BACKGROUND-AIM:
In September–October 2025, the European Commission opened a public consultation on the revision of the EU Tobacco Tax Directive (TTD), which received an exceptionally high volume of submissions across EU Member States and beyond. Public consultations have increasingly been used by the tobacco industry and its allies to amplify opposition narratives, dilute public health voices and strategically influence fiscal and regulatory reforms. This study systematically analyses the full dataset of consultation responses to identify sentiment patterns, recurring arguments, coordinated submission clusters and country-level trends in support for or opposition to the proposed reform.

METHODS:
We compiled and cleaned a dataset of 18477 consultation submissions scraped from the Commission’s “Have Your Say” portal. After pre-processing, we applied a mixed-methods natural language processing (NLP) pipeline comprising transformer-based sentiment analysis, BERTopic-based topic modelling, dense-embedding and hybrid retrieval–based similarity scoring, and near-duplicate detection to flag coordinated or templated responses. A stratified sample of 1,000 records was manually annotated for stance and key argument categories (e.g., illicit trade, cross-border shopping, economic impact, revenue loss, harm-reduction narratives), enabling supervised multi-label classification across the full dataset. Country-level sentiment and argument distributions were examined to explore geographic clustering of opposition.

RESULTS:
Analysis reveals strong cross-country heterogeneity: while some countries exhibited clusters of supportive responses to excise reform, others showed concentrated opposition to the TTD revision. Topic and word-frequency analyses highlight recurrent themes around illicit trade, novel products, and harm reduction. Similarity clustering identified multiple large groups of near-identical submissions, indicating coordinated or mass-distributed templates, particularly in countries with active tobacco or nicotine lobbies.

CONCLUSIONS:
Using advanced NLP techniques alongside systematic manual coding provides strong evidence of duplication and clustering patterns that may distort genuine public engagement. These findings highlight the importance of critically assessing the authenticity of consultation feedback and reinforce the need for greater transparency, appropriate weighting, and strengthened verification mechanisms in policy consultations. By revealing potential influence strategies, this approach enhances transparency in EU tobacco excise policymaking and supports more data-driven public health governance. Overall, the study offers empirical insights that can help strengthen the governance of public consultations in future tobacco tax reforms.