1 From early adoption to current practices

Since their massive diffusion during the first decade of the 2000s, it became clear that social media could serve as a valuable source of real-time information about any topic, acting as crowdsourced newsfeeds [1], and being able to support situational awareness and decision-makers during mass emergencies [2]. While almost any available social media has been used in literature as an information source, Twitter quickly emerged as the most suitable option due to several factors, including its micro-blogging format based on frequent short posts, and the availability of user-friendly APIs [3]. Most importantly, even if a subscription fee was required for full access, the free version of the APIs allowed enough data to be crawled to support research work in different domains. Other popular sources struggled to gain the same success in the scientific community, either for difficulties in accessing the data (Facebook) or for the challenges of processing multi-modal data with the methodologies available at that time (YouTube, Instagram, Flickr). In fact, early research primarily relied on network analysis to explore community structures and information diffusion, alongside Natural Language Processing (NLP) to identify crisis events and assess public perception, among other tasks. Along with the use of social media data, researchers have characterised their limitations. These include various inherent biases in users’ selection (e.g., demographic), as well as technical challenges in data analysis due to the volume, unstructured nature and multilingualism of their content. Additionally, the overflow of unverified content also facilitates the dissemination of misinformation [4]. More recently, the integration of transformer-based architectures and Large Language Models (LLMs) has marked a significant shift in crisis monitoring. BERT-based models have dominated event detection and classification tasks, while decoder-only models such as GPT-4 and LLaMA-2 have gained traction for their reasoning and few-shot capabilities [6,5]. Despite their promising performance, these models still struggle with the informal language and domain-specific jargon that are characteristic of crisis communication.

However, the popularity of using social media data for crisis management (i.e., prevention, preparedness, response, adaptation) has decreased over the years, facing substantial disruption following Elon Musk’s acquisition of Twitter in 2022 and the subsequent operational changes, including the platform’s rebranding to X, altered content moderation policies, and changes in the API access (i.e., end of free API in February 2023) [7]. These developments have prompted many scientists and professionals to reconsider their reliance on the platform for crisis communication, marking a potential end to the era when Twitter served as a primary conduit for scientific discourse during public emergencies. Following these events, alternative micro-blogging platforms are emerging (e.g., Bluesky, Mastodon, Threads) and have been the subject of massive user migration from Twitter/X (Fig 1). However, there is still uncertainty about when (and if) some of these networks will reach a critical mass to make them worth considering in the context of crisis management [8].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Number of articles per year, including the name of an Online Social Network in the title, alongside a crisis management keyword.

Articles including the generic “Social Network” mention in the title are also included. Titles were retrieved via the Scopus API in February 2026, limiting the search to four subject areas involving data oriented/computational approaches: Computer Science, Mathematics, Engineering, Decision Sciences. Code and list of keywords at https://github.com/interdonatos/ScopusCrawl/.

https://doi.org/10.1371/journal.pcsy.0000112.g001

Many other challenges in this context are related to the General Data Protection Regulation (GDPR) implemented by the EU since May 2018. The GDPR is probably the first (and only) example of a unified law about the privacy and use of personal data at the continental scale. While this law is nowadays a fundamental component of European Union (EU) privacy and human rights laws (also regulating the extent to which data can be used outside the EU), it also represents a significant limitation on how social media data can be used in the context of research projects (and especially EU-funded ones such as Horizon or ERC projects). In particular, it may hinder reproducibility, as researchers are often unable to archive and share datasets containing information that could directly or indirectly identify individuals. Moreover, researchers must ensure that when users withdraw their consent from a platform, all corresponding records are removed from the datasets. This implies the development of monitoring systems capable of detecting consent withdrawal events (e.g., account deletions) and updating datasets accordingly. Over time, as more users delete their accounts or revoke consent, datasets must be continuously updated and may consequently shrink, further complicating longitudinal analyses and reproducibility. Some further concerns have emerged about the new Twitter/X policies no longer fully fulfilling key obligations required by the GDPR. While the existing literature on the application of the GDPR to the humanitarian aid sector is rather limited [9], some studies suggest that people may agree to share their personal data during emergency situations [10].

2 Future directions for crisis monitoring

The recent increase in the cost of access to Twitter/X’ APIs, combined with regulatory constraints, strongly questions the sustainability of relying on a single dominant platform for crisis monitoring. A natural alternative to Twitter/X is leveraging other social media platforms such as Mastodon, Reddit, Telegram, or TikTok. However, existing studies show that none of these platforms can fully substitute Twitter/X’s role in disaster management due to lower and more uneven coverage and strong access limitations. To name a few, TikTok lacks a real-time public stream, Reddit and Telegram rely on fragmented access (content is dispersed across independent subreddits and channels that must be individually identified and monitored, rather than aggregated into a single public stream), Bluesky and Mastodon have extremely limited adoption. These limitations position these platforms as valuable complements rather than standalone replacements [8]. Moreover, each social media platform must be used according to its audience and user behaviors, to match regional contexts and analytical objectives.