Study
EU bank PDF accessibility study: 2,571 documents tested
Since 28 June 2025 the European Accessibility Act has covered consumer banking. FileSeen downloaded 2,571 public PDFs from 151 banks in 21 EU countries and checked each one. Only 24.5% passed five basic accessibility checks.
In short
- Only 630 of 2,571 bank PDFs (24.5%) passed five basic accessibility checks.
- One in three (32.7%) had no tags at all, so a screen reader gets no headings, lists or table structure.
- Only 6.0% passed the machine-checkable rules of PDF/UA-1, tested with the open-source veraPDF validator.
- Files dated from 28 June 2025 did better (33.5% basic pass against 15.5%), but two in three still failed.
- Automated checks find minimum failures only: a pass does not prove a file is accessible.
Key findings
Results for 2,571 PDFs from 151 banks in 21 EU member states, fetched on 24 October 2026.
| Measure | Result | Files |
|---|---|---|
| Passed all five basic checks | 24.5% | 630 of 2,571 |
| Tagged (structure tree and Marked flag) | 67.3% | 1,730 of 2,571 |
| Document language declared | 76.5% | 1,968 of 2,571 |
| Title set and shown in the title bar | 31.0% | 797 of 2,571 |
| Real, extractable text | 98.0% | 2,519 of 2,571 |
| Figure tags with alt text (tagged files) | 20.4% | 1,785 of 8,754 figures |
| Bookmarks in files of 10+ pages | 29.9% | 221 of 739 |
| Claim PDF/UA in their metadata | 11.9% | 306 of 2,571 |
| Passed veraPDF PDF/UA-1 machine checks | 6.0% | 153 of 2,568 |
What the European Accessibility Act says about bank PDFs
The European Accessibility Act (EAA) is Directive (EU) 2019/882. Article 2(2) lists “consumer banking services” and “e-commerce services” among the services it covers, and Article 31 says Member States apply the rules from 28 June 2025.
Annex I asks service providers to make information available “in text formats that can be used to generate alternative assistive formats”, to supplement “any non-textual content with an alternative presentation”, and to make electronic information “perceivable, operable, understandable and robust”. For consumer banking, information must also be understandable at no more than level B2 of the Common European Framework of Reference for Languages.
The Directive does not name a PDF standard. PDF/UA (ISO 14289-1) and the PDF guidance in WCAG and EN 301 549 are the usual technical references; this study uses them as a yardstick, not as a legal test.
The Dutch Authority for the Financial Markets (AFM), designated to supervise banking and financial e-commerce services under the EAA, states: “In cases where it is mandatory to provide information via a durable medium, and this is done via a pdf, then the pdf file must also be accessible.” It adds that contracts “that are modified (for example, when general terms and conditions are changed) from June 28, 2025” must meet the accessibility obligations.
Two caveats
Article 2(4)(b) exempts “office file formats published before 28 June 2025” on websites, and Article 32 lets older service contracts run on for up to five years.
This study measures accessibility features, not legal compliance. It does not say that any bank breaks the law.
Newer documents are better, but most still fail
To allow for the exemption for older files, the sample is split by the date stored inside each PDF (the later of its creation and modification date).
| PDF internal date | Files | Tagged | Basic pass | PDF/UA-1 |
|---|---|---|---|---|
| From 28 June 2025 | 1,322 | 73.3% | 33.5% | 7.5% |
| Before 28 June 2025 | 1,145 | 61.2% | 15.5% | 4.6% |
| No date in the file | 104 | 57.7% | 8.7% | 1.0% |
Files dated after the EAA started applying pass the basic checks more than twice as often. Still, two in three of them fail at least one check, and 353 of them have no tags at all. Among fee and price documents dated from 28 June 2025, 99 of 397 are untagged.
Results by country
German banks had the highest basic pass rate (62.1%) among countries with five or more banks sampled; Portugal (2.1%), Luxembourg (1.1%) and France (5.1%) the lowest. With 2 to 12 banks per country, these figures describe the banks sampled, not each national market.
| Country | Banks with PDFs | PDFs | Tagged | Language set | Basic pass | PDF/UA-1 |
|---|---|---|---|---|---|---|
| Germany | 12 of 14 | 195 | 71.3% | 79.5% | 62.1% | 25.1% |
| Ireland * | 2 of 6 | 34 | 85.3% | 97.1% | 58.8% | 0.0% |
| Austria | 7 of 10 | 132 | 81.1% | 91.7% | 50.0% | 34.1% |
| Finland | 9 of 10 | 163 | 92.0% | 92.6% | 47.2% | 11.0% |
| Denmark | 7 of 11 | 140 | 57.9% | 62.9% | 32.9% | 10.7% |
| Netherlands | 7 of 10 | 137 | 48.9% | 88.3% | 31.4% | 5.1% |
| Czechia | 7 of 9 | 118 | 54.2% | 81.4% | 29.7% | 3.4% |
| Poland | 10 of 12 | 143 | 79.7% | 83.2% | 26.6% | 6.3% |
| Italy | 11 of 14 | 159 | 55.3% | 66.7% | 25.8% | 0.6% |
| Slovenia | 5 of 5 | 82 | 84.1% | 80.5% | 23.2% | 0.0% |
| Belgium | 6 of 11 | 102 | 82.4% | 78.4% | 18.6% | 0.0% |
| Sweden | 7 of 11 | 110 | 68.2% | 72.7% | 16.4% | 2.8% |
| Spain | 11 of 13 | 191 | 60.2% | 60.2% | 15.2% | 1.1% |
| Croatia * | 4 of 6 | 80 | 67.5% | 88.8% | 13.8% | 0.0% |
| Slovakia * | 3 of 6 | 60 | 68.3% | 88.3% | 13.3% | 0.0% |
| Hungary | 7 of 8 | 121 | 71.1% | 80.2% | 10.7% | 0.0% |
| Romania | 7 of 8 | 122 | 66.4% | 72.1% | 8.2% | 0.0% |
| Greece * | 3 of 6 | 60 | 70.0% | 71.7% | 6.7% | 0.0% |
| France | 9 of 14 | 138 | 50.0% | 74.6% | 5.1% | 0.0% |
| Portugal | 11 of 11 | 192 | 63.0% | 62.5% | 2.1% | 0.0% |
| Luxembourg | 6 of 6 | 92 | 58.7% | 67.4% | 1.1% | 0.0% |
| All 21 countries | 151 of 201 | 2,571 | 67.3% | 76.5% | 24.5% | 6.0% |
Results by bank
Bank figures use only the 140 banks with at least five analysed PDFs.
40 banks (28.6%)
had no sampled PDF that passed the basic checks.
15.0%
of its sampled files passed for the median bank.
31 banks
had at least one file that passed veraPDF PDF/UA-1.
1 bank
passed on every sampled file: ING Italia (5 files).
Examples you can reproduce from the dataset
- Postbank (Germany): 19 of 20 sampled PDFs passed the basic checks, and 12 of 20 passed veraPDF PDF/UA-1.
- S-Pankki (Finland): 19 of 20 passed the basic checks; all 19 files dated from 28 June 2025 were tagged.
- ASN Bank (Netherlands): 19 of 20 passed the basic checks, yet none passed PDF/UA-1. Having the basics is not the same as meeting the standard.
- Abanca (Spain): none of 20 sampled PDFs was tagged, including 9 dated from 28 June 2025.
- Česká spořitelna (Czechia): none of 14 sampled PDFs was tagged; 13 were fee or price documents.
- Handelsbanken (Sweden): 1 of 20 sampled PDFs was tagged, and 3 had no text layer at all.
Results for every bank, one row each, are in banks.csv.
Banks that were not checked
50 of the 201 banks on the list produced no PDF. They were not checked and are not scored: no conclusion is drawn about their documents.
| Reason | Banks (not scored) |
|---|---|
| Refused automated access (17)robots.txt denied access or disallowed the homepage | Jyske Bank (Denmark), BNP Paribas (France), Hello bank! (France), La Banque Postale (France), Alpha Bank (Greece), Eurobank (Greece), K&H Bank (Hungary), AIB (Ireland), EBS (Ireland), Haven Mortgages (Ireland), PTSB (Ireland), Rabobank (Netherlands), BOŚ Bank (Poland), Banca Transilvania (Romania), Tatra banka (Slovakia), BBVA (Spain), ICA Banken (Sweden) |
| Homepage unreachable (7)timed out or returned an error | Volksbank Wien (Austria), BNP Paribas Fortis (Belgium), ING Belgium (Belgium), Ringkjøbing Landbobank (Denmark), Sydbank (Denmark), Attica Bank (Greece), ING Nederland (Netherlands) |
| No PDF found (26)PDFs disallowed in robots.txt, links that did not return a PDF, or links shown only with JavaScript | Erste Bank und Sparkassen (Austria), bank99 (Austria), Beobank (Belgium), KBC (Belgium), Keytrade Bank (Belgium), Erste Bank Croatia (Croatia), Zagrebačka banka (Croatia), MONETA Money Bank (Czechia), ČSOB (Czechia), Lunar (Denmark), OP (Finland), Banque Populaire (France), LCL (France), Berliner Volksbank (Germany), N26 (Germany), Banca Mediolanum (Italy), Banca Popolare di Sondrio (Italy), Credem (Italy), bunq (Netherlands), Bank Pekao (Poland), 365.bank (Slovakia), Slovenská sporiteľňa (Slovakia), ING España (Spain), Avanza (Sweden), SEB (Sweden), Sparbanken Skåne (Sweden) |
Why files fail
Missing structure
One file in three has no tags. Without tags, assistive technology cannot tell a heading from a footnote or read a fee table cell by cell.
Images without descriptions
Tagged files hold 8,754 figure tags; only 1,785 have alt text.
No language
603 files (23.5%) declare no language, so a screen reader may read Polish or Portuguese with English pronunciation.
Title not shown
Only 31.0% set a title and tell the viewer to show it, so many files announce themselves by a file name.
Scans are rare
Only 41 files (1.6%) were images with no text layer. The problem is structure, not scans.
Export settings
PDFs made with Microsoft Word or Office were 88.5% tagged but only 21.2% passed, usually for a missing title setting, language or alt text.
The most common PDF/UA-1 failures, as a share of failing files: no PDF/UA identifier (76.9%), title display flag missing (52.2%), content neither tagged nor marked as decoration (52.0%), links or annotations without a description (about 40%), fonts not embedded (32.8%).
Method
Banks. A list of 201 retail banks in 21 EU member states: for each country, the major consumer banking brands (national leaders, the main savings and cooperative bank sites, the post-office bank and large digital banks), 5 to 14 per country. It is a purposive list, not a random sample or an asset ranking. Each national site of a cross-border group counts as its own bank.
Finding PDFs. A crawler started at each bank’s consumer homepage, read its XML sitemaps and up to 35 pages, and collected PDF links on the bank’s own domain. It obeyed robots.txt, waited 2 to 3 seconds between requests to the same site, and did not log in, submit forms or run scripts. Investor relations, annual reports, ESG, press and careers files were excluded. Up to 20 PDFs per bank were downloaded, spread across fees and prices, terms and conditions, pre-contract information, forms and other customer documents, preferring current files. Document types are a keyword guess from the link and file name. All files were fetched on 24 October 2026 between 17:12 and 17:26 UTC.
| Check | Rule |
|---|---|
| Tagged | A structure tree is present and the file is marked as tagged |
| Language | The document language is declared |
| Title | A title is set and the viewer is told to display it |
| Real text | At least 20 extractable characters per page across the first 10 pages |
| Alt text | Every figure tag carries alternative text |
| Bookmarks | Files of 10+ pages have bookmarks (reported, not part of the basic pass) |
| PDF/UA-1 | veraPDF 1.30.3, PDF/UA-1 profile: all machine-checkable rules pass |
A file passes the basic checks when it meets the first five. Structural checks used pikepdf and pypdf; each file was also validated with veraPDF, the open-source PDF validator.
Sanity check. A fixed random sample of 40 files was re-checked with independent code (a raw read of the PDF catalog and the pdfium text engine): all 40 agreed on tagging, language and text. All 41 files flagged as image-only were rendered and inspected by eye.
Limitations
- Automated checks show minimum failures only. They cannot judge whether alt text is meaningful, whether headings and reading order are right, whether tables make sense to a screen reader, colour contrast, or the plain-language rule. Every pass rate here is an upper bound on real accessibility.
- Coverage. 50 of the 201 banks yielded no PDF and are listed above as not checked. Banks with strong bot protection may differ from the rest.
- A sample, not an inventory. Up to 20 files per bank, selected by keyword. Large banks publish thousands.
- Dates come from file metadata, which can be missing or wrong. Re-saving an old file updates its modification date.
- Legal scope. Not every PDF on a bank website is covered by the EAA, and the study does not assess any bank’s legal compliance.
- Point in time. Bank websites change daily. Results describe the files fetched on 24 October 2026.
Download the dataset
Free to reuse under the Creative Commons Attribution 4.0 licence. Credit FileSeen.
- pdfs.csvOne row per PDF (2,571 rows): bank, country, source URL, document type, fetch time and every check result, including veraPDF. 2.0 MB
- banks.csvOne row per bank (201 rows), including banks that could not be checked and why. 39 KB
- summary.jsonEvery aggregate on this page, with definitions. 36 KB
Licence: CC BY 4.0. You may share and adapt the data for any purpose, including commercially, if you credit FileSeen and link to this page.
Suggested credit: FileSeen, “EU bank PDF accessibility study”, 24 October 2026, https://fileseen.com/studies/eu-bank-pdf-accessibility/
The PDFs themselves are not redistributed: each row links to the bank’s original URL. Data last checked 24 October 2026.
Check a PDF yourself
veraPDF is free and runs the same PDF/UA-1 tests used here. FileSeen’s free PDF tools do not test or fix accessibility, and two of them change it: the PDF compressor turns every page into an image, so the smaller copy has no selectable text, and the redaction tool turns redacted pages into images. Use them on a copy for sharing, not on the version published for customers.
FAQ
Questions about the study
Does the European Accessibility Act apply to bank PDFs?
The EAA covers consumer banking services from 28 June 2025 and requires service information to be perceivable, operable, understandable and robust. The Dutch regulator AFM says that where information must be given on a durable medium and a PDF is used, the PDF must also be accessible. Office files published on websites before 28 June 2025 are exempt under Article 2(4)(b).
What is a tagged PDF?
A tagged PDF contains a hidden structure that marks headings, paragraphs, lists, tables and images, and the order to read them in. Screen readers and reflow tools rely on it. An untagged PDF can look perfect on screen and still be very hard to use with assistive technology.
Does passing these checks make a PDF accessible?
No. The checks find the most basic failures. A file that passes still needs a person to confirm that its alt text, headings, reading order and tables are correct.
What is PDF/UA?
PDF/UA (ISO 14289-1) is the international standard for accessible PDF. veraPDF, an open-source validator, tests the parts of it that a machine can check.
Which banks were tested?
201 retail banks across 21 EU member states were on the list; 151 produced at least one PDF. The full list, with the outcome for each bank, is in banks.csv.
Why are some large banks missing?
Their websites refused automated requests or did not expose PDFs to a polite crawler. They are listed as not checked, with the reason. Missing data says nothing about how accessible their documents are.
Are newer documents better?
Yes. Files dated from 28 June 2025 passed the basic checks 33.5% of the time, against 15.5% for older files. Two in three still failed.
Can I reuse the data?
Yes, under CC BY 4.0: credit FileSeen and link to this page. A rerun on another day would collect different files and give somewhat different numbers.