PaperlayerView exact revision on arXiv

Source-linked AI summary

Senseful Consense: Towards Simplified Cookie Banners using Plain Language

Minela Bećirović, Ha Dao, Mannat Kaur, Martin Johns, Alexandra Dirksen

arXiv:2609.10271v1cs.CYcs.HC

TL;DR

Cookie banners often use technical or vague language, leaving users with content they may not understand. This feasibility study applies plain language within the IAB TCF; it reports that comprehension barriers fell from a college-graduate level to a 7th-grade level.

  • Problem

    Cookie-banner language can be vague and technical, and prior work had not established how plain language could improve comprehension within the IAB TCF.

    Source: §1 · p. 2

  • Method

    The authors developed plain-language descriptions for TCF purposes, used AI-based mapping to classify extracted banner text, and generated modified texts for comparison.

    Source: §1 · p. 2

  • Results

    Comprehension barriers fell from a college-graduate level to a 7th-grade level in the study of 181 websites.

    Source: §7 · p. 10

  • Takeaways & Limitations

    Plain language is a tool for digital accessibility, but cookie banners also need standardised implementation guidelines to support informed decisions.

    Source: §7 · p. 10

    Website implementation and wording diversity limited text extraction and consistent assignment of banner text to purposes.

    Source: §6.3 · p. 10

Abstract

from arXiv · show

While the GDPR and ePrivacy Directive mandate that consent information must be clear and accessible, most modern cookie banners remain obscured by technical jargon, vague phrasing, and frequent content overload or underload. This feasibility study investigates the impact of applying plain language (Einfache Sprache) to cookie banners within the IAB Transparency & Consent Framework (TCF). In our study, we analysed cookie banner texts from 200 websites, using AI-based mapping to categorise extracted content into standardised processing purposes. By substituting complex legal terms with simplified descriptions, we successfully demonstrated that the comprehension barrier can be lowered from a college-graduate level to a 7th-grade level. However, the effectiveness of plain language is inherently constrained by the informativeness of the original content; it cannot compensate for banners that omit legally required details. We conclude that while plain language is a vital tool for digital accessibility, it must be paired with standardised implementation guidelines to ensure that cookie banners are both readable and informative.

1 Introduction

Cookie banners must support informed consent through clear, accessible information, yet technical language and vague phrasing often make their content difficult to understand. This study asks whether plain language can improve cookie-banner comprehensibility within the IAB TCF.

1 Introduction

  • GDPR and ePrivacy rules call for consent information to be clear and accessible, but cookie banners vary in design and in the consent options they offer.Some banners merely inform users, while others offer granular choices.Source: §1 · p. 1
  • Technical terms, vague wording, and complex language make cookie banners hard to understand and can function as textual dark patterns.The banners are often users’ first line of defence for informed online privacy decisions.Source: §1 · pp. 1–2
  • Browser extensions can apply users’ privacy preferences automatically, but they do not make banner content more understandable.The paper argues that banners should help everyone make informed privacy decisions rather than be treated only as a nuisance.Source: §1 · p. 2
  • The feasibility study investigates whether plain language can improve cookie-banner comprehensibility while accounting for the IAB TCF.Its question concerns current cookie-banner practices.Source: §1 · p. 2
  • The authors developed plain-language descriptions for TCF purposes, mapped banner text to those purposes, and compared original with modified texts.The comparison also indicated that simplification cannot compensate for missing or unclear information.Source: §1 · p. 2

2 Background & Related Work

This background section connects consent law and the TCF with concerns about comprehension, dark patterns, and user interactions with cookie banners.

2.1 The Legal Basis for Processing Personal Data

  • GDPR consent must be freely given, specific, informed, and unambiguous, and obtained in an intelligible, accessible form using clear and plain language.Processing includes operations such as collection, storage, and disclosure.Source: §2.1 · p. 2

2.2 Cookie Banners and the IAB TCF

  • The TCF provides eleven standardised processing purposes and lets users consent or object on legal bases including consent and legitimate interest.Despite standardisation, its descriptions retain technical complexity that can hinder comprehension.Source: §2.2 · pp. 2–3

2.3 The Crux with Cookie Banners

  • A gap remains between legal transparency requirements and users’ actual comprehension of cookie banners.Technical standardisation alone does not close this gap.Source: §2.3 · p. 3, §2.2 · p. 3

2.3.1 Dark Patterns in Banner Designs.

  • Cookie-banner dark patterns include highlighted accept buttons, hidden options, and preselected choices that influence users’ consent decisions.One study found that 86% of German websites offering rejection used visual nudging to encourage acceptance.Source: §2.3.1 · p. 3

2.3.2 Interaction of Users with Cookie Banners.

  • Studies report substantial banner avoidance and annoyance, alongside varied acceptance, rejection, and customization behaviours.One behavioural study found that 68% ignored or closed the banner; its results warrant caution because they lacked peer review.Source: §2.3.2 · p. 3

2.4 The Concept of Plain Language

The section distinguishes plain-language approaches and describes the paper’s operational definition, scope, and tools.

2.4 The Concept of Plain Language

  • Plain-language efforts emphasize communication that readers can understand, while simplification approaches differ in target audience and linguistic implementation.The section distinguishes Germany’s Leichte Sprache from Einfache Sprache.Source: §2.4 · p. 3, §2.4.1 · p. 3

2.4.1 Defining the Model: Einfache Sprache (Germany).

  • The study uses Einfache Sprache for a broad audience, allowing technical terms while simplifying syntax and grammar and targeting B1-B2 language levels.The paper ties this operational definition to DIN ISO 24495-1 and DIN 8581-1.Source: §2.4.1 · p. 3

2.4.2 The Scope of Plain Language.

  • The authors use capito.ai and Wortliga tools to rephrase text in plain language.Wortliga is described as recommended by Germany’s certification authorities.Source: §2.4.3 · p. 4

2.4.3 Tools for Plain Language.

  • The tools offer AI-based rewriting and analysis, with capito.ai providing three language levels and Wortliga supporting rewriting in over 50 languages.The passages also describe their text-handling and data-protection practices.Source: §2.4.3 · p. 4

2.5 Assessment of Textual Comprehensibility

The section describes readability measures and identifies plain-language implementation in cookie banners as an underexplored research area.

2.5 Assessment of Textual Comprehensibility

  • Textstat computes readability measures including Flesch Reading Ease, Gunning FOG, and New Dale-Chall scores, whose directions indicate easier reading or lower required grade levels.Higher Flesch scores and lower FOG or Dale-Chall scores indicate improved readability.Source: §2.5 · p. 4
  • The authors also consider capito.ai’s B1-B2 percentage and Wortliga’s comprehensibility and readability scores.Wortliga considers scores from 60 to 100 comprehensible, and its Rewriter’s 100% indicates best possible readability.Source: §2.5 · p. 4

2.6 Research Gaps

  • Prior cookie-banner research addressed legal validity, design dark patterns, and complex-language detection, but plain-language implementation remained underexplored.This study addresses that gap in the context of the IAB TCF.Source: §2.6 · p. 4

3 Comprehensibility of Cookie Banners

This section evaluates whether plain language can improve comprehensibility within the IAB TCF, assessing the framework’s existing descriptions before comparing revised descriptions and their application to cookie banners.

3.1 Enhancing TCF’s Efforts

  • TCF’s existing “user-friendly descriptions” received poor comprehensibility scores; revised versions were iterated toward B1–B2 and stable readability scores.The authors used Wortliga Rewriter and, in some cases, capito.ai or manual adjustments to refine the descriptions.Source: §3.1 · p. 4

3.2 Enhancement Comparison

  • Comprehensibility was compared using Textstat, capito.ai, and Wortliga.The comparison assessed the user-friendly descriptions against the revised plain-language descriptions.Source: §3.2 · p. 5
  • Flesch Reading Ease for P9 improved from -17.18 for the user-friendly description to 50.02 for its plain-language version; the FOG Index for P3 fell from 15.81 to 7.25.The FOG change corresponds to a shift from college-junior to seventh-grade readability.Source: §3.2 · p. 5
  • The Dale-Chall Readability Score for P5 fell from 9.58 for the user-friendly description to 7.35 for its plain-language version.The paper associates these scores with college-level and 9th- or 10th-grade readability, respectively.Source: §3.2 · p. 5
  • Capito.ai’s B1–B2 percentage for P5 rose from 0% to 97%, while Wortliga levels shifted from C2 to B1–B2 and scores from 0–32 to 85–96.The authors therefore consider the plain-language descriptions comprehensible and the user-friendly descriptions not comprehensible by Wortliga’s criteria.Source: §3.2 · p. 5

3.3 Application of Plain Language Descriptions

  • The proposed cookie-banner concept combines IAB TCF purposes with plain-language principles to improve comprehensibility.The section frames this as a feasibility concept for applying purpose descriptions to cookie banners.Source: §3.3 · p. 5

4 Methodology

The methodology samples English-language cookie banners, maps their content to IAB TCF purposes, creates modified plain-language versions, and assesses comprehensibility. A pre-study selected o3 for the main mapping, while the analysis also revealed vague wording and purpose assignments based on implication.

4.1 Study Design

  • The study processes first-level banner text through extraction, IAB TCF purpose mapping, plain-language modification, and comprehensibility assessment.Source: §4.1 · p. 5, Figure 3 · p. 5

4.2.1 Selection of the Website Sample.

  • The sample comprises 200 English-language banners selected by manually reviewing the Majestic Million list, excluding duplicate site versions and nonqualifying websites.Banners were considered only when natively available in English; no translation tools were used.Source: §4.2.1 · pp. 5–6

4.2.2 Cookie Banner Text Extraction.

  • A Firefox extension required manual banner selection and extracted text from 181 of 200 sites; extraction failed on 19, often with Shadow DOM or iFrames.Only the first banner level was analyzed, and partially extracted text was retained as what users see before interaction.Source: §4.2.2 · pp. 6–7

4.3 IAB TCF Purpose Mapping

  • In the pre-study, AI-based methods aligned more closely with manual mappings than statistical methods; OpenAI’s o3 was most accurate and was applied to the main study’s reduced sample of 181 websites.Source: §4.3 · p. 7
  • P1 and P10 were assigned most often, P2 and P11 least; paired P3/P4 and P5/P6 assignments predominated, though single-purpose exceptions occurred.P7 mapped to advertising performance, P8 to content performance or traffic analysis, and P9 to audience insights, preferences, or partner sharing.Source: §4.3 · p. 8
  • Banners used vague and varied wording; purpose assignments often relied on implications, including inferred profiles for personalized advertising or content.Withdrawal or subscription text generally received no purpose assignment, while banners explicitly listing the TCF purposes sometimes received all 11.Source: §4.3 · p. 8

4.4 Modification of Cookie Banner Text

  • A Python script replaced each extracted banner text with the plain-language description corresponding to every mapped purpose.Source: §4.4 · p. 8

5 Evaluation

Modified banner texts generally scored better on readability and comprehensibility measures than originals. Improvements appeared across several tools, though some cases showed no improvement or lower scores.

5 Evaluation

  • 167 of 181 modified banners had higher capito.ai B1–B2 percentages; 158 of 181 had improved Wortliga comprehensibility scores.Source: Table 5 · p. 9, §5 · p. 9
  • The Flesch Reading Ease Score rose from 8.74 to 57.98 in one case, while the FOG Index fell from 17.33 to 8.1 in another.Only 13 of 181 original banners scored higher on Flesch Reading Ease, and only 4 of 181 had lower original FOG scores.Source: Table 5 · p. 9, §5 · p. 9
  • Wortliga assigned 142 of 181 modified banners a lower language level, while 23 stayed the same and 16 received a higher level.Modified versions’ Wortliga comprehensibility scores ranged from 75 to 89; 22 cases had lower scores after modification.Source: Table 5 · p. 9, §5 · p. 9

6 Discussion

The discussion reports improved readability from plain language while emphasizing that comprehension scores do not establish whether banners provide enough information for informed decisions. It also identifies mapping, implementation, standardization, measurement, and data-collection limitations.

6.1 IAB TCF Purpose Mapping

  • Vague wording and varied phrasing led the AI model to make assumptions, creating inconsistent purpose assignments and possible omissions or incorrect assignments from incomplete texts.The authors note these issues can hinder both automated and human understanding.Source: §6.1 · p. 9

6.2 Application of Plain Language

  • B1–B2 descriptions and modified banners were generally more readable, but high comprehensibility scores do not establish that content informs decisions about personal-data processing.One banner scored A1 and 100 for comprehensibility despite providing no specific information about processing.Source: §6.2 · pp. 9–10

6.3 Limitations

  • Implementation variety limited both text extraction and purpose assignment, constraining analysis of the cookie-banner landscape.Observed variations included dynamic structures, accessibility-related attributes, iFrames, and Shadow DOM.Source: §6.3 · p. 10
  • Manual mapping was subjective and potentially erroneous, and because it informed mapping-method selection, that subjectivity qualifies interpretation of the findings.Source: §6.3 · p. 10
  • Plain language lacks fixed rules and consistent language-level definitions, while its practical use on cookie banners also needs legal consideration.The study follows capito.ai’s B1–B2 definition, while other sources associate plain language with A2–B1.Source: §6.3 · p. 10
  • Comprehensibility scores do not show whether users can make informed decisions; the authors recommend a user study comparing interactions with original and modified banners.Source: §6.3 · p. 10

6.4 Ethical Consideration

  • Data collection covered first-level banner text on public websites and excluded sites where extraction was not readily feasible or required bypassing technical protections.The study did not collect or process personal data.Source: §6.4 · p. 10

7 Conclusion

The study examined plain language for cookie banners within the IAB TCF and reports a reduction in comprehension barriers from college-graduate to 7th-grade level. It concludes that plain language needs standardised implementation guidelines and cannot compensate for missing information.

  • Comprehension barriers fell from college-graduate level to 7th-grade level, while plain language could not compensate for banners lacking necessary information.Source: §7 · p. 10
  • The study examined the feasibility of using plain language (Einfache Sprache) to improve cookie-banner comprehensibility within the IAB TCF.Source: §7 · p. 10
  • The authors conclude that standardised cookie-banner guidelines are required alongside plain language to prevent complex or vague wording from functioning as a dark pattern.Source: §7 · p. 10

A USER-FRIENDLY descriptions for the purposes defined by the IAB TCF

Table 6 presents the IAB TCF’s 11 standard purposes and their user-friendly explanations.

B Language levels as defined by capito.ai

Figure 5 describes capito.ai’s three language levels, from very easy to simple language. The German page uses Leichte Sprache for “easy language” and Einfache Sprache for “simple language,” matching the paper’s plain-language definition.

  • Figure 5 defines A1–A2 as very easy, A2–B1 as easy, and B1–B2 as simple language, with examples, descriptions, and typical use cases for each.The levels differ in complexity, target audience, and estimated comprehension rates.Source: B Language levels as defined by… · p. 14
  • In the German page version, Leichte Sprache means “easy language” and Einfache Sprache means “simple language,” corresponding to the paper’s definition of plain language.Source: Figure 5 · p. 14

C Details on Statistical Methods

The methods combine semantic similarity with keyword matching to detect processing purposes in cookie-banner text.

C Details on Statistical Methods

Loading 2609.10271v1…