The Opaque Path to Transparency: The DSA, VLOPs, and the Struggle for Democratic Integrity

Team Members

Facilitator: Varvara Boboc

Participants: Meret Baumgartner, Gizem Brasser, Afina Krisdianatha, Lisabeth Matyash Leonsins, Anders Puck Nielsen, Renze van de Putte, Ulrikke Dybdal Sørensen, Adel Tayebi

Designers: Sofia Bonfanti, Giulia Costanzo

Key findings

The multi-layered analysis reveals a persistent gulf between the Digital Services Act’s (DSA) ambition for auditable openness and the realities of TikTok ’s disclosures. Legally, the platform must meet strict obligations under Articles 14, 15, 24 and 42, yet its submissions oscillate between value-laden public claims (“transparency has been at the heart of our approach since 2019”) and narrowly framed, statistic-only filings that frustrate meaningful oversight. This evidence shows a shift from voluntary “trust-and-safety” rhetoric to what scholars label transparency-washing: the production of voluminous but low-value data that satisfies form while obscuring substance.

Quantitative comparison of January – March 2025 moderation data underscores this point. TikTok self-reports 211 million removals – eleven times the figure recorded in the EU’s DSA Transparency Database – because schemas, localisation fields and content categories fail to align. Vague statement labels such as STATEMENT_CATEGORY_SCOPE_OF_PLATFORM_SERVICE and the near-exclusive focus on “removed” videos (versus EU tracking of demotions, age-gating and adverts) block cross-verification, while 87%–96% automation rates remain impossible to audit without complete metadata.

Website-mapping amplifies these structural mismatches: EU DSA pages form a densely interlinked web that allows serendipitous discovery, whereas TikTok ’s Transparency Center and Community Guidelines sit in tightly gated sub-clusters, guiding users along pre-set paths and siloing critical material. Category labelling of 130 EU and 170 TikTok documents confirms divergent communication styles – Brussels balances enforcement, oversight and PR, while TikTok ’s corpus skews toward rule-citing compliance pieces and promotional messaging.

Finally, public-discourse mining shows two parallel lines of critique. Media and civil-society commentators fear the DSA could invite over-removal and political overreach, yet they simultaneously fault platforms for opaque, bad-faith reporting and restricted data access. Together, these findings suggest that without enforceable common data standards, independent audit rights and open researcher APIs, the DSA risks enshrining a new form of compliance theatre rather than delivering genuine democratic accountability.

1. Introduction

With global tech giants exercising increasing influence over public discourse, elections, and access to information, the European Union’s Digital Services Act (DSA) represents a landmark attempt to recalibrate this power dynamic. It has been hailed as the world’s most ambitious bid to make online platforms legibly accountable – introducing mandatory transparency reports, searchable ad-libraries and researcher‐access APIs, with the threat of fines up to 6% of global turnover, the regulation seeks to convert years of voluntary “trust-and-safety” marketing into auditable legal duties (Allen, 2022).

Among the twenty-three Very Large Online Platforms (VLOPs) designated under the DSA, TikTok stands out for both its scale, about 159 million monthly active recipients in the EU, and its rapid entanglement with the new rule-book. In December 2024, the European Commission opened formal proceedings over suspected failures to safeguard election integrity during Romania’s presidential vote, and in May 2025 it issued preliminary findings that TikTok ’s advertisement repository “does not fulfil DSA transparency obligations,” a charge that could trigger multi-billion-euro penalties (Chee, 2025).

Yet even before regulators weighed in, journalists described TikTok ’s glossy Los Angeles “Transparency & Accountability Center” as “smoke and mirrors” - a physical manifestation of transparency theatre designed to impress visitors without exposing the workings of the platform (Heath, 2023). Academic work labels such rituals as transparency-washing: the production of voluminous but low-value disclosures that satisfy formal rules while obscuring underlying logics (Zalnieriute 2021; Urman & Makhortykh 2023; Bassan 2025). Early independent audits lend credence to this critique: a recent study found that TikTok ’s Research API fails to return metadata for roughly one in eight requested videos, undercutting its utility for external verification (Entrena-Serrano et al., 2025). Similarly, the European Digital Media Observatory’s (EDMO) latest compliance review highlights gaps between company claims and verifiable evidence across all major VLOPs, including TikTok.

This study addresses the disjuncture between such corporate disclosures and the DSA’s ambition for verifiable, machine-readable, and researcher-accessible data. By comparing TikTok ’s reporting practices against the requirements and spirit of the DSA, we explore the evolving landscape of platform accountability. To do so, we conducted a multi-pronged digital methods study integrating document classification, natural language processing (NLP), transparency report analysis, and website architecture mapping. Specifically, we:

• Compiled and annotated documents from three key stakeholder sets: EU regulatory materials, TikTok ’s transparency and governance outputs, and critical public discourse from specialized media;

• Applied NLP and network analysis to identify thematic patterns and discursive trends across over 1,000 pages of text;

• Compared TikTok ’s transparency reports with entries in the EU’s DSA Transparency Database for Q1 2025, highlighting disparities in moderation data, reporting categories, and coverage; and

• Used web crawling to visualize and compare the user journey to access transparency-related content across TikTok and EU platforms.

These analytical lenses allowed us to triangulate legal obligations, public narratives, and actual information accessibility to better understand how transparency is constructed, communicated, and contested in the context of the DSA. They allow us to move beyond the platform’s carefully staged dashboards to an integrated view of legal obligations, public narratives, and lived information architecture.

Ultimately, this research seeks to clarify whether the DSA’s transparency regime can move beyond symbolic compliance and deliver genuine oversight, or whether it risks entrenching a new form of regulatory theater. By usingTikTok as a case study, the project contributes to wider debates on digital governance, algorithmic accountability, and the role of law in reshaping power in the platform economy.

2. Initial Data Sets

The initial dataset consisted of an expert collection of resources from the two key stakeholders: the EU and TikTok. The structure of the initial dataset was used as a backbone for the final dataset collection. The resources were considered and expanded once the direction was chosen. Thus, the initial data collection served as the starting point of the exploratory analysis. An inductive snowballing method was used to reach the final dataset for each entity. Most resources made their way into the final dataset collection, which was further enriched with a third stakeholder: specialized media as a representative of informed public discourse.

3. Research Questions

RQ: How effectively does TikTok ’s transparency reporting and content moderation align with the transparency requirements of the DSA?

• SRQ1: How does the DSA impart responsibility on transparency enforcement, and what is the main critique brought forth by the specialized public discourse?

• SRQ2a: How are different document categories (e.g., policies, articles, terms of service) used by the stakeholders to communicate transparency?

• SRQ2b: What are the key thematic issues raised by the authoring entities concerning TikTok ’s transparency and content moderation practices?

• SRQ3a: How can users access transparency information on the EU, EC and TT websites?

• SRQ3b: What information do TikTok ’s transparency reports disclose about its content moderation decisions, and how comprehensive and accessible are these reports?

4. Methodology

Methodological pipeline

The logic of the methodology stems from the multiple analyses which reflect various angles to unpacking transparency regulation and enforcement discourse under DSA for the key entities: the EU, the key VLOP – TikTok, and specialized media as a representation of informed public discourse. The methodological pipeline diagram below (fig. 1) showcases the steps, methods and outputs. Thus, starting from a provisional expert list of relevant EU and TikTok sources, an inductive approach was used to reach the final dataset. The DSA policy document was also used as a starting point, aiming to produce a transparency enforcement guide for the various responsible entities. In line with these starting points, four analyses were conducted:

• Document category labelling. Based on the final corpus, all sources were assigned a category. The category labels were developed through deductive manual annotation based on the content of the documents, which was then further refined to reflect more general themes.

• NLP textual analysis. Additionally, the content of the documents was used to perform various NLP based and network analyses to determine the key meaningful words and discursive arguments based on word frequency and various word window-collocation indicators.

• Transparency reports comparison. The transparency reports were collected from the EU website - the EU Transparency Database, and from TikTok ’s Transparency Center. The period chosen was the latest one available, January-March 2025. While the initial aim was a cross-verification of content moderation decision making, the data structures and categories did not allow for any meaningful comparison of numbers. What was interesting, however, was the discrepancy in the type of issues used to ground the content moderation decisions (statements of reason or community guidelines principles). Thus, we focused on comparing these types of issues and on drawing further conclusions based on the lack of comparability across transparency reports.

• Website mapping. Moreover, based on two starting points per each of the core entities (EU and TikTok), four website maps were developed through webpage crawling. The aim was to compare how access to transparency-related information is designed within the website architecture of TikTok and EU/EC. Upon selecting the starting points, the navigation of information is showcased.

Figure 1. Methodological pipeline

Analyses performed

Contextual Roadmap to Transparency

This research project focused on the DSA as the primary regulatory instrument and used a doctrinal, legal research methodology by examining the provisions of the DSA relevant to transparency reporting and content moderation obligations for VLOPs. The regulatory approach-based analysis systematically identified and interpreted relevant Articles of the DSA to determine the scope and procedural requirements and mechanisms applicable to TikTok as a VLOP. Furthermore, interpreting legislative texts under the DSA is substantially relevant to clarify TikTok ’s legal transparency reporting and content moderation obligations, as is the context of this project. The findings are structured to map these legal obligations onto TikTok ’s practices to form the legal basis for subsequent analysis on platform accountability and transparency within the EU's digital services and platform governance regulatory framework that is the DSA.

Website Map Comparison

Website Crawling

For the website crawling to compile the sitemaps, we first defined four starting points. For the DSA, the crawl starts at the main DSA page and the DSA FAQ, while for TikTok it starts at the Transparency Center and Community Guidelines.

From the starting page, the links pointing to other pages are extracted and then taken as a starting point for the next crawling iteration. This is repeated three times to build the sitemap from the starting page. As the EU websites have versions of each page for different languages, links referring to the same page in different languages are aggregated into one page.

For the visualisation of these networks, post-processing of the networks is done by determining whether a link between two pages has a parallel counterpart (the two pages refer to each other), and by designating pages as internal (part of the same website as the startpage, i.e. europa.eu or tiktok.com) or external.

Transparency Report Comparison

TikTok’s Community Guidelines Enforcement Report includes information about the platform’s moderation efforts and technologies. For the sake of comparison, we filtered by only including the period January - March 2025.

In addition to the daily CSV dumps available on the Transparency Database data download page, the DSA provides an aggregated report of the overall count of Statements of Reasons (SoRs) provided by all the different platforms. From this dataset we filtered by only including the entries submitted by TikTok from January - March 2025.

Initially, we attempted to map the two dataset schemas to each other in order to compare moderation results reported by TikTok itself and through the DSA. However, when exploring the data using Python analysis methods, we found too many discrepancies to merge the data in a meaningful way for a proper quantitative analysis.

EU and TikTok Category Assignation

ChatGPT-03 was used to read the content of all of the EU DSA documents listed in the EU final data set. The prompt was “create a table categorizing the following documents. Include the title and which category it belongs to. Start with technical, definitional, enforcement and regulation, and PR. Add any relevant categories you find.”

ChatGPT-03 produced a table with the information and identified the following categories: enforcement and regulation, technical/analytical, oversight and review, and public relations (PR):

• Enforcement and regulation included operative rules, obligations or compliance duties that a service must follow, binding rules, delegated and implementing acts, official guidances and guidelines, templates, or tools that operationalise compliance.

• Technical/analytical encompasses data-driven cost models, methodologies, evidence-based studies, impact assessments, annexes, or other in-depth technical material supporting policy choices.

• Oversight and review includes quality-control audits and audit reports, dashboards, risk assessment reviews, and any monitoring or scrutiny of the analytical base. PR covers public-facing framing such as press releases, FAQs, summarising external feedback, consultation summaries, calls for evidence, stakeholder events, recruitment notices, and any general explainer pieces. The same application was applied to the TikTok final dataset. Both resulting lists were verified manually to ensure that category assignation matched the document.

Specialized Media Discourse Analysis

Category Assignation

A selection of relevant tech-related publications and media sources was chosen with the purpose of providing different perspectives on the DSA from think tanks, mainstream media, civil society, and tech blogs. In addition, it was a priority to include both European and non-European perspectives. The chosen outlets were Algorithm Watch, Daring Fireball, DSA Observatory, Politico, Techdirt, Tech Policy Press, The Verge, Wikipedia, and Wired.

The search function was used to identify DSA related articles that were scraped using python scripts. The html code of the pages was saved, and relevant data was later extracted into a combined CSV file using a Jupyter Notebook.

Two separate tracks of analysis were performed. One filtered for those articles that mentioned TikTok, which allowed a comparison with the documents from EU and TikTok mentioned above. The other track was a broader analysis of the public discourse about the DSA and the transparency requirements.

We identified different formats of documents in public discourse. First using ChatGPT to attribute categories and then manually annotating them for meaningful attribution, which revealed the following categories: regulatory coverage, policy advocacy and commentary, and policy monitoring and evaluation. Regulatory coverage included articles and other documents reporting official actions or news on regulatory updates, codes, designation decisions, and others regarding both the European Union and platforms. Policy advocacy and commentary encompassed essays, blog posts, and documents arguing a position or influencing the debate on the digital services act, transparency, and/or content moderation. This includes calls to action, reporting opinions, and advocacy efforts. Policy monitoring and evaluation documents include outputs that track, benchmark, or evaluate platform and regulatory behaviour.

Textual Analysis for Public Discourse about TikTok

To analyze the three corpora, a hybrid approach using a generative AI model and a combination of Python libraries was applied to verify the semantic aspects of the quantitative outputs and extract concepts. Data for network analysis was also gathered using traditional statistical NLP methods. Nodes and edges were used to visualize raw data. A central objective of the hybrid script is to conduct a deep thematic analysis on a per-file basis. By moving beyond simple word counting, it identifies and validates multi-word phrases and concepts that are central to the document's meaning. The script processes each file individually through a multi-stage pipeline:

Statistical Candidates: preprocesses the text (lemmatization, stop-word removal) and identifies all two to five word sequences (n-grams) that appear more than ten times. This captures statistically significant repeated phrases.

1. Semantic Candidates: uses the KeyBERT model to analyze the text and extract the top 500 most semantically relevant phrases, regardless of their frequency.

2. Unification: These two lists are combined and de-duplicated to create a comprehensive pool of candidate phrases.

3. LLM Validation: the unified list of candidates is passed to a LLM (using Mistral 7b through Ollama). For each phrase, the LLM is asked: "Is this a meaningful multi-word concept?" This acts as a powerful qualitative filter, removing grammatically correct but nonsensical n-grams.

4. Output Generation: validated phrases; the final LLM-approved phrases are saved to a CSV file.

5. Network Files: calculated word frequencies and saved two node and edge files to represent the relationships between top 100 frequented words. This process was also used to generate nodes and edges for all words and words with more than ten frequencies.

The goal is to visualize and compare the lexicon of the language used by each stakeholder. Its output is designed to be imported directly into network analysis tools to map the relationships between words, identify central terms, and discover linguistic clusters, which directly support the research question about "discursive trends" and "network analysis."

Textual Analysis for Public Discourse about the DSA Transparency Initiatives

The public discourse part of the analysis consisted of NLP processing using LLM machine learning. The Gemma3:12b model was used through Ollama to create summaries, provide keywords, and assess the sentiments towards the DSA in each article. Semantic groupings of similar keywords were made using a combination of manual group assignment (e.g. "Digital Services Act": ["Digital Services Act", "DSA", "Digital Services Act (DSA)"]) and automatic grouping using sentence transformers.

The grouped keywords were used to generate maps of the most discussed topics in the public debate, which were processed in Gephi. A total of 352 keyword groups were identified with a total of 6123 co-occurrence edges.

A list of points of criticism of the DSA was identified by running a CSV file with the summaries of the articles classified as negative or somewhat negative toward the DSA through ChatGPT -4o. The points on the list were verified manually for both relevance and accuracy, and the list was finalized manually for identification of the most important points related to the DSA transparency initiatives.

5. Findings

Contextual Roadmap to Transparency

Contextualizing the DSA

The DSA was created to provide a safer and more transparent digital environment within the EU. As a key piece of legislation, the DSA aims to harmonise the legal framework for digital services and platforms, ensuring that users are better protected and that platforms are held accountable for the risks they may pose in the online environment. It entered into force on 16 November 2022, establishing clear regulatory obligations for online service providers operating within its broad scope, including online content, transparency for users, and managing systemic platform risks.

TikTok falls under the DSA because it has been designated by the European Commission as a VLP under Article 33, meeting the threshold of at least 45 million average monthly active users in the EU. Furthermore, the DSA has a broad territorial scope (Article 2(1)), which means it applies to platforms regardless of where they are established, as long as they have users in the EU. As a result, despite being headquartered outside the EU, TikTok is subject to the DSA and must comply with its regulatory requirements.

Identification of applicable provisions

In terms of specific legal obligations under the DSA, Article 14 requires TikTok to set out clear and lawful terms and conditions, including all information that must be made available to users. This obligation connects with Article 27 on recommender system transparency, requiring TikTok to disclose which parameters are used within its recommender systems and to incorporate this information into its terms and conditions. Additionally, TikTok must provide functionality that allows users to select and modify the types of recommender systems they wish to use, increasing user autonomy over content curation on the platform.

When it comes to transparency reporting, the DSA sets out clear obligations for platforms like TikTok. Under Article 15, TikTok is required to report on its content moderation activities, providing insight into how the platform handles illegal and harmful content. Article 24 adds to this by requiring TikTok to publish reports every six months on the average number of monthly active users in the EU, while also including information on the number of disputes brought to out-of-court settlement bodies, actions taken against repeat offenders, and user appeals concerning content moderation decisions.

Article 42 brings these requirements together by ensuring that platforms comply with the reporting obligations in Articles 15 and 24, in addition to publishing audit reports (as required under Article 37(4)) and the outcomes of risk assessments (Article 34). However, Article 42(5) allows TikTok to withhold certain sensitive information from the public versions of these reports if disclosing it could compromise business confidentiality, user privacy, or public safety. Even so, TikTok is still obligated to provide the complete reports to the European Commission and the Digital Services Coordinator. This balance between transparency and confidentiality illustrates a tension within the DSA framework, showing the need for effective regulatory oversight to ensure that platforms – i.e. TikTok – remain accountable.

Figure 2. The Contextual Roadmap of the DSA

Website Map Comparison

DSA Network Maps

For the EU’s main page describing the DSA, we can see that the sitemap contains a lot of pages, with many parallel links which describe pages directly pointing to each other. This high degree of referentiality reflects the experience of navigating the DSA pages on the EU website, where pages sometimes refer to themselves or lead you on a journey that ends where you began - almost like a snake eating its own tail.

For the DSA FAQ page, the number of pages is smaller and some central pages that are more often referred to can be identified. Furthermore, this network contains relatively many external pages, which, as we can see by examining the dataset, refer to the EU’s social media channels, but also other external sources such as government websites of EU countries.

Figure 3. EU Network Maps

TikTok Network Maps

The network maps generated from both the TikTok Transparency Center home page and the TikTok Community Guidelines are very similar, with it being possible to identify the same subclusters across both maps. This suggests that these two startpages are (indirectly) referring to the same pages and are closely clustered together within the TikTok information structures. There are some external pages that are being linked in these maps, but the majority of the pages are internal to TikTok. There are parallel links that indicate that pages directly refer to each other, but these are mainly localized to sub-clusters.

Figure 4. TikTok Network Maps

Comparison between DSA and TikTok

While the DSA-oriented site maps have one big centralized cluster of nodes with direct cycles across many pages in this cluster (fig. 3), the TikTok sitemaps are made out of clearly identifiable sub-clusters within which parallel links exist (fig. 4). This could indicate that the TikTok website guides, or steers, the user more than the EU, which seems to connect many pages to many other pages. While this could make it easier to use the TikTok website to retrieve certain information, it could also mean that certain information can only be found if the user knows in what sub-cluster (or area of information) to look, and they are more likely to incidentally “happen upon” information in the EU website. Furthermore, it could reflect differences in the ways that TikTok and the EU portray data transparency. The sitemaps suggest that TikTok portrays the topic as a group of loosely connected subjects, while the EU portrays data transparency as one intricate and complex topic.

Transparency Report Comparison

The DSA Transparency Database provided by the EU and the January – March 2025 Community Guidelines Enforcement Report independently provided by TikTok don’t neatly overlap in the types of data collected.

Localization

Users and creators inside and outside the EU interact with each other on TikTok. This however, complicates content moderation and data collection. If a video by an EU creator is restricted in, for example, the USA, this isn’t included in the DSA dataset (DSA Article 17). If instead, a US creators’ video is restricted in the EU, this is included in the DSA dataset (Drolsbach & Pröllochs, 2024). Besides the data being of EU origin, the EU dataset doesn’t provide any specific geographic information such as country of origin.

TikTok has two types of localization categories: language and location. For location, it only has four EU countries: Italy, Spain, Germany and France. TikTok claims that the 50 worldwide locations included in the dataset contribute to 90% of globally removed content.

This makes it impossible to compare the two datasets. To illustrate this, we compared the total number of videos removed in the period of January – March 2025. In the TikTok dataset, 211 193 115 videos were removed, while in the DSA dataset, 18 605 201 videos were removed; meaning that the TikTok dataset contains more than eleven times the number of videos removed than what the DSA dataset shows over the same period.

Statement Categories

A drawback of the DSA datasets that we encountered and that has also been noted by other researchers are the Statement Categories. According to the DSA, these categories should be “clear and specific” (Kaushal et al., 2024), but one of the most used categories for TikTok, as well as YouTube, Facebook, Instagram, and LinkedIn, is STATEMENT_CATEGORY_SCOPE_OF_PLATFORM_SERVICE (Trujillo et al., 2025). This category, which does not have a definition in the DSA documentation, gives platforms a possibility of obscuring their moderation practices and the rationale behind them by grouping most decisions under this category (transparency washing).

Type of Moderation

When it comes to the types of moderation included in the two datasets, the EU DSA dataset is more comprehensive. It includes five types of moderation, namely: remove, disable, demote, age restrict, and other. However, the TikTok dataset for the most part merely covers ‘removed’ content.

Type of Data

Another asymmetry between the two datasets was the type of data collected. This is because the EU dataset encompasses far more types of data. The EU dataset includes audio, image, text, account, sticker, and advertisements. In contrast, the TikTok dataset doesn’t include content moderation affecting audio, stickers, and images.

Type of Interaction

The datasets also differ when it comes to measuring interaction with content. Firstly, the TikTok dataset includes more statistics about intricate interactions such as ‘percentage of moderators assigned’ or ‘fake follow requests prevented.' Secondly, TikTok provides more overarching statistics that encompass interaction with many different types of content such as ‘removal rate before any views.'

Automated Detection

Both the DSA and TikTok datasets include statistics about automated versus human moderation. When comparing the two from January - March 2025, we found a considerable discrepancy. TikTok’s data specifies that 211 193 115 total videos were removed, of which 184 378 987 videos were removed through automated detection – 87.3%. Whereas, according to the DSA, 96.34% of videos were automatically removed.

Ads

In the DSA dataset for TikTok, there is no moderation for ads included, although ads are defined as a content category in the DSA schema. However, TikTok does indeed moderate ads as 1 700 504 were removed from January - March 2025. Furthermore, the European Commission has started proceedings against TikTok for failing to disclose an ad repository.

Given the lack of consistency in numbers reporting across the EU and TikTok transparency reports, the only comparable element is the types of issues surrounding content moderation decision-making. Figure 5 indicates partial overlap between the DSA and TikTok content moderation guidelines issues, but TikTok has a more granular approach.

Figure 5. Content moderation decision-making issues: a discrepancy comparison between the EU and TikTok transparency reports

EU and TikTok Category Assignation Results

Based on the prompt, ChatGPT -03 identified the categories: enforcement and regulation, technical/analytical, oversight and review, and PR. It discarded definitional, expanded technical to include analytical, and added oversight and review. Below are tables delineating the number of documents and percentages identified in each category.

This includes the full text of the DSA is listed in the EU dataset as well as the individual articles. For the purpose of our analysis, the DSA full text was counted but not the individual articles as listed in the complete dataset. Discounting the individual articles delineated (109 total, all assigned enforcement and regulation), the EU dataset has a total of 45 documents and the TikTok dataset contains a total of 85 documents.

Table 1: EU Category Assignation

Table 2: TikTok Category Assignation

Discourse Analysis

Category Assignation Results

Based on the categorisation our data was distributed in the categories according to the table below, delineating the number of documents and percentages identified in each category.

Table 3: DSA Discourse Category Assignation

Table 4: TikTok Discourse Category Assignation

The distribution of categories in the above showcase the full dataset for the public discourse on the DSA. However, the distribution of categories for articles mentioning TikTok follows the one for public discourse on the DSA.

Figures 6 & 7. Category assignation of public discourse on EU and TikTok sources

The dominating category is regulatory coverage with more than half or almost half of the articles reporting official actions or news on regulatory updates, codes, designation decisions, and others regarding both the European Union and platforms. This can be due to the articles tracing back to 2020 when the DSA was first proposed by the European Commission, introducing the regulatory framework.

Following its relatively recent adoption, the regulatory coverage is meaningful as the regulatory framework itself has been up for discussion in terms of efficiency and legislative quality. Furthermore, there have been continuous updates in relation to the DSA including the Code of Practice in Disinformation, Implementing Regulation for harmonising standards for transparency reportings, and the Delegated Act on data access.

Public Discourse on the DSA

The keyword analysis of the public discourse about the DSA in general revealed a range of topics discussed but as the fig. 8 shows, the most prevalent keywords related to topics such as content moderation, transparency obligations, disinformation, and free speech. This indicates that transparency and content moderation were big issues in the public discourse. Furthermore, there was a range of prominent keywords related to specific platforms such as TikTok, Facebook, and X.

Figure 8. Keyword analysis of the public discourse about the DSA

Public Criticism of the DSA Transparency Initiatives

The sentiment analysis revealed overwhelmingly negative or somewhat negative coverage of the DSA in the public discourse. This is perhaps not surprising, given that the media tends to focus on controversial aspects of regulation, but it is still noteworthy that over 73% of the articles were assessed to have a negative or somewhat negative angle on the DSA.

Figure 9. Public discourse sentiment analysis regarding DSA measures

The LLM analysis revealed a list of different types of criticism. Many articles discussed the concern that the DSA can be a threat to freedom of expression and potentially be a censorship tool. A dispute between EU commissioner Thierry Breton and Elon Musk spurred discussion about how the DSA can be used as a tool for silencing political opponents and gives too much discretionary power to EU officials. There was also a discussion about how the purpose of the DSA as a tool of regulatory harmonization across the EU was undermined by states implementing additional rules on top of the DSA. This was specifically exemplified by Romania’s reaction to the election interference.

The qualitative analysis narrowed the criticism of the DSA down to the following four points:

1. Promoting Censorship: It introduces impractical and unrealistic standards for content moderation, leading to discretionary overreach and over-removal, resulting in censorship to avoid penalties.

2. The Opacity and Power of the Regulator: The EU Commission are misusing the DSA for political pressure and have opaque decision-making processes. Additionally, the DSA gives too much discretionary power to EU officials, contains vague or inconsistent enforcement, and provides vague definitions and methodology, leading to disproportional application.

3. Contradictory Harmonisation: Member States initiate additional national rules and efforts, leading to a patchwork of platform rules which contradicts the harmonising purpose of the DSA.

4. Red Tape: The extensive obligations put on platforms lead to overreliance on automated systems, especially as it contains technically unrealistic demands of content removal and burdensome moderation requirements.

We used a similar approach to identify the most important criticisms raised against the VLOPs in the public discourse. They were regularly criticized for a failure to tackle disinformation effectively, having a resistance to oversight, lacking transparency, and undermining their legal responsibilities. After the qualitative analysis we narrowed down the criticism of the VLOPs to the following points:

1. Lack of good faith in reporting: Platforms' reporting practices are inconsistent and opaque due to selective disclosures that often lack real auditability.

2. Insufficient data access: It is not possible to monitor that the platforms do as they claim to do, since there is no useful or meaningful data access granted to scrutinize.

Together, these two sets of criticism illustrate vantage points in the debate. One looks at the EU from the outside and highlights risks in the implementation of the DSA. The other focuses on the platforms and their resistance to the transparency initiatives.

Comparative Content Analysis on TikTok under the DSA

TikTok Textual Analysis

Figure 10 visualizes, via a Gephi network, the most frequent terms occurring in the TikTok textual dataset, revealing a lexicon that closely echoes the wider DSA conversation while remaining platform-specific. TikTok ’s dataset suggests that the platform attempts to shape its discourse around its VLOP service provider attribution. Relevant nodes include “platform,” “data,” “content,” “company,” “user,” “transparency,” “report,” “research/researcher,” and “risk,” indicating that public discussion of TikTok is produced discourse that attempts to be in alignment with governance, compliance, and accountability frames rather than product features or culture. The prominence of “report” and “researcher” positions TikTok as an object of study and oversight; “risk” and “transparency” tie that scrutiny to DSA expectations of disclosure and mitigation. Although this vocabulary overlaps with both the EU and broader public-discourse corpora, term frequencies differ, suggesting partially distinct emphases in how platform governance is approached across communities. The network’s dense central cluster reflects argumentative, policy-oriented language, consistent with advocacy and commentary around costs, burdens, and promised benefits of regulation. In short, the TikTok map shows a discourse gravitating toward accountability and risk framing, with co-occurrences that bind platform operations (“content,” “user,” “data”) to DSA-inflected expectations of transparency and evidence-based assessment.

Figure 10. Network graph of the most frequent words used in the textual dataset from the TikTok dataset

EU Textual Analysis

Figure 11 presents the EU textual dataset’s most frequent words as a Gephi network. Its core vocabulary substantially overlaps with the TikTok corpus: terms such as “platform,” “data,” “content,” “user,” “transparency,” “report,” and “risk”, but the relative weighting of these words shifts, indicating that the institutional discourse places slightly different stress across shared themes. The configuration centers on a compact argumentative cluster, reflecting the policy-commentary register that characterizes EU communications about platform governance under the DSA. This cluster structure signals how concepts are mobilized together: “transparency” and “risk” circulate alongside references to reporting and research, aligning oversight with evidence production and comparative assessment. While these overlaps complicate claims of wholly distinct language communities, the frequency differences imply divergent vantage points: EU texts formalize governance and monitoring, whereas other corpora invoke similar terms to debate implications or evaluate platforms. Overall, the EU network illustrates a discourse that organizes around transparency, risk, and reporting requirements, linking user- and content-level concerns to regulatory objectives without departing from the shared lexicon observed across datasets.

Figure 11. Network graph of the most frequent words used in the textual dataset from the EU dataset

Specialized media textual analysis

Figure 12 shows a Gephi network of the most frequent words in the specialized media dataset. The vocabulary mirrors the wider DSA discussion and the other corpora in this study. Core terms include “platform,” “data,” “content,” “DSA,” “company,” “user,” “transparency,” “European,” “report,” “research/researcher,” and “risk.” This overlap indicates that specialized outlets frame TikTok mainly through regulation and oversight, rather than product features or culture. At the same time, the relative weight of these words differs from the TikTok and EU networks, suggesting a slightly different emphasis in how topics are approached. As with the other figures, node size should be read according to the sizing rule used in the graph (term frequency or co-occurrence degree), and edges represent co-appearances of terms in the same texts. The network structure features a compact central cluster, consistent with argumentative, policy-oriented language that highlights costs and benefits of the DSA. Read alongside Figures 10 and 11, the message is consistent: specialized media take up the same core concepts – transparency, risk, reporting, users and content – but adjust their emphasis when discussing platform governance. Overall, the figure depicts a discourse that remains close to regulatory themes while reflecting the editorial focus of outlets that track policy and industry developments. Moreover, the map in fig. 12 shows a relatively large central cluster. Exploring this cluster, many words are argumentative in tone, consistent with policy advocacy and commentary within the discourse. In the negative framing, specialized outlets surface concerns such as “burden,” “compliance cost,” “regulatory overreach,” “censorship,” “stifle innovation,” and “competitive disadvantage.” In the positive framing, they highlight expected benefits including “enhance transparency,” “improve accountability,” “strengthen oversight,” “democratic oversight,” “public interest,” and “user rights.” Together, these poles organize debate around trade-offs between compliance costs and regulatory goals. Read alongside Figures 10 and 11, the figure shows that specialized media work with the same core concepts but adjust emphasis when weighing risks and benefits of DSA implementation.

Figure 12. Network graph of the most frequent words used in the textual dataset from the specialized media dataset

Figure 13 summarizes these trends using a comparative keyword analysis across stakeholders. Taken together with the previous textual analyses, the results show considerable discursive similarity between the EU and TikTok, indicating how the VLOP is mimicking compliance by replicating language (figures 10-12). The lack of technical details to support the compliance perspective highlights a lack of transparency around the practicalities of transparency enforcement.

The document categories further prove the trend. Created based on the documentation purpose, the categories were inductively developed to keep track of the types of content being produced to frame transparency regulation and enforcement.

Figure 13. Comparative keyword analysis

6. Discussion

The DSA was sold as Europe’s answer to the chronic accountability gap of global platforms: it converts voluntary “trust & safety” marketing into enforceable duties with audit trails and sanctions. Yet a body of scholarship warns that mandatory disclosure can still be gamed through transparency washing - the strategic production of voluminous but low-value data that looks open while shielding the core logic of the business (Zalnieriute 2021; Urman & Makhortykh 2023; Bassan 2025). TikTok ’s first DSA cycle illustrates how that risk plays out in practice.

Although the DSA positions transparency as a cornerstone of platform accountability, its implementation raises structural and practical concerns. Critics warn that stringent moderation requirements could incentivise over-removal of content, effectively fostering censorship. At the institutional level, the regulation is seen as concentrating discretionary power within the EU Commission, with decision-making processes and enforcement standards remaining vague. The intended harmonisation is further undermined by divergent national measures, creating a fragmented regulatory landscape. The compliance burden, particularly the technically unrealistic demands for rapid content removal, risks driving overreliance on automated moderation. From a transparency standpoint, the regulation’s efficacy is questioned: platform reporting remains selective and difficult to audit, while limited data access precludes independent verification of compliance. In this view, the DSA may function less as a levelling mechanism and more as an instrument for expanding EU regulatory reach.

As such, the question becomes – rather than providing a regulatory roadmap to transparency, does the DSA introduce a guide to transparency washing?

Value-inflated discourse vs. compliance-sounding PR

TikTok’s outward-facing language is drenched in values. In its fourth DSA transparency report, it claims that “transparency has been at the heart of our approach since 2019,” pairing the statement with glossy metrics of 18 million removals and 99 % accuracy.

A similar aesthetic animates the company’s Los Angeles “Transparency & Accountability Center,” which reporters described as “smoke and mirrors designed to give the impression that it really cares” (Heath, 2023). Such value-inflated discourse tells lawmakers and users, in effect, “trust us—we share your democratic ideals.”

When Brussels knocks, however, the tone changes. Submissions to the Commission rely on tightly framed legal citations (“we comply with Article 24/39”) and statistical snapshots devoid of context – what Bassan (2025) calls the next wave of “letter-of-the-law” transparency washing. The switch from lofty values to minimalist legalese exemplifies how platforms toggle between legitimacy-seeking rhetoric and compliance-sounding PR as audiences change.

Decision-maker vs. enforcer

The DSA tries to invert the historical power balance - moving platforms from quasi-regulators to regulated entities - yet TikTok ’s case shows the struggle is ongoing. In February 2025 the Commission opened formal proceedings over election-integrity risks, citing suspected failures to mitigate foreign interference in Romania’s presidential vote (Blenkinsop, 2024). Three months later it issued preliminary findings that TikTok ’s ad-repository - the very database meant to let outsiders track political targeting - omitted essential fields and was difficult to query, a potential breach of Articles 39 and 40.

These episodes highlight a structural tension. TikTok still controls the data architecture and chooses what to reveal (decision-maker), while the Commission can only react post-hoc (enforcer). Fabbri (2025) already warned that vague requests for information under Article 54 risk becoming “data dumps”: voluminous spreadsheets that satisfy the letter of the law yet bury the signal regulators need.

Obstacles to verifying veracity and cross-platform comparison

Even when data is supplied, validating them is arduous. Independent audits of early DSA reports show gaps in comparability and reproducibility (Urman & Makhortykh 2023; Kaushal et al. 2024). TikTok ’s Research API – touted as a transparency milestone - has been labelled an “access illusion.” Researchers report a single credential per organisation, opaque approval criteria, and missing metadata for roughly 12% of requested videos (Entrena-Serrano et al., 2025).

Meanwhile, EDMO June 2025 assessment finds a “clear gap” between company commitments and verifiable evidence across Google, Meta, Microsoft and TikTok, cautioning that the Code of Practice (soon to be a DSA code) risks remaining “performative.” Taken together with Nahrgang et al. (2025) mapping of labyrinthine community-guideline prose, the picture is one of information oversupply but insight scarcity – classic transparency-washing.

Implications for democracy

The stakes are not academic. Romania became the first EU member to annul an election over suspected manipulation “particularly on TikTok,” prompting the Commission’s emergency statement on mis- and disinformation. Algorithmic curation that is opaque to researchers but finely tuned for engagement can accelerate polarisation and disenfranchise already-marginalised voices - especially when hateful or gender-targeted harassment is left under-moderated (Urman & Makhortykh 2023).

If platforms can meet disclosure obligations through symbolic gestures, democratic oversight suffers on multiple fronts:

• Risk assessment becomes one-sided. Regulators rely on platform-filtered data, weakening their ability to set evidence-based policy.

• Research pluralism is curtailed. Opaque APIs and legalistic terms crowd out independent replication and cross-platform studies, a prerequisite for systemic risk mapping.

• Public trust erodes. Voters see headlines about investigations and fines, yet remain in the dark about what actually happens backstage – a breeding ground for conspiracy theories that further fracture the infosphere.

Looking ahead

The DSA gives the Commission sharper teeth – fines up to 6% of global turnover and the power to impose “periodic penalty payments” (Bassan 2025; EU Commission 2025). But enforcement will be a running battle until meaningful transparency: machine-readable ad libraries, standardised metrics, independent audits, and open researcher access – replaces ornamental disclosure. Otherwise, Europe risks codifying a new compliance theatre where corporate disclosure dashboards facade as democratic accountability.

7. Conclusions

Taken together, our findings show that TikTok ’s transparency narrative is lexically aligned with the DSA but practically contested in emphasis and use. Across EU texts, TikTok materials, and specialized media, the same vocabulary dominates – platform, data, content, DSA, company, user, transparency, European, report, research/researcher, risk – which signals convergence on what counts as “transparency.” Yet the network maps also reveal differing frequencies and a dense, argumentative core in which the specialized press openly weighs costs (e.g., burden, compliance cost, regulatory overreach, censorship, stifle innovation, competitive disadvantage) against benefits (e.g., enhance transparency, improve accountability, strengthen oversight, democratic oversight, public interest, user rights). This pattern answers SRQ1 and SRQ2a–b: enforcement is publicly framed as a trade-off; stakeholders communicate via familiar policy, reporting, and explanatory genres; and the same themes are mobilized to different ends. Crucially, these dynamics illuminate the RQ: while TikTok speaks the DSA’s language, the emphasis on producing documents and statements can shade into transparency-washing: the appearance of openness through volume and repetition rather than disclosures that are comparable, verifiable, and useful for oversight. In other words, alignment in words does not guarantee alignment in practice.

SRQ3a-b extend the point from rhetoric to user experience. Transparency information is accessible, but it is distributed across EU/EC pages, TikTok ’s own materials, and specialized coverage; users typically reach it by navigating reports and policy statements rather than a single, user-facing path. This dispersion, coupled with argumentative framing, helps explain why the comprehensiveness and usability of TikTok ’s transparency reporting remain debated: stakeholders can find material, but not always in a way that resolves key questions about decision rationales or comparative scope. That is precisely where transparency-washing gains traction—when disclosures exist but do not reduce uncertainty or enable like-for-like reading across sources. Implications follow for society, academia, and technical practice: publish plain-language summaries that tie each document to a specific user question; maintain consistent labels and categories across stakeholders; and interlink EU, platform, and media documents so claims can be traced across the ecosystem (addressing RQ and SRQ3a-b).

Looking ahead, future work should expand the time window, add stance/claim annotations to distinguish description from advocacy, and pair textual mapping with user-journey tests of where people actually succeed, or fail, in finding answers. Another further research direction would be to compile more granular sets of stakeholders and compare their discourse around particular DSA-enforcement measures. Alternatively, a longitudinal analysis of EU Transparency reports versus TikTok Transparency reports, against TikTok API data would help triangulate content moderation decision-making navigation. These attempts might shed more light into how the phenomenon of transparency washing prevents meaningful analysis from researchers.

TikTok is the stress-test. Whether the platform remains the decision-maker or the Commission graduates to genuine co-governance will signal the DSA’s trajectory, and, by extension, the resilience of democratic oversight in the attention economy.

8. References

n.d. Aggregated data. DSA Transparency database tool. https://dsa.pages.code.europa.eu/transparency-database/dsa-tdb/data_sources.html#aggregated-data

n.d. DSA Transparency Database: Download. European Commission. https://transparency.dsa.ec.europa.eu/explore-data/download

n.d. Digital Services Act: Questions and answers. European Commission.

https://digital-strategy.ec.europa.eu/en/faqs/digital-services-act-questions-and-answers

n.d. The Digital Service Act. European Commission. https://commission.europa.eu/strategy-and-policy/priorities-2019-2024/europe-fit-digital-age/digital-services-act_en

n.d. The final text of the Digital Services Act (DSA). European Union. https://www.eu-digital-services-act.com/Digital_Services_Act_Article_17.html?utm_source=chatgpt.com

n.d. Transparency center. TikTok. https://www.tiktok.com/transparency/en

(2025, May 14). Commission preliminarily finds TikTok 's ad repository in breach of the Digital Services Act. European Commission. https://ec.europa.eu/commission/presscorner/detail/en/ip_25_1223

(2025, June 26). Community guidelines enforcement report. TikTok Transparency Center. https://www.tiktok.com/transparency/en-us/community-guidelines-enforcement-2025-1

(2024 April 17). Community guidelines: Overview. TikTok. https://www.tiktok.com/community-guidelines/en

(2025 February 28). Digital Services Act: Our fourth transparency report on content moderation in Europe. TikTok. https://newsroom.tiktok.com/en-eu/digital-services-act-our-fourth-transparency-report-on-content-moderation-in-europe

(2025 June 26) EU report finds tech giants failing to meet disinformation commitments. EU Insider. https://www.euinsider.eu/news/eu-report-finds-tech-giants-failing-to-meet-disinformation-commitments

(2025 June 24). Implementing the EU code of practice on disinformation: An evaluation of VLOPSE compliance and effectiveness (Jan–Jun 2024). European Digital Media Observatory. https://edmo.eu/publications/implementing-the-eu-code-of-practice-on-disinformation-an-evaluation-of-vlopse-compliance-and-effectiveness-jan-jun-2024/

(2025 May 15). Violation of DSA? EU Commission takes further action against TikTok. Ailance. https://2b-advice.com/en/2025/05/15/violation-of-dsa-eu-commission-takes-action-against-tiktok/

Allen, A. (2022 July 8). Europe's big tech law is approved. Now comes the hard part. Wired. https://www.wired.com/story/digital-services-act-regulation/

Bassan, S. (2025). Transparency ≠ accountability? Rethinking voluntary vs. mandatory content moderation reports. SSRN. http://dx.doi.org/10.2139/ssrn.5143075

Blenkinsop, P. (2024 December 17). EU opens investigation into TikTok over election interference. https://www.reuters.com/business/eu-opens-investigation-into-tiktok-over-election-interference-2024-12-17/

Chee, F.Y. (2025 May 15). TikTok charged with breaching EU online content rules. Reuters. https://www.reuters.com/sustainability/boards-policy-regulation/tiktok-charged-with-breaching-eu-online-content-rules-2025-05-15/

Drolsbach, C. P., & Pröllochs, N. (2024). Content moderation on social media in the EU: Insights from the DSA transparency database. Companion Proceedings of the ACM Web Conference 2024, 939–942. https://doi.org/10.1145/3589335.3651482

Entrena-Serrano, C., Degeling, M., Romano, S., & Çetin, R.B. (2025). TikTok 's research API: Problems without explanations. AI Forensics. https://www.arxiv.org/abs/2506.09746

Fabbri, M. (2025). The role of requests for information in governing digital platforms under the digital services act: The case of X. Journalism and Media, 6(1). https://doi.org/10.3390/journalmedia6010041

Heath, A. (2023 February 2). TikTok ’s transparency theater. The Verge. https://www.theverge.com/2023/2/2/23583491/tiktok-transparency-center-tour-photos-bytedance

Kaushal, R., Van De Kerkhof, J., Goanta, C., Spanakis, G., & Iamnitchi, A. (2024). Automated transparency: A legal and empirical analysis of the digital services act transparency database. Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency, 1121–1132. https://doi.org/10.1145/3630106.3658960

Nahrgang, M., Weidmann, N. B., Quint, F., Nagel, S., Theocharis, Y., & Roberts, M. E. (2025). Written for lawyers or users? Mapping the complexity of community guidelines. Proceedings of the International AAAI Conference on Web and Social Media, 19(1), 1295-1314. https://doi.org/10.1609/icwsm.v19i1.35873

Naja, B. (2024 November 12). Mis- and disinformation on social media and related risks to election integrity. European Parliament. https://www.europarl.europa.eu/thinktank/en/document/EPRS_ATA%282024%29767150

Trujillo, A., Fagni, T., & Cresci, S. (2025). The DSA Transparency Database: Auditing self-reported moderation actions by social media. Proceedings of the ACM on Human-Computer Interaction, 9(2), Article CSCW187. https://doi.org/10.1145/3711085

Urman, A., & Makhortykh, M. (2023). How transparent are transparency reports? Comparative analysis of transparency reporting across online platforms. Telecommunications Policy, 47(3). https://doi.org/10.1016/j.telpol.2022.102477

Zalnieriute, M. (2021). “Transparency washing” in the digital age: A corporate agenda of procedural fetishism. Critical Analysis of Law : An International & Interdisciplinary Law Review, 8(1), 139–153. https://doi.org/10.33137/cal.v8i1.36284

9. Appendix

Final Datasets