Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper presents the corrrelation between authoritarian discourse and several discoursive practices, including the rethoric of D. Trump. The paper contributes to the political sciences theoretical approach by ilustrating through rethoric design the recurrence of authoritarian perspective in contemporary leadership. 

As regards the introduction part we have several onservations:

1. the paper must present in 2-3 subsections the academic perspectives related to decline or erosion of democracy
2. the authors have to introduce academic refferences related to political communication, political marketing and the importance of political psychology perspectives to understand the authoritarian rethoric.

As regards materials and methods is neccesary to stress research hypotheses, questions and conceptual framework. Also, is important to highlight the thesis: what paper aims to demonstrate?

Empirical findings and methodology are accurate.

In the discussion section is necessary to interpret results and to create the nexus between results and research hypotesisis.

Present the limits of research and further directions.

Is necesaary to have a robust conclusion. 

Author Response

Dear Reviewer 1,

We sincerely thank you for your highly encouraging assessment of our manuscript. We are particularly pleased that you found the empirical findings and our methodological approach to be accurate and robust. Your constructive suggestions regarding the structural organization of the paper have been extremely helpful in improving the clarity, flow, and theoretical framing of our arguments. Below, we detail how we have addressed each of your observations.

Reviewer Comment 1: As regards the introduction part we have several observations: the paper must present in 2-3 subsections the academic perspectives related to decline or erosion of democracy. The authors have to introduce academic references related to political communication, political marketing and the importance of political psychology perspectives to understand the authoritarian rhetoric. Authors Response: we fully agree that a clearer structure and a broader theoretical framing enhance the manuscript. To improve readability, we have restructured the Introduction into three distinct subsections:

• 1.1. Theoretical Perspectives on Democratic Decline and Erosion
• 1.2. The Political Psychology and Communication of Authoritarian Rhetoric
• 1.3. Research Questions, Hypotheses, and Thesis Furthermore, in the new Section 1.2, we have expanded our theoretical framework to explicitly integrate perspectives from political marketing, political psychology, and communication. We added new key references (e.g., Iyengar et al., Norris & Inglehart) to explain how authoritarian rhetoric strategically exploits cognitive biases, out-group threat perception, and identity-based loyalties to bypass rational deliberation.

Reviewer Comment 2: As regards materials and methods is necessary to stress research hypotheses, questions and conceptual framework. Also, is important to highlight the thesis: what paper aims to demonstrate? Authors Response: thank you for pointing this out. We have now added a dedicated subsection at the end of the Introduction (1.3. Research Questions, Hypotheses, and Thesis) where we explicitly state our central thesis: that authoritarian democratic erosion leaves a mathematically measurable rhetorical footprint based on the structural alignment and intensity of extreme persuasion peaks. To formalize our conceptual framework, we have explicitly listed our primary Research Question (RQ1) and our two guiding Hypotheses (H1 regarding Structural Alignment, and H2 regarding the Intensity Threshold).

Reviewer Comment 3: In the discussion section is necessary to interpret results and to create the nexus between results and research hypothesis. Authors Response: we have revised the opening of the Discussion section to explicitly create this nexus. The discussion now systematically links our interpretation of the AI clustering and the Authoritarian Reference Index (ARI) directly back to the validation of our formalized hypotheses (H1 and H2), stating clearly how the results confirm our initial theoretical premises.

Reviewer Comment 4: Present the limits of research and further directions. Authors Response: we have highlighted this critical aspect by creating a dedicated subsection within the discussion (4.1. Limitations and Future Directions). In this new subsection, we explicitly discuss the constraints of maxima-vector representations, transcriptional noise, and the necessary future directions for testing this framework across cross-cultural and non-Western rhetorical ecosystems.

Reviewer Comment 5: Is necessary to have a robust conclusion. Authors Response: following your valuable advice, we have extracted the concluding thoughts from the Discussion and expanded them into a standalone, robust 5. Conclusions section. This new section summarizes our theoretical contribution to political science, reaffirms the core thesis, and highlights the practical value of the ARI framework as an early-warning system to guard democratic institutions.

We believe these structural and theoretical additions fully address your concerns and have significantly strengthened the manuscript. We thank you again for your time and excellent feedback.

Reviewer 2 Report

Comments and Suggestions for Authors

Let me start by saying that I am not an expert on LLMs and thus suggest strongly to the editors that they have this paper reviewed by someone with expertise to evaluate the computational methods directed at devising quantitative measures from speeches.

From what I can evaluate, I am of two minds about this paper.    On the positive side, the basic idea is intriguing, and it seems like a useful methodology for measuring authoritarian elements in speech. The authors make a case for analyzing certain moments in speeches, the high intensity, most authoritarian passages, instead of focusing on averages.   They belabor the case with a lot of repetition, but I can see the logic.   (That said, I strongly suggest that they incorporate into their analyses speeches by others, like Barack Obama, Tony Blair, or some other leader who no one would suspect of backsliding on democracy. This is an alternative baseline to Hitler, and perhaps endpoints two sides.).

 On the negative side, I would wish for two things.  First, the authors could compare speech elements to what the leaders actually do.   I understand that speeches matter and they help shape discourse and the mental frameworks people have when they listen to those speeches.   But the actions of leaders likely matter much more and this methodology offers an opportunity to link speeches to actions over time, and that would be a much more powerful statement.  That’s perhaps beyond the methodological goals of this paper, but I don’t fully accept all the claims made in the paper about descent into authoritarianism just based on speeches alone. 

Second, the authors strongly overstate their case in terms of the scientific validity of what they are doing. Ultimately, they are using quantitative measures but then applying qualitative categorizations.  The language used throughout the paper implies more rigor than I think is warranted. (e.g., Her inclusion proves mathematically that high political conflict and polarization do not automatically equate to authoritarian magnitude.  THIS IS NOT A CREDIBLE THING TO SAY, AND UNDERMINES THE PAPER QUITE A LOT.  A PROOF IS A THEOREM OR SOMETHING OF THAT SORT.)

My overall assessment is that this paper has promise but it needs to find a comparison to an obviously not authoritarian leader, and to soften its tone about the scientific rigor by which they authors ultimately use qualitative classifications.   The authors should also pay attention to actions of leaders.  The tone and attention should be more about the usefulness of the toolkit to discover patterns of speech that accord with actions of leaders. 

Note, there is a link in the paper which sends you to a website that ends up identifying the authors of this paper, or at least the researchers who are proposing this methodology. 

Author Response

Dear Reviewer 2,

We sincerely thank you for your critical reading and your assessment that the basic idea is intriguing and serves as a useful methodology. Your feedback has been instrumental in recalibrating the tone of the manuscript and identifying key areas for future research. Below we detail our responses to each of your points.

Reviewer Comment 1: Second, the authors strongly overstate their case in terms of the scientific validity of what they are doing... The language used throughout the paper implies more rigor than I think is warranted. (e.g., Her inclusion proves mathematically that high political conflict... THIS IS NOT A CREDIBLE THING TO SAY). / The tone and attention should be more about the usefulness of the toolkit to discover patterns of speech that accord with actions of leaders. Authors Response: we completely agree with this assessment. We realize that our previous phrasing inadvertently conveyed a sense of absolute mathematical proof, which is inappropriate for social science categorizations. Following your excellent advice (and in agreement with the Academic Editor’s request), we have thoroughly revised the manuscript to soften the tone. Specifically, we removed the phrase "proves mathematically" and substituted words like "proves" with "demonstrates" or "illustrates". We have also explicitly reframed our conclusions to emphasize the usefulness of the computational toolkit as a diagnostic instrument rather than claiming absolute algorithmic proof.

Reviewer Comment 2: I strongly suggest that they incorporate into their analyses speeches by others, like Barack Obama, Tony Blair, or some other leader who no one would suspect of backsliding on democracy. This is an alternative baseline to Hitler... Authors Response: we greatly appreciate this suggestion. In our research design, both Nicola Sturgeon and Giorgia Meloni were specifically selected to serve as our control points, effectively anchoring the "democratic-institutional pole". Sturgeon functions as our strict, "obviously non-authoritarian" democratic baseline, yielding very low intensity scores (22.3%) despite governing during a period of extreme political and territorial polarization. Meloni complements this as a critical test case for the "grey zone" (radical right populism); her profile yields an even lower intensity score (16.4%), mathematically proving that the model successfully differentiates between latent ideological alignment and active authoritarian magnitude. We recognize that analyzing leaders like Obama or Blair would be an outstanding addition; however, doing so would require a complete recalculation of the multi-dimensional space, PCA, and clustering. We have, therefore, clarified in the Corpus and inclusion criteria section how Sturgeon and Meloni effectively function as our control cases, while explicitly acknowledging that extending this toolkit to leaders like Obama is a logical and necessary next step for future research.

Reviewer Comment 3: First, the authors could compare speech elements to what the leaders actually do... this methodology offers an opportunity to link speeches to actions over time, and that would be a much more powerful statement. That’s perhaps beyond the methodological goals of this paper... Authors Response: you raise a crucial point. While creating a longitudinal correlation between our rhetorical index and empirical policy actions is indeed beyond the methodological scope of this specific paper, we completely agree that it is the most important next step. To address this, we have added a new paragraph in the newly created 4.1. Limitations and Future Directionssection. We now explicitly state that future research must systematically link these computational speech metrics with databases of empirical executive actions (such as V-Dem) to definitively measure how rhetorical peaks correlate with concrete institutional dismantling.

Reviewer Comment 4: Note, there is a link in the paper which sends you to a website that ends up identifying the authors of this paper, or at least the researchers who are proposing this methodology. Authors Response: we thank you for pointing this out. The link to the AI Democracy Observatory was included as an example of the practical implications of the toolkit. We apologize if this compromised the double-blind review process. We have ensured that the final publication version handles author identification in accordance with the journal's editorial guidelines.

Thank you again for your valuable critique, which has undeniably resulted in a more mature, balanced, and useful manuscript.

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for the opportunity to review this manuscript. I found the study both timely and engaging, and I appreciate the originality of applying LLM-based coding to a question of clear academic and societal relevance. I hope the following comments may be helpful as you continue to develop the manuscript.

First, the operationalization of the analysis through an explicit set of indicators is a clear strength of the research design, as it adds transparency and structure to the coding process carried out by the LLMs. At the same time, I wonder whether the absence of speeches drawn from other authoritarian traditions articulated through different ideological registers may place some limits on the conceptual robustness of the framework. This concern becomes more pronounced if Hitler serves as the historical “gold-standard” benchmark, since this may orient the measurement toward a historically specific prototype of authoritarianism rather than the broader range of forms identified.

Second, I would encourage the authors to reflect more fully on the decision to exclude Nicola Sturgeon’s COVID-19 emergency communications. I understand the methodological rationale behind this choice, particularly given the exceptional nature of the context and the need to preserve corpus comparability and avoid distorting the behavioural peaks later used in the similarity analysis. That said, it may have been useful either to retain these speeches or, at minimum, to conduct a sensitivity analysis including them. Because emergency contexts temporarily extend executive communication into the realm of crisis governance, they can provide a particularly revealing setting in which to observe tensions, shifts, and controversies within democratic systems themselves. In this sense, while the exclusion helps maintain the comparative homogeneity of the corpus, it may also leave aside a politically significant type of discourse for assessing how restrictions, appeals to authority, or exceptional measures are articulated within democratic settings. A brief additional discussion of this point, or even a complementary exploratory analysis, could further strengthen the interpretive scope of the study.

I hope these comments are received in the constructive spirit in which they are intended and prove useful in the revision of the manuscript.

Sincerely,

Author Response

Dear Reviewer 3,

We sincerely thank you for your thoughtful and constructive review. We are delighted that you found the study timely, engaging, and that you highlighted our explicit 11-indicator operationalization as a clear strength of the research design. Your comments regarding the conceptual limits of our baseline and the role of emergency communications are incredibly insightful. Below, we detail how we have integrated your reflections into the revised manuscript.

Reviewer Comment 1: I wonder whether the absence of speeches drawn from other authoritarian traditions articulated through different ideological registers may place some limits on the conceptual robustness of the framework. This concern becomes more pronounced if Hitler serves as the historical “gold-standard” benchmark, since this may orient the measurement toward a historically specific prototype of authoritarianism... Authors Response: you raise an excellent theoretical point. Our rationale for using Hitler as the "gold-standard" was purely mathematical: we needed a historical case that unequivocally maxed out all variables to establish the absolute ceiling (the outer boundary) of the 11-dimensional vector space. However, we completely agree that authoritarianism manifests through various ideological registers and that a single prototype does not capture the full spectrum of autocratization. To address this, we have expanded the newly created 4.1. Limitations and Future Directions section. We now explicitly acknowledge that relying on a single historical prototype is an empirical constraint, and we note that expanding the framework to include diverse authoritarian traditions (including far-left or non-Western autocratic models) is a necessary next step to fully validate the conceptual robustness of the model across different ideological geometries.

Reviewer Comment 2: Second, I would encourage the authors to reflect more fully on the decision to exclude Nicola Sturgeon’s COVID-19 emergency communications. I understand the methodological rationale... That said, it may have been useful either to retain these speeches or, at minimum, to conduct a sensitivity analysis including them. Because emergency contexts temporarily extend executive communication into the realm of crisis governance, they can provide a particularly revealing setting... Authors Response: we greatly appreciate this nuanced observation. You are absolutely correct that crisis governance presents a fascinating stress test for democratic rhetoric, specifically regarding how restrictions and appeals to authority are articulated without breaking the democratic framework. While we opted to maintain the exclusion in this foundational paper to preserve strict corpus homogeneity under routine political conditions (ensuring that the baseline clustering was not distorted by epidemiological mandates), we agree that this exclusion leaves aside a politically significant type of discourse. Following your suggestion, we have added a dedicated theoretical reflection in the Corpus and inclusion criteria section. We now explicitly discuss the unique value of emergency contexts and highlight that running a complementary analysis on crisis governance rhetoric is a critical avenue for future research to understand how democratic systems manage structural stress.

We hope these additions adequately reflect your excellent feedback. Your insights have undoubtedly broadened the interpretive scope of the study, and we thank you again for your constructive spirit.

Reviewer 4 Report

Comments and Suggestions for Authors

The submitted paper is focused on research on a significant topic related to the communication studies – the rhetoric of political leaders and comparison of their authoritarian behaviour. The research method includes LLMs and partially quantitative analysis of a topic which could be often researched through qualitative methods, which is an interesting approach.

The introduction is clear and well referenced and while the main hypothesis is clearly described, the research questions could be more clearly stated. The materials and methods are described to a level of impressive detail and address most possible questions or concerns regarding the chosen topic and research approach. The discussion part clearly states the risks of generalisation based on the limitations of the research, so the conclusions are only outlined in the context of the specific approach and selected research objects.

Besides the selected political leaders and the polarised messages they send, according to the results, an interesting analysis (for the sake of future research on the topic) would be one of the “evolution” of authoritarian messages over time – for example, using the same approach speeches from the first and second term of Mr. Trump could be compared, as well as speeches of Mr. Orban before and recently after he heads government and during and towards the end of his last term.

The GPT-computed graphics (especially Figure 1 and Figure 2) are well visualising the achieved results, but the images are of very poor quality and impossible to read.

Author Response

Dear Reviewer 4,

We sincerely thank you for your highly positive evaluation of our manuscript. We are thrilled that you found the application of LLMs to this topic interesting, and we deeply appreciate your encouraging words regarding the impressive level of detail in our materials and methods section. Your constructive feedback has been vital in putting the final touches on this article. Below we detail how we have addressed your observations.

Reviewer Comment 1: The introduction is clear and well referenced and while the main hypothesis is clearly described, the research questions could be more clearly stated. Authors Response: we completely agree. To address this, and in alignment with suggestions from other reviewers, we have restructured the Introduction to make the research questions mathematically and theoretically explicit. We have added a dedicated new subsection at the end of the introduction (1.3. Research Questions, Hypotheses, and Thesis), where we now formally outline our primary Research Question (RQ1) alongside our specific hypotheses regarding structural alignment and intensity thresholds. We believe this significantly improves the clarity of the paper’s objectives from the outset.

Reviewer Comment 2: Besides the selected political leaders and the polarised messages they send... an interesting analysis (for the sake of future research on the topic) would be one of the “evolution” of authoritarian messages over time – for example, using the same approach speeches from the first and second term of Mr. Trump could be compared, as well as speeches of Mr. Orban before and recently after he heads government...Authors Response: this is an outstanding suggestion. In fact, we share your intuition regarding the evolutionary nature of authoritarian rhetoric, which is why we included a preliminary comparison between Trump's first presidential term (2017–2021) and his early second term (2025) in our current dataset (as shown in Table 1 and the PCA clustering). To fully embrace your excellent point for future research, we have added a concluding sentence to our new 4.1. Limitations and Future Directions section. We now explicitly propose that applying this framework longitudinally to track the evolution of messages across consecutive terms (or pre/post-government stages) is a critical next step to understand how authoritarian persuasion adapts to changing institutional pressures over time.

Reviewer Comment 3: The GPT-computed graphics (especially Figure 1 and Figure 2) are well visualising the achieved results, but the images are of very poor quality and impossible to read. Authors Response: we sincerely apologize for this. The legibility issues were caused by automatic image compression in the word processor during the creation of the previous draft. Following your observation (and a concurrent request from the Academic Editor), we have entirely regenerated Figures 1 and 2. We have inserted high-resolution, uncompressed versions (300 DPI, vector-based where possible) into the revised manuscript to ensure maximum clarity and readability of the data clusters and radar charts.

Thank you again for your constructive, insightful, and encouraging review, which has helped us ensure the manuscript is presented at its absolute best.

 

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

For the points raised in my previous review:

Tone:  the paper is improved in backing away from overclaims about the rigor of the conclusions based on mathematical logic.  I can attest that they improved the paper in accordance with my suggestions.

Comparisons:  I had suggested that they find another politician to anchor the other side of the spectrum (opposite Hitler).  I am persuaded by their answer to my suggestion, however, that their existing analysis includes such politicians.

Language versus actions:   my suggestion to incorporate a comparison of actions to language was made with the knowledge this was likely beyond the scope of the paper.  The authors confirm that this is indeed the case.   It would be a more compelling paper if they could do this comparison but they have decided to limit their analysis to language. 

 

Overall: 

The paper is not much changed except for additional writing and some editing to soften the tone in places.     The authors did not do a major revision but more a minor one.  It’s up to the editors to decide based on my review above   I think accepting the paper for publication is reasonable, but with the following caveat:

I repeat my earlier suggestion in my first review that someone with LLM expertise review the methods carefully.  I do not vouch for the methods used and the conclusions drawn from the data analysis as I am simply not an expert. 

 

Author Response

Dear Reviewer 2,

 

We would like to sincerely thank you once again for your time, your thorough evaluation of our manuscript, and your constructive feedback during both rounds of review. Your insights have undeniably helped us present a more balanced and mature paper. Below we address your final observations.

 

• Reviewer comment 1 (tone and comparisons). Tone: the paper is improved in backing away from overclaims about the rigor of the conclusions based on mathematical logic. I can attest that they improved the paper in accordance with my suggestions. Comparisons: I had suggested that they find another politician to anchor the other side of the spectrum (opposite Hitler). I am persuaded by their answer to my suggestion, however, that their existing analysis includes such politicians.

Authors response: We are very pleased that our revisions regarding the tone of the manuscript successfully met your expectations. We also deeply appreciate your understanding and acceptance of our research design regarding the inclusion of Nicola Sturgeon and Giorgia Meloni as the necessary democratic/institutional anchors for our comparative spectrum.

 

• Reviewer comment 2 (language versus actions). My suggestion to incorporate a comparison of actions to language was made with the knowledge this was likely beyond the scope of the paper. The authors confirm that this is indeed the case. It would be a more compelling paper if they could do this comparison but they have decided to limit their analysis to language.

Authors response: We completely agree that correlating these rhetorical signals with concrete political actions is the most compelling next step for this line of research. While we had to limit the current foundational study to the validation of the rhetorical metrics, it was your excellent suggestion that prompted us to explicitly map out this exact trajectory in our new Limitations and Future Directions section. We thank you for highlighting this crucial avenue for future scholarship.

 

• Reviewer comment 3 (overall assessment and LLM expertise). The authors did not do a major revision but more a minor one. It’s up to the editors to decide based on my review above I think accepting the paper for publication is reasonable, but with the following caveat: I repeat my earlier suggestion in my first review that someone with LLM expertise review the methods carefully. I do not vouch for the methods used and the conclusions drawn from the data analysis as I am simply not an expert.

Authors response: We are grateful for your conclusion that accepting the paper for publication is reasonable. We fully respect and understand your caveat regarding LLM expertise. To ensure methodological rigor, our computational protocol was strictly designed following the latest established best-practice guidelines for evaluating Large Language Models in social-science research (specifically, Abdurahman et al., 2025). We have fully documented the prompt templates, scoring procedures, and version-controlled audit logs to ensure total reproducibility. Naturally, we leave any further methodological evaluation entirely at the discretion of the Academic Editor.

 

 

Thank you again for your critical reading and your vital role in improving this article.