Methodology
The 2026 pilot study systematically examines how accessible important Swiss websites really are in everyday use. It combines manual and automated testing to provide an overall picture. The study is carried out under the auspices of the Swiss Digital Inclusion Alliance (ADIS). The «Access for all» foundation leads the study and conducts the tests, with academic support from Bern University of Applied Sciences (BFH).
The pilot study is the starting point for ADIS's regular monitoring programme, Digital Inclusion Monitoring Switzerland (DIMoS).
Websites examined
The study examines publicly accessible websites of particular relevance to the population. Selection follows three rules:
- Continuity: Wherever possible, the same websites are tested again as in 2016.
- Relevance: Websites that have become more important since 2016 are added, such as key e-government services.
- Coverage: The selection broadly covers public administration, higher education, media and online retail.
The test plan calls for at least four sample pages to be examined in detail per website: the home page, a form page and two pages from a typical user scenario. Representative PDF documents are also included. Each test records at the outset which pages and document versions it examines. Later changes to the website do not change the result.
Testing framework
The benchmark is version 4 of the Swiss eCH-0059 standard. It incorporates WCAG 2.2 requirements at level AA and adds Swiss requirements, including easy-to-read language, sign language and accessibility statements. The study uses separate checklists:
- eCH-0059 – Web: WCAG 2.2 requirements for web pages;
- eCH-0059 – Additional web requirements: additional requirements of the standard, including those for accessibility statements;
- eCH-0059 – Documents: requirements for PDF documents, supplemented by automated testing against PDF/UA-1 and document-related WCAG requirements;
- eCH-0059 – Easy-to-read language: tested systematically for the first time in a Swiss study and reported separately.
The checklists remain frozen throughout data collection. Results from different checklists are not combined into a single indicator.
Test specifications: making the testing itself traceable
An innovation of this study, new to this field, is the use of test specifications. A standard such as eCH-0059 describes what must be fulfilled. A test specification also defines how to test: which test cases belong to each criterion, which rating scale applies, which results count as fulfilled, and how results are aggregated into indicators.
- Versioned and frozen: Each test specification has a version number. Each test run records the version used. Later changes to a checklist do not alter earlier results.
- One specification per testing method: Manual web testing, automated web testing, document testing, easy-to-read language and accessibility statements each have their own test specification. All refer to the same criteria in the standard. The methods can therefore be compared by criterion without mixing their results.
- Reusable: Test specifications are published as packages. Other studies, administrations or companies can test using exactly the same method, making their results directly comparable.
- Transparent: Anyone wishing to trace a result can follow the path from criterion to test case, test run and individual finding.
The method itself thus becomes a verifiable, reusable part of the study. This is a prerequisite for monitoring that provides comparable statements over many years.
Two testing methods: manual and automated
Manual testing. Certified accessibility specialists test the defined sample of 25 of the 102 websites against the eCH-0059 – Web checklist, following real user scenarios and a second-person review process. The team includes specialists with disabilities, particularly for screen reader, keyboard and magnification testing. Each finding receives a rating from 1 to 10, a short explanation and concrete examples identifying the page and describing the issue.
Automated testing. Established open-source testing software checks the sample pages against those WCAG requirements that can be detected automatically with a high degree of confidence. Each result is stored with its rule, software version and timestamp. Results requiring human judgement are marked «Review required» and are not counted as violations. PDF documents are automatically tested against PDF/UA-1 and document-related WCAG requirements.
Comparing the two methods
Results from both methods are compared by website and WCAG criterion. The comparison shows where both methods report a violation, where only manual testing identifies one, and where no automated rule exists. The tested pages and scope remain traceable for each run. Both views use the manually selected websites, including those without automated results. Analyses requiring evidence from both methods state their own denominator. What is counted is criterion-level signals, not matching individual errors. The absence of an automated signal does not prove that the tool missed the very same individual error. This makes the complementary contributions of the two methods visible.
Easy-to-read language
The 2026 pilot study is the first Swiss accessibility study to include easy-to-read language.
Easy-to-read language supports people with limited reading skills. It simplifies texts and makes them easy to understand. It aims for clear statements and leaves out unnecessary information. Difficult words are explained and abstract concepts illustrated with examples. Particular attention is paid to content structure and layout, which are equally important for comprehension. Easy-to-read language corresponds to levels A1 to A2 of the Common European Framework of Reference for Languages. It is not the same as plain language, which is less simplified and is more closely associated with level B1.
People with cognitive disabilities generally need easy-to-read language; information in plain language excludes them. Conversely, the target audience for plain language also understands texts at A2 level. This is why the study focuses on easy-to-read language.
People with limited reading skills are often excluded from digital services because they cannot understand their content. Including easy-to-read language gives a more complete picture of the groups affected by digital exclusion and of digital inclusion in society.
Unlike the largely quantitative measures used in traditional accessibility testing, the assessment of easy-to-read language relies mainly on qualitative criteria. The study follows two steps:
- Entry point on the home page: Every website is checked for a recognisable entry point to easy-to-read content on its home page.
- Detailed assessment: For websites in the manual testing groups, a specialist assesses three criteria. If no easy-to-read content is found, findability is considered unfulfilled; language level and recognisability are then not assessed:
- Findability: The link to easy-to-read information is clearly visible on the home page.
- Recognisability: Easy-to-read information is marked with the easy-to-read icon.
- Language level: Texts do not exceed language level A2.
These criteria show whether existing easy-to-read information can be found, recognised and understood by people with very limited reading skills. The results also support the further development of testing criteria for future measurements.
Accessibility statements: first assessment against the new criteria catalogue
An accessibility statement explains how accessible a website is, which limitations are known and how people can report barriers. Version 4 of eCH-0059 introduces a binding criteria catalogue for this purpose. The pilot study is the first to assess statements against this catalogue:
| Criterion | Requirement |
|---|---|
| 16.1 | Compliance status and scope are stated. |
| 16.2 | Inaccessible content and exemptions are documented. |
| 16.3 | A feedback mechanism and escalation contact are provided. |
| 16.4 | The statement is available in German, French and Italian. |
| 16.5 | The statement is available in Romansh where required. |
| 16.6 | The statement is available in English where required. |
| 16.7 | A machine-readable version is available. |
| 16.8 | The statement can be reached from every page. |
The main verdict on accessibility statements comes from 28 completed manual tests: 23 are recorded as non-conformant and 5 as conformant. The report retains the verdict stored under the statement test’s rules, without applying a new threshold or the Web G classes. All 46 rated observations in these tests concern statement criteria; other supplementary requirements are marked not applicable. Recorded ratings cover 16.1, 16.3, 16.4 and 16.8; missing ratings on other criteria are not treated as fulfilment.
A supplementary automated check covers 101 websites, recording whether a statement and its information can be detected. Each criterion is recorded as found, unclear, not found or not checkable. This describes detection, not the manual assessment of the content. Both methods’ results are presented with their respective scope.
Comparison with the 2016 study
Common rating scale
The historical overall rating of 1–5 stars is multiplied by 2 and compared with the mean observation rating for 2026 on the 1–10 scale. G remains a separate measure. We report the median of paired changes, not the difference between two group medians.
Assessment requirements over time
The current Web assessment uses WCAG 2.2 and eCH-0059, version 4. The historical comparison uses the ratings recorded in each study. A pair refers to the same website in both years. Requirements, assessed content and testing conditions are not necessarily identical. The common numerical scale allows the ratings to be shown together.
How to read the manual document rating
The document Quick Test comprises 25 assessments of 20 distinct documents. Several websites use the same document; identical content counts once in the document-level analysis. Ratings of the same document agree.
The Quick Test is a sample assessment at the tester’s discretion. The stored rating is the mean of numerical observation ratings on a 1–10 scale. Under the specification used, 6 points or more, including the boundary, receives the label “conforming”: 4 of 20 documents meet it. This classification does not certify full conformance. The manual Web assessment’s G classes do not apply to PDFs.
98 selected test runs cover 93 distinct documents. All 98 validations completed; no selected execution failed or remained unvalidated. Of the 102 websites, 4 have no selected automated PDF test run; they do not count as passes.
91 of 93 documents have at least one automatically detected rule violation. 2 have no detected violation; this does not certify full PDF/UA conformance. The median is 7 failed rules per document, ranging from 0 to 17.
Testing used PDF/UA-1, with veraPDF 1.30.2 (96 runs) and 1.30.3 (2 runs). The results therefore come from two software versions.
Analysis and evaluation metrics
The analysis answers three different questions: How accessible is a website overall? How many websites are insufficiently accessible? And on how many was at least one deficiency observed? A score, a proportion of websites and the frequency of deficiencies are different indicators.
K, S and G: three complementary perspectives
| Score | What it describes | Role in the assessment |
|---|---|---|
| K – WCAG criterion score | Manually assessed criteria, averaged within user aspects and then combined with equal weight per aspect. | Shows the criterion assessment independently of user scenarios. |
| S – scenario score | The lowest result across all recorded user scenarios in the included manual runs. | Reveals difficulties in completing connected user tasks. |
| G – overall score | 0.5 × K + 0.5 × min(K, S). | Combines criteria and scenarios to determine the website’s overall class. |
All three scores use 0 to 100 points. K is neither a simple average of all findings nor the percentage of WCAG criteria passed. User aspects receive equal weight even when they contain different numbers of criteria. If S is at least as high as K, G remains equal to K; a lower S reduces G. Scenarios therefore cannot improve a weak criterion score.
Because user aspects receive equal weight, an applicable criterion in an aspect containing few criteria has more weight than one in an aspect containing many. Non-applicable criteria and entirely non-applicable aspects are excluded, so the effective weights can differ between websites.
The analysis shows K, S and G for each website and adds the table “Criteria score K — without scenarios”. This supplementary view assigns the existing, unrounded K scores to the same four point ranges as G: good from 90, conditional from 80, insufficient from 60 and poor below 60. It changes neither K, S nor G and does not replace the overall assessment. Its classes describe the criterion assessment within the chosen study scope, not formal WCAG conformance.
In the K distribution, each website with a fully assessable K counts once. Counts and percentages refer only to this group; websites without an assessable K are reported separately. A partial score from incomplete criterion evidence is not assigned a class. Missing S, however, does not prevent a K-only classification. The assessable groups for K and G may therefore differ in size. The following sections explain the calculation in detail.
Overall score and accessibility classes
G describes recorded accessibility on a scale of 0–100 points. It is not a percentage of fulfilled criteria. This edition uses four classes: good accessibility from 90 points, conditional accessibility from 80, insufficient accessibility from 60, and poor accessibility below 60. The last two count as insufficiently accessible. The class table below shows the effective thresholds. The median G is the middle score of assessable websites; the profiles on “The study” page describe a different measure.
The classes summarise the requirements and usage scenarios tested:
- Good accessibility: The requirements and usage steps tested are largely fulfilled.
- Conditional accessibility: Remaining barriers restrict use.
- Insufficient accessibility: Substantial barriers make use difficult.
- Poor accessibility: Pronounced barriers severely restrict use.
Individual barriers may exist in every class.
Selection and criterion scope
The overall rating includes all completed manual web test runs selected for domain compliance. Each website counts once. Multiple selected runs for one website are evaluated together; the version number of a test specification does not determine inclusion. The original specification and its rating scheme remain as provenance. Automated testing, PDF documents, easy-to-read language and accessibility statements are reported separately.
The common criterion scope is defined in the study configuration. Criterion 4.1.1 «Parsing» is explicitly excluded from this analysis. The remaining criteria remain visible even when evidence is missing. «Not applicable» is an explicit assessment and is different from a missing result.
K: criterion score
The applicable test specification determines which observation ratings indicate a deficiency. A sufficiently evidenced fulfilled criterion receives 10 points on the observation scale. For a sufficiently evidenced unfulfilled criterion, the lowest associated deficiency rating across all included runs applies. Each criterion counts only once per website; additional observations do not increase its weight. Without sufficient evidence, it is considered unknown or partially assessed.
Within each configured usability aspect, the mean of its applicable criteria is calculated. K is the mean of these aspect scores, converted from a 0–10 scale to 0–100 points. Entirely inapplicable aspects are omitted. Missing evidence is not counted as fulfilled: known partial values may be shown, but do not produce an overall class.
S: tested user scenarios
We include all recorded user scenarios from the included manual test runs. The lowest scenario result is combined with the criterion score. Scenarios can lower the score but cannot increase it.
Default scenario mapping in this study:
| Scenario result | Fraction | S in points |
|---|---|---|
| Accomplished | 6/6 | 100 |
| Partially accomplished | 4/6 | 66⅔ |
| Not accomplished | 1/6 | 16⅔ |
The mapping uses stored result keys and their configuration identity, not words from free text. With multiple scenarios, the minimum applies. S is unavailable if a recorded scenario has no result, cannot be mapped, or no scenario is recorded. A test run without its own scenario remains included in the criterion analysis.
G: overall score and classes
G = 0.5 × K + 0.5 × min(K, S)
A successful scenario cannot improve a weaker criterion score. Conversely, a weaker scenario lowers the overall score. A single deficiency does not trigger an additional blanket penalty. The score describes the tested criteria and recorded user scenarios; it is neither a declaration of conformity nor a guarantee of usability for every person.
Worked example: With K = 80 points and a partially accomplished scenario, S = 66⅔ and G = 73⅓ points. This is a calculation example, not a study result. The corresponding class follows from the class table shown below for this edition.
Class thresholds are study configuration. The class with the highest lower bound reached by the unrounded G score applies. The table at the end of this page is generated from the same stored rule as the results; it is the authoritative mapping for this edition. Websites that cannot be evaluated belong to no class.
Proportions and frequencies
The proportion of insufficiently accessible websites counts websites in the classes configured for this purpose and divides that count by the number of evaluable websites. The display always states the numerator and denominator, as well as the additional websites that cannot be evaluated. The class distribution and median G complement this proportion. A score of 80 does not mean that 80 per cent of websites are accessible.
Deficiency frequency answers a separate question: On how many websites was at least one deficiency observed within the examined criterion scope? A known deficiency remains an observed deficiency even when other evidence is missing. «No deficiency observed» requires complete evaluable evidence; otherwise this part of the analysis remains unevaluable. Its denominator is stated separately.
Scenario frequencies show the number of scenarios and the number of affected websites separately. Category comparisons use the same individual evaluations and state the group size. A website is not counted more than once within a category.
Average observation rating and comparison over time
The existing average observation rating remains a separate indicator. It summarises stored observation ratings and is neither the new G score nor a proportion of accessible websites. The comparison with 2016 continues to use its previous calculation; no new accessibility classes are assigned to it. Minima, medians, maxima and counts refer to the stated testing method and selected group.
Quality assurance
- Training and calibration of all testers before testing starts
- Second-person review of all manual assessments
- Independent double assessment of selected websites
- Plausibility checks between automated and manual testing
- Academic support from Bern University of Applied Sciences (BFH) and a project advisory board
Use of the results
The study takes a constructive approach: it raises awareness of digital barriers and identifies concrete opportunities for improvement. Results are published at sector and group level, without ranking individual organisations against one another. Each tested organisation receives a confidential report documenting the barriers found and providing specific starting points for improvement.
Limits of interpretation
- A sample of pages is tested for each website, rather than the entire website.
- Automated testing covers only some requirements.
- The criteria for easy-to-read language and accessibility statements are based on version 4 of eCH-0059 and are applied for the first time in this study.
- Websites change continually; results apply to the documented data collection period.
Scope, software and limitations
The manual Web assessment covers 25 of the 102 websites in this evaluation. It examines the defined sample and recorded usage scenarios; “fully assessed” does not mean every page of a website. The planned minimum of four sample pages is an assessment plan, not a statement of how many pages were actually tested.
Automated Web assessments use axe-core, and PDF tests use veraPDF. A test-specification version identifies the checklist, not the executing software. The PDF tests used veraPDF 1.30.2 (96 selected runs) and 1.30.3 (2 runs). Identical document content counts once in the document-level summary, even when tested on several websites. The sector analysis instead counts websites. Having no automatically detected violation does not imply full PDF/UA conformance.
Partially executed automated tests provide results for targets actually assessed. Failed or skipped targets have not passed. The specification’s selection policy may allow this partial evidence; it changes neither the original observations nor the scoring formulas.
Evidence behind the automated Web testing scope
The automated testing scope and reasons for missing results come from the operational log of 16 November 2026, 07:14–07:44 UTC. This log was extended by one additional completed test run of 17 November, with four of four pages tested. The 94 websites with results and eight without were reconciled with this edition’s selected run identities. All planned pages were tested for 66 websites, and some planned pages for 28. Reasons for missing results appear in the overview on “The study” page.
For the criterion comparison, results are assigned to common criteria using defined mappings. This comparison contains 93 automatically tested websites. One further website has an automated overall result but is not included in that criterion mapping. Direct manual/automated rankings use the 25 manually selected websites; missing automated evidence remains visible.
Means in the accessibility profile
The profile bars and radar charts show the arithmetic mean of the percentages for individual websites. Every website has equal weight. For each website, the percentage is the proportion of fulfilled criteria among the sufficiently assessed, applicable criteria for that aspect. Inapplicable criteria are excluded from the denominator; websites without a numeric value do not enter the mean as zero. The display states the number of included websites (n).
These profiles describe recorded criterion results. They are neither the G score nor the proportion of sufficiently accessible websites. Selecting which aspects to display does not change the underlying criterion evaluation. «Mobile operability» is not shown in the profile charts; the aspect remains part of the unchanged criterion score K and therefore the overall score G.
Accessibility classes applicable to this edition
| Overall class from G | Score interval (points) |
|---|---|
| Good accessibility | 90 ≤ G ≤ 100 |
| Conditional accessibility | 80 ≤ G < 90 |
| Insufficient accessibility | 60 ≤ G < 80 |
| Poor accessibility | 0 ≤ G < 60 |
Mapping criteria to aspects and impairment groups
The table shows the accessibility-profile aspect and impairment groups assigned to each manually tested WCAG criterion in this study. Membership is defined by the template configuration; a criterion can belong to several impairment groups.
| Criterion | Accessibility-profile aspect | Motor skills | Vision | Hearing | Cognition |
|---|---|---|---|---|---|
| 1.1.1 Non-text Content | Non-text content (graphics) | – | yes | – | yes |
| 1.2.1 Audio-only and Video-only (Prerecorded) | Multimedia and two-senses principle (audio/video) | – | yes | yes | yes |
| 1.2.2 Captions (Prerecorded) | Multimedia and two-senses principle (audio/video) | – | – | yes | – |
| 1.2.3 Audio Description or Media Alternative (Prerecorded) | Multimedia and two-senses principle (audio/video) | – | yes | – | yes |
| 1.2.4 Captions (Live) | Multimedia and two-senses principle (audio/video) | – | – | yes | – |
| 1.2.5 Audio Description (Prerecorded) | Multimedia and two-senses principle (audio/video) | – | yes | – | yes |
| 1.3.1 Info and Relationships | Semantic structure | – | yes | – | yes |
| 1.3.2 Meaningful Sequence | Semantic structure | – | yes | – | yes |
| 1.3.3 Sensory Characteristics | Contrast and sensory characteristics | – | yes | yes | – |
| 1.3.4 Orientation | Mobile operability | yes | yes | – | – |
| 1.3.5 Identify Input Purpose | Interaction assistance | yes | – | – | yes |
| 1.4.1 Use of Color | Contrast and sensory characteristics | – | yes | – | – |
| 1.4.2 Audio Control | Display flexibility | – | yes | yes | – |
| 1.4.3 Contrast (Minimum) | Contrast and sensory characteristics | – | yes | – | – |
| 1.4.4 Resize text | Display flexibility | – | yes | – | – |
| 1.4.5 Images of Text | Display flexibility | – | yes | – | yes |
| 1.4.10 Reflow | Display flexibility | – | yes | – | yes |
| 1.4.11 Non-Text Contrast | Contrast and sensory characteristics | – | yes | – | – |
| 1.4.12 Text Spacing | Display flexibility | – | yes | – | yes |
| 1.4.13 Content on Hover or Focus | Display flexibility | yes | yes | – | – |
| 2.1.1 Keyboard | Keyboard operability | yes | yes | – | – |
| 2.1.2 No Keyboard Trap | Keyboard operability | yes | yes | – | – |
| 2.1.4 Character Key Shortcuts | Voice control | yes | – | – | – |
| 2.2.1 Timing Adjustable | Interaction assistance | yes | yes | yes | yes |
| 2.2.2 Pause, Stop, Hide | Display flexibility | – | yes | – | yes |
| 2.3.1 Three Flashes or Below Threshold | Display flexibility | – | – | – | yes |
| 2.4.1 Bypass Blocks | Keyboard operability | yes | yes | – | yes |
| 2.4.2 Page Titled | Understandability | – | yes | – | yes |
| 2.4.3 Focus Order | Keyboard operability | yes | yes | – | yes |
| 2.4.4 Link Purpose (In Context) | Understandability | yes | yes | – | yes |
| 2.4.5 Multiple Ways | Understandability | – | yes | – | yes |
| 2.4.6 Headings and Labels | Understandability | – | yes | – | yes |
| 2.4.7 Focus Visible | Keyboard operability | yes | yes | – | yes |
| 2.4.11 Focus Not Obscured (Minimum) | Keyboard operability | yes | yes | – | – |
| 2.5.1 Pointer Gestures | Mobile operability | yes | – | – | – |
| 2.5.2 Pointer Cancellation | Interaction assistance | yes | yes | – | yes |
| 2.5.3 Label in Name | Voice control | yes | yes | – | – |
| 2.5.4 Motion Actuation | Mobile operability | yes | – | – | – |
| 2.5.7 Dragging Movements | Mobile operability | yes | – | – | – |
| 2.5.8 Target Size (Minimum) | Mobile operability | yes | yes | – | – |
| 3.1.1 Language of Page | Understandability | – | yes | – | yes |
| 3.1.2 Language of Parts | Understandability | – | yes | – | yes |
| 3.2.1 On Focus | Consistency and predictability | yes | yes | – | yes |
| 3.2.2 On Input | Consistency and predictability | – | yes | – | yes |
| 3.2.3 Consistent Navigation | Consistency and predictability | – | yes | – | yes |
| 3.2.4 Consistent Identification | Consistency and predictability | – | yes | – | yes |
| 3.2.6 Consistent Help | Consistency and predictability | – | – | – | yes |
| 3.3.1 Error Identification | Interaction assistance | – | yes | – | yes |
| 3.3.2 Labels or Instructions | Interaction assistance | yes | yes | – | yes |
| 3.3.3 Error Suggestion | Interaction assistance | yes | yes | – | yes |
| 3.3.4 Error Prevention (Legal, Financial, Data) | Interaction assistance | yes | yes | yes | yes |
| 3.3.7 Redundant Entry | Interaction assistance | yes | – | – | yes |
| 3.3.8 Accessible Authentication (Minimum) | Interaction assistance | yes | yes | – | yes |
| 4.1.2 Name, Role, Value | Compatibility with user agents | yes | yes | – | yes |
| 4.1.3 Status Messages | Compatibility with user agents | – | yes | – | – |
