Prashant’s Backstory

It started off as a different project during the pandemic, when the aim was to use geolocated Twitter data to understand social distancing. While working on this project, I (Prashant) realised how valuable Twitter data could be. My pivot to academics started when my supervisor suggested creating a ranking of how sustainable university communications were. To do this, I began collecting Twitter accounts of academics. Soon after, I came across a dataset merging Twitter IDs with another bibliometric dataset. This discovery spurred me to collect large-scale data on as many aspects of social media as possible for these academics, including their tweets, retweets, replies, and other metadata. Fortunately, all this was collected just before Twitter changed its policy and ended free academic access to Twitter data in 2023.

This all happened against the backdrop of rapidly growing technologies like Large Language Models (LLMs). There are countless text analysis methodologies that traditionally required researchers to develop highly specific skills in narrow text analysis methods. The trained machine learning models were also typically applied for very specific purposes. In my early experience using Twitter data, I employed a variety of dictionary-based methods and machine learning models that used the traditional test-and-train sample approach. With the introduction of LLMs, this changed fundamentally, significantly lowering the barriers to entry. Now, researchers who are non-experts in text methods can easily enter this space. 

While earlier methods still have value, they can be effectively augmented with LLMs. For example, one can use dictionary-based methods to classify topics in data—we do the same, but we use LLMs to craft those dictionaries, making them specific to context (e.g., time period, medium such as Twitter, and specific topics). 

Classifying political stance has also become easier with LLMs, as traditional models may not fully capture the nuances of context, particularly on platforms like Twitter, where sarcasm is common. From 2023 until today, the cost of using LLMs has decreased exponentially, while the quality of outputs has greatly improved. This democratising trend of LLMs is likely to continue. Without these advancements, our project would have taken significantly more time and would have constrained what we could accomplish, such as limiting the number of topics and stances we could study, as well as the accuracy of our findings.

Later in 2023, I met my co-author Thiemo Fetzer, who is now a co-author on many other papers and a close friend. With his guidance and experience, we focused the research more specifically on understanding how and what academics discuss on social media, comparing differences across subgroups and over time.

There are many decisions required to discipline oneself in descriptive research, especially for publication in a general-interest journal like Nature Human Behaviour. Being trained in economics, and with Thiemo frequently publishing in economics journals, this was a new experience for both of us. The editorial process was extremely helpful and efficient in guiding us in this direction.

With recent changes in Twitter (now X) and a shift among academics towards Bluesky, our findings may become increasingly relevant, especially as this platform transition appears largely driven by political factors.

Thiemo’s Backstory

I learned to code as a young kid when the internet was a creative playground—a space open to experimentation and genuine knowledge creation, long before it became a marketplace engineered to capture our attention and separate us from our data.

That early exposure made me notice not only the explosion of social media but also its impact on traditional media. News outlets increasingly leaned toward opinion over in‐depth reporting, while social platforms—once promising global connection—evolved into arenas of performative content, where algorithms and advertising metrics reward outrage rather than understanding. Even academics now adopt similar marketing strategies, sometimes at the expense of rigorous scientific inquiry.

This trend challenges the very notion of academic work. Research should not be “sold” like a product; truly good work will find its audience without compromising quality. Yet, a generation lacking proper digital literacy is more vulnerable to manipulation, struggling to navigate information ecosystems critically. Academics, who should be role models of critical thought, are also tempted by the instant rewards of likes and clicks over the slow, deliberate pace of deep research. This shift pushes scholars toward producing content designed to appeal broadly instead of focusing solely on knowledge creation.

At the same time, social media remains a powerful tool when used responsibly.  In fact, I got to know Prashant through social media. Having been a bit of an odd one in the economics profession due to my academic socialisation across computer science, maths and economics, it was so refreshing to finally be able to connect with early stage academics that share the passion and where this passion is substantiated with similar technical skills and journey. This has led to an incredibly rewarding journey of fruitful academic collaboration with LLMs now enabling a lot of research that previously was difficult to do, essentially working ourselves down a long list of research ideas that have been left out due to lack of computational and other resources. 

My own work, which gained attention through traditional media engagement, was catalyzed at times by the social sphere. This dual exposure has deepened my perspective on the challenge of transmitting research into wider society and the responsibility that comes with it.

Ultimately, this digital shift reshapes science by privileging visibility and immediate rewards over the invisible rigor of anonymous review. As social media platforms amplify emotional responses, biases creep in—reminding us that personal branding can, at times, outshine substantive contribution. By documenting these shifts in how we communicate on politically salient topics, we can better understand the reconfiguration of knowledge transmission and ensure that substance, not spectacle, remains our guiding principle.

Globally, some of the most exciting work in this sphere is unfolding right before us.

The Paper

Garg, P. & Fetzer, T. (2025), ‘Political Expression of Academics on Twitter’.  Nature Human Behaviour. Preprint available here.