The massive use of social media networks for the dissemination of content has led to a significant increase in the reach and impact of disinformation, understood as any type of false, inaccurate or misleading information, created with the intention of causing public harm and obtaining economic gain (European Commission, 2018). This phenomenon constitutes a major risk for society as a whole, as it increases polarisation and undermines trust in both the media (Altay et al., 2024) and the institutions alike (Ecker et al., 2024), affecting, in particular, certain segments of the population (Budak et al., 2024). In addition, disinformation is particularly relevant in election periods, as fake news is used to manipulate public perception and disrupt democratic processes (McKay & Tenove, 2021). 

This article focuses on the spread of multimodal disinformation on social media over the past few years. Specifically, messages – accompanied by images and videos – published since the end of 2019 on open and Spanish-language Telegram channels have been analysed. The reason behind this study is that, although Telegram is a messaging platform that has been used for the public dissemination of disinformation since the beginning of the pandemic, there is little research that has analysed this (Bodaghi et al., 2024). Moreover, the scientific literature has suggested that multimedia elements can amplify the negative effects of disinformation (Dan et al., 2021), so it is important to know how they influence the spreading of messages.

The results show that, on Telegram, the number of messages accompanied by images and video has increased significantly in the last two years, especially around certain topics. In addition, multimedia messages reach a wider audience than those only containing text. Likewise, the difference in reach and virality between multimedia and non-multimedia messages is more significant when they are published on channels that have more subscribers. Combined, these two facts pose a significant challenge for initiatives fighting against disinformation.

Many of the downloaded Telegram messages contained some kind of multimedia content or a link to access it. 

However, if media messages are considered separately, it is observed that text or web messages predominate. On the other hand, alternative channels mostly contain messages that incorporate audio and video. 
 

The number of messages published by the alternative channels analysed increased in 2020 and 2021, when most of them began their activity. At the end of 2019, the average number of messages published on these channels was about 2,000 per month, a figure that in 2022 increased to 25,000 and is currently maintained. 
 

The number of messages including multimedia elements has also increased. While, at the beginning of 2020, 82% of messages were non-multimedia and 18% contained multimedia (the former exceeded the latter by more than four times), in four years, the situation has practically reversed: in July 2024, 23% of messages were non-multimedia while 77% did contain multimedia.

Images – static or animated – are the multimedia elements that have increased most. Thus, by early 2020, they had ceased to represent under 10% of the total to become, at over 40%, the most common content. The inclusion of videos in messages on Telegram channels has also experienced a notable boom, from being present in one in six publications to appearing in one in three. In contrast, publications containing web links, which accounted for over half of the total, have been reduced to approximately 10%. 
 

The study shows that multimedia messages on alternative media accumulate, as a whole, more views than those that only contain text. This makes sense, given that the number of multimedia media messages has increased over time, which also increases the cumulative total number of views. 

The average number of views, on the other hand, is higher for non-multimedia messages if we consider all channels (Figure 6.A). However, if the calculation is limited to the channels with the most views (Figure 6.B), the opposite result is obtained: the average number of views for multimedia messages is higher. This seems to indicate that the most successful channels are able to use visual resources to attract a wider audience, exploiting users’ preference for more dynamic and entertaining content (Dan et al., 2021). In contrast, the least watched channels do not achieve this and cause the global average of the reach of multimedia messages to fall.
 

Based on the metric used in this study, multimedia messages tend to be more viral; that is, they tend to stand out more frequently than non-multimedia messages published on the same channel. 

In any case, the number of viral messages is not very high, so there are generally not many messages with a disparate number of views compared to those belonging to the same context. One possible explanation for this is that, on Telegram, information is not consumed in the same way as on other platforms, such as X. Although in some aspects Telegram behaves like a social media network, the way of viewing messages is similar to that of a messaging application; that is, sequentially and, normally, in blocks of attention. This behaviour implies that, unlike what happens with other social networks (Vosoughi et al., 2018), all messages on the same channel spread at a similar speed. The accumulation of views occurs in a short time interval, reaching a total similar to that of other messages in the channel, and the lifetime of the messages is not very long.

The evolution over time of the number of messages published on alternative channels regarding various topics reveals distinctive patterns of activity that seem to be related to significant events in each area of discussion. The graphs show that some topics, such as "Covid-19 vaccination campaigns" and "Biden-Trump controversies and legal problems" have pronounced peaks in messages that coincide timewise with critical events, such as vaccination spikes in 2021 and presidential campaigns or appointments in the United States. This suggests a strong correlation between activity on Telegram and socio-political events with high media relevance. 

For all topics, the proportion of multimedia messages increases over time, as had already been detected without this segmentation. This is what happens, for example, in "Events and politics in Spain" and in "Climate change and weather". This second case is especially illustrative: currently, practically all messages in this area contain photos or videos of atmospheric phenomena that emphasise the minimisation or denial of climate change. Moreover, it should be noted that some topics have included a large number of multimedia messages since their inception. In this regard, worth noting are "War in Ukraine: conflict between Ukraine and Russia" and "Escalation of the Israel-Palestine conflict and humanitarian crisis in Gaza". Both cases exemplify the visual nature of some topics and therefore, they have an even greater impact on the audience. In addition, the topic of the war in Ukraine is interesting because it involves a much higher volume of multimedia messages than expected in that initial period. 
 

El The reach and virality of the messages segmented by topic do not differ substantially from what has already been explained. If the entire dataset is considered, the metrics for multimedia messages are not significantly different from those for non-multimedia messages. However, if the analysis is limited to alternative channels with a higher number of subscribers, then it can be seen that multimedia messages have greater reach and virality. The difference increases when it comes to topics that a priori are less visual, such as "News events and politics in Spain" or "Abortion and gender today". This reinforces the hypothesis that the most successful alternative channels are able to activate their audiences even on less attractive topics by taking advantage of visual resources.

Analysis of messages posted on Telegram channels from late 2019 to 2024 has revealed that multimodal disinformation, especially when it includes images and videos, is reaching more people than messages made up solely of text or web links. This behaviour is particularly noticeable in channels with a high number of subscribers, where multimedia content manages to attract greater attention from the audience. In addition, a sustained increase in the proportion of multimedia messages has been observed over time, as has the fact that certain specific topics, such as "War in Ukraine" and "Covid-19 vaccination campaigns", show patterns of high activity at critical times. These results underline the need to develop specific strategies to address disinformation on messaging platforms, considering the differentiated impact that different types of multimedia content have on the perception and dissemination of information.

To conduct this study, messages were downloaded that had been published in a set of Spanish-language channels on Telegram and selected by researchers from Maldita, an independent, transparent, and non-profit fact-checking organisation, accredited by the International Fact-Checking Network (IFCN), based on previous findings (Maldita.es, 2022) and recent verifications. This list included traditional media channels from the SCImago Media Ranking (Trillo-Domínguez et al., 2023), which the authors call media, and channels suspected of spreading disinformation or framed within so-called parajournalism, which are referred to as alternative channels. Using the pytopicgram library, more than 1.5 million messages published between 1 December 2019 and 31 August 2024 were obtained through the Telegram API, of which around half a million corresponded to media. The groupings of the messages into topics were obtained using the BERTopic algorithm (Alammar & Grootendorst, 2024), with a threshold of 1,000 topics. The generation of the descriptions was carried out using OpenAI GPT-4o.

The differentiation between multimedia messages and non-multimedia messages was made from the data provided by the Telegram API. The metadata associated with each downloaded message include several fields that indicate, among other things, whether the message contains an image, video, audio, or document. When the message includes a web link to an image or video, Telegram can embed the multimedia content in the message, in which case it is also marked as such in the message data. Web documents not embedded in messages were considered non-multimedia, even though the website accessed through the message may include this type of elements, as well as attachments of an unknown type. Each message was assigned only one type: image, video, audio (multimedia); web, text, document (non-multimedia).

To evaluate the reach of the messages, the number of views was considered, a figure that is also provided directly by the Telegram API. This number represents the unique views that a message has received. Telegram does not offer a detailed official explanation of how this value is calculated. For example, it is not known if there is a minimum time during which a user must see a message for it to be considered a view. If the same user sees a message multiple times, it only counts as once. To measure virality, a metric was developed that quantifies the effectiveness of the spread of a message by comparing its number of views with that of those received by messages published on the same channel a little earlier and a little later.

In order to comply with legal regulations on data protection, several measures were implemented to minimise the data used in the research. In this way, only strictly necessary metadata from the messages was used: channel ID, message ID, publication date and associated type of multimedia content. Comments associated with messages were not collected, nor were data about channel managers or subscribed users. The content of each message was used only to automatically identify the topic covered and then it was deleted. With regard to compliance with the EU Artificial Intelligence Act (Regulation 2024/1689), developments made for research purposes are outside of the regulation as they are not prohibited applications nor classified as high risk.
 

The authors wish to express their gratitude to Dr Javier Valls Prieto for his valuable contributions to the analysis of the ethical and legal aspects related to the collection and processing of data for this research.

ALAMMAR, J., and M. GROOTENDORST (2024): Hands-on. Large language models, O’Reilly Media, Inc.

ALTAY, S., B. A. LYONS, and A. MODIRROUSTA-GALIAN, A. (2024): «Exposure to Higher Rates of False News Erodes Media Trust and Fuels Overconfidence», Mass Communication and Society, 1-25.

BODAGHI, A. et al. (2024): «A Literature Review on Detecting, Verifying, and Mitigating Online Misinformation», IEEE Transactions on Computational Social Systems, 11(4), 5119-5145.

BUDAK, C. et al. (2024): «Misunderstanding the harms of online misinformation», Nature, 630, 45-53.

DAN, V. et al. (2021): «Visual Mis- and Disinformation, Social Media, and Democracy», Journalism & Mass Communication Quarterly, 98(3), 641-664.  

ECKER, U. et al. (2024): «Misinformation poses a bigger threat to democracy than you might think», Nature, 630(8015), 29-32. 

EUROPEAN COMMISSION, DIRECTORATE-GENERAL FOR COMMUNICATIONS NETWORKS, CONTENT AND TECHNOLOGY (2018): A multi-dimensional approach to disinformation. Report of the independent High level Group on fake news and online disinformation, Publications Office.

GRINBERG, N. et al. (2019): «Fake news on Twitter during the 2016 U.S. presidential election», Science, 363(6425), 374-378.

MALDITA.ES (2022, 19 de diciembre): El clan del negacionismo en Telegram.  

MCKAY, S., and C. TENOVE (2021): «Disinformation as a Threat to Deliberative Democracy», Political Research Quarterly, 74(3), 703-717.

MONTORO-MONTARROSO, A. et al. (2023): «Fighting disinformation with artificial intelligence: fundamentals, advances and challenges», Profesional de la información, 32(3).

NATURE (2024, 6 de junio): «Data access needed to tackle online misinformation», The International Journal of Science.

NOBARI, A. D. et al. (2021): «Characteristics of viral messages on Telegram; The world’s largest hybrid public and private messenger», Expert Systems with Applications, 168, 114303.

REGLAMENTO (UE) 2024/1689 DEL PARLAMENTO EUROPEO Y DEL CONSEJO, de 13 de junio de 2024, por el que se establecen normas armonizadas en materia de inteligencia artificial y por el que se modifican los Reglamentos (CE) n.º 300/2008, (UE) n.º 167/2013, (UE) n.º 168/2013, (UE) 2018/858, (UE) 2018/1139 y (UE) 2019/2144 y las Directivas 2014/90/UE, (UE) 2016/797 y (UE) 2020/1828 (Reglamento de Inteligencia Artificial).

RÍOS, C. (2024, 26 de marzo): «Paraperiodismo», El País.

TRILLO-DOMÍNGUEZ, M. et al. (2023): «SCImago Media Rankings (SMR): situation and evolution of the digital reputation of the media worldwide», Profesional de la información, 32(5).

URMAN, A., y S. KATZ (2020): «What they do in the shadows: examining the far-right networks on Telegram», Information, Communication & Society, 23(12), 1979-1998.

VOSOUGHI, S., D. ROY and S. ARAL (2018): «The spread of true and false news online», Science, 359(6380), 1146-1151.  

WARDLE, C. (2017, 14 de marzo): «Noticias falsas. Es complicado», First Draft.