Telegram has become one of the main platforms for sharing scientific information, but also a place where misinformation abounds. impersonation of academic publishersA new study conducted in Spain has put very concrete figures to a phenomenon that until now was suspected, but had not been measured in such detail.
According to this research, led by the University of Granada (UGR)Almost eight out of ten channels operating on Telegram using the names of major international scientific publishers are not official. We're talking about a 78% of fake channelsThis fact raises serious concerns in a context where scientific misinformation is already a major problem in Europe and the rest of the world.
A map of fraud in scientific publishers' channels
The research, conducted by the Computational Humanities and Social Sciences Unit (U-CHASSResearchers from the University of Granada (UGR) set out to map the Telegram ecosystem linked to major academic publishers. Those responsible for the work were the researchers. Victor Herrero Solana and Carlos Castro CastroThey wanted to verify to what extent the channels that present themselves as official really are.
To do this, they selected 13 leading international scientific publishersAmong them are leading names such as Elsevier, Springer, Wiley-Blackwell, Nature, and Cambridge University Press. The selection was not random: their volume of indexed publications on the SCImago portal, one of the leading international references in scientific metrics.
Once the publishers to be analyzed were defined, the team located a total of 37 channels that could be associated with these sealsThe objective was twofold: on the one hand, to determine if these channels were truly official; on the other, to study what type of content and practices were being developed by those impersonating identities.
The results were conclusive: of the 37 channels reviewed, only 8 turned out to be legitimate and be directly and verifiably linked to the corresponding publishers. That is, only 21,62% of the channels were authentic, compared to 78,38% of fraudulent channels that operated using names, logos or references of these institutions without any type of authorization.
A pioneering study using ChatGPT and DeepSeek
One of the most striking aspects of the work is the methodology employed. The UGR researchers have used, in a a pioneer in this field, artificial intelligence language models such as ChatGPT and DeepSeek to help detect these fake channels. The study was published in the academic journal "IDB: University Texts on Library and Information Science", in its December 2025 issue.
Far from being limited to a simple manual search, the team designed a multiple case scheme in which each identified channel was analyzed following a standard procedure. For this purpose, a standardized prompt which was applied equally to ChatGPT and DeepSeek, with the option of web search enabledso that these systems could compare information in real time.
The task of the AI ​​models was to value authenticity of each Telegram channel, taking into account indicators such as the relationship with official websites, the existence of verified accounts, the consistency between the published content and the brand's editorial line, or the presence of reliable corporate links.
After receiving the rankings provided by ChatGPT and DeepSeek, the researchers conducted a independent manual verificationwhich acted as the final reference (ground truth). That is, the final decision on whether a channel was fake or real was not left in the hands of the AI, but rather the AI ​​was used as a support tool that was then compared with expert human judgment.
How fake channels operate on Telegram
The analysis of the 37 channels allowed for the identification of a fairly clear pattern regarding How do those who impersonate scientific publishers operate?The most common tactic is the mass distribution of books and articles in digital format without authorizationoften presented as "free access" or "direct download" of titles that are actually subject to copyright.
In addition, many of these channels offer editorial services of dubious credibilitysuch as the promise of publishing scientific articles in high-impact journals in extremely short timeframes and with review processes that have little to do with actual academic practice. These types of proposals can be especially confusing to young or less experienced researcherswho are looking for quick ways to expand their resumes.
Another detected feature is the use of highly promotional and lacking rigorThe messages are more reminiscent of aggressive marketing campaigns than typical communications from a scientific publisher. The University of Granada (UGR) points out that this rhetoric, full of promises and discounts, is a poor fit for the way the academic publishing sector usually communicates.
In some cases, fake channels use logos, collection names, or shortened links that appear legitimate, which means that, at first glance, they can seem convincing to a user unfamiliar with the inner workings of publishing houses. This mix of professional appearance and irregular practices fosters an environment especially vulnerable to disinformation.
All of this, the study concludes, shapes a distorted ecosystem on Telegramwhere the presence of fraudulent actors far exceeds that of legitimate publishers. This imbalance increases the risk to academic integrity and intellectual property, both in Spain and in the rest of Europe, by facilitating the circulation of unauthorized content and misleading editorial promises.
What does artificial intelligence get right and where does it get wrong?
In terms of performance, the study indicates that both ChatGPT and DeepSeek showed a high capacity to detect clearly fake channelsWhen signs of impersonation were evident—for example, a total absence of official links, excessive promises, or openly pirated content—the models agreed to classify them as illegitimate.
However, the study also raises the structural limitations of these systems when it comes to validating authentic channels. The cases that generated the most doubts were those in which the channel did appear to be linked to a publisher, but lacked clear check marks, such as the blue checkmark on Telegram or explicit links to easily traceable corporate pages.
The authors detected that DeepSeek tended to focus more on contextual consistency Regarding the content: it checked whether the messages, the type of publications, and the tone matched what one would expect from an established scientific publisher. ChatGPT, for its part, placed more emphasis on the formal verification of institutional affiliationsprioritizing signals such as presence on official websites, linked profiles, or verified mentions.
This dual approach made it possible to observe that, although both models are useful for a initial screening of large volumes of channelsThey are not infallible. In particular, when strong signals of authenticity are lacking, AI can have difficulty differentiating between a real channel with little public information and a well-constructed fake one.
The report notes that, for now, the reliability of these models as standalone detectors For users without specific training, its use is limited. According to research, its best application is in hybrid systems where the massive analysis capabilities of AI are complemented by... expert opinion of librarians, documentalists and academic staff.
Biases in sources and hegemony of English content
Beyond the detection of fake channels, the UGR study focused on analyzing What type of sources do ChatGPT and DeepSeek consult? to support their answers. One of the conclusions was the strong presence of Western references versus other geographical areaseven in the case of DeepSeek, which might be presumed to be more oriented towards Asian sources.
This imbalance illustrates the hegemony of English content on the web, especially regarding scientific and academic information. Because they are mostly trained on data in that language and from specific regions, the systems tend to reproduce that distribution, which translates into a structural bias when it comes to identifying or assessing sources from other areas, such as China or other non-Western countries.
In practice, this can have significant implications for the evaluation of channels linked to non-Western publisherswhose websites, communication patterns, or verification systems may not align as well with the prevailing criteria in the English-speaking world. As a result, some legitimate channels may be classified with greater uncertainty or suspicion.
The authors of the study believe that this finding should be taken into account when design global monitoring tools AI-based research, especially in Europe, where scientific actors from diverse linguistic and cultural backgrounds coexist, risks exacerbating inequalities in the visibility and recognition of some institutions without specific corrective measures.
The study suggests that future research should explicitly address these shortcomings, whether training models with more balanced corpora or by adjusting the evaluation criteria to better adapt them to the diversity of the international academic system.
A high-risk environment for academic integrity
With all the data on the table, the research concludes that the universe of Telegram channels related to scientific publishers is deeply distortedThe majority presence of fake channels, compared to a small number of official accounts, creates a high-risk scenario for academic integrity and the protection of intellectual property.
Among the dangers detected is the uncontrolled dissemination of scientific materialThis not only violates copyright but can also encourage the circulation of outdated, incomplete, or manipulated versions of articles and books. At the same time, fraudulent publishing services erode trust in the scientific publishing system and can seriously damage the careers of those who fall into these traps.
The authors of the study speak of a genuine institutional paradoxWhile Telegram offers great potential as a robust communication and scientific dissemination channel, limited active and verified presence of the publishers themselves This leaves a void that malicious actors are exploiting without much resistance.
In the European context, where the fight against scientific misinformation While this is already a political and regulatory priority, these types of poorly regulated environments present an additional challenge. The ease with which channels can be created and content distributed on Telegram makes it particularly attractive to those seeking to leverage the brand of prestigious institutions.
Therefore, the work of the University of Granada functions not only as a diagnosis, but also as a call to attention to academic communities, libraries and regulatory bodies, who should consider these types of practices when designing integrity and open access policies.
Towards hybrid surveillance systems and new lines of research
Faced with this situation, the UGR researchers advocate for the development of hybrid detection systems that combine the capabilities of artificial intelligence with specialized human oversight. The idea is to leverage the computational scale of language models to comb large volumes of channels and content, but reserving the final decision for expert teams.
In this hybrid approach, AI would serve as initial mapping toolThis involves identifying suspicious patterns, repeated fraud tactics, or new accounts that mimic the identity of established publishers. From there, documentalists, librarians, and staff from the publishers themselves can confirm or rule out the detected cases.
The study also points to the possibility of expanding this type of methodology to other areas of disinformation, beyond the publishing field. The authors expressly mention the detection of fake news and conspiracy narratives on Telegram, both scientific and political in nature, which opens the door to future research that could be of direct interest to European institutions.
The progressive integration of advanced text and context analysis functions into language models offers a opportunity to develop proactive monitoring systemsThese systems could provide early warning of the emergence of new fake channel networks, facilitating a faster response from publishers, universities, or public bodies.
At the same time, the need for scientific publishers to become more involved in the building a solid official presence on TelegramVerified accounts, clear communication policies, and greater transparency in authorized channels would help users better identify reliable sources and reduce the scope for impersonators.
The work of the University of Granada highlights that the problem of the Fake channels of scientific publishers on Telegram It is not anecdotal, but structural, and addressing it requires combining technology, expert judgment, and the active involvement of the academic institutions themselves to regain ground in a digital space where, today, fraudulent actors have a clear advantage.
