is a need for research to adapt rapidly to respond to the evolving media landscape and changes in the tactics of information threat actors.

Future research needs cross-platform examination, longitudinal data access, integration of mixed methods and social theory.

Elon Musk’s takeover of Twitter (now X) (Barrie 2022; Marzouk 2025) has radically altered the culture ‘on platform’, and changed data accessibility for research, impacting the long-term comparability of data over time (Blakey 2024). There has been a failure of regulators involved in the Online Safety Act and EU Digital Services Act to ensure data availability to facilitate consistent comparison and understanding of social media data over time. These implications must be considered when attempting to estimate information threat effects over longer time-frames. 

Conclusion

As these examples illustrate, digital social media data can be an ‘indicator’ but not a ‘measure’ of information effects. It does not and cannot tell the full story when used in isolation, or without theoretical interpretation. There are clear difficulties when translating social media data from crude engagement metrics to more nuanced effect indicators such as belief in a news story or estimating offline behaviour from online actions. Future research needs cross-platform examination, longitudinal data access, integration of mixed methods, and social theory to contextualise digital social media data. To achieve this, research and policy environments must adapt to a rapidly changing media ecosystem with constantly shifting cultural preferences and technological affordances. It is important to recognise both the opportunities offered and limitations inherent when analysing digital data and utilising it as an indicator of information threat effects.