Alt Text for Scientific Figures: The Complete Guide

Diese Seite auf Deutsch

A blind researcher opens your paper with a screen reader. The text reads well. Then the screen reader reaches Figure 3 and says: "Figure 3. Image." Whatever that figure showed, the reader now has two options: trust your discussion section or skip the result.

The short answer: write the alt text yourself, at the same time as the caption. A chart or other complex figure needs more than the alt text. The alt text is one to three sentences that name the figure type, what it measures and the main finding. The rest goes into the visible document: a long description and, where readers need exact values, a table. The author knows what a figure is meant to show and has the data behind it, so the author is best placed to write the content of the description. An accessibility specialist can then check its form and build it into an accessible PDF.

This guide is for researchers, doctoral and master's students, supervisors and university accessibility teams. The practical part comes first: the method, a worked example, figure types and software settings. AI, costs, standards and law follow as reference. For a short version with many good and weak examples, read How do you write good alt text for a chart? DTP Services Berlin turns research documents into tagged, tested PDF/UA files; the PDF accessibility service page explains that work.

Why should researchers write their own alt text?

Because almost nobody else does it, and describing a research figure well takes subject knowledge. Of 11,397 papers published between 2010 and 2019, only 7.5% had any alt text for their figures^[1]. Having alt text is not the same as having useful alt text: a follow-up study of 19,997 PDFs from 2014 to 2023 could extract alt text from 1,536 papers. 95% of those contained placeholders such as "Image" or "Figure", and only 5 papers, about 0.03% of the whole sample, had alt text that described a figure^[2]. Even in human-computer interaction, a field that studies accessibility, only 31% of chart alt texts mentioned the main trend and 50% mentioned the extremes or outliers^[3].

Describing someone else's figure is slow without the background. In a 2026 study, university disability-services staff had 50 minutes to describe four research figures without AI help. All twelve used the full time, two did not finish, and their descriptions often repeated the caption and lacked detail. With AI help, the same staff worked faster and scored higher overall, although they often accepted the AI text without checking it against the paper^[4]. The study is small and covers one field, but it shows what subject knowledge adds. The University of Washington came to a similar view for its central teams: staff may lack the subject expertise to understand and translate complex images^[5].

Publishers have drawn the same line. Taylor & Francis, AIP Publishing, ACM and De Gruyter ask authors to write the alt text for their figures^{[6, 7, 8, 9]}. Springer Nature keeps authors responsible for the accuracy of alt text even when an AI tool suggests it, and does not correct alt text after publication^[10].

What are the three layers of a figure description?

DTP Services Berlin recommends three layers for research figures. They follow the W3C advice for complex images, a short alternative plus a long description that can include a table, and what blind readers asked for in studies^{[11, 12]}. They are a working method, not a structure that WCAG or PDF/UA prescribes.

A simple photo needs only the first layer. Most charts need the first two. Add a table when readers need exact values or the values are not reported anywhere else; if they already sit in a table in the paper, point to that table instead of repeating it.

The caption stays a separate element and is read by everyone. It often names the variables and the context already; the alt text adds what a sighted reader gets from looking at the figure. Repeating the caption in the alt text makes screen reader users hear it twice^{[13, 14]}. Lists, tables and line breaks do not belong in the alt text field: in a PDF, alt text is one plain string that a screen reader reads in one go, so anything with structure goes into layers 2 and 3^{[8, 55]}.

How long should alt text be?

No standard sets a character limit. The often-quoted limit of 125 characters does not come from WCAG or PDF/UA; neither sets a length^{[11, 23, 47]}. Where a publisher uses the number, it applies to one sentence: ACM asks for a first sentence under 125 characters^[8]. Published guidance gives ranges instead: at most 80 to 100 characters in the German federal guidance for teaching materials^[16], 100 to 150 characters at De Gruyter^[9], 25 to 50 words at AIP Publishing^[7], and up to 200 characters at Elsevier^[15], plus a long description it aims to keep under 2,000 characters^[55]. In a study with blind and low-vision participants, people asked for two to eight sentences for a chart, depending on its complexity^[17]. The layers resolve the conflict: keep the alt text short and put everything else in the long description.

How do you write a figure description, step by step?

Keep three kinds of statement apart. An observation is what the chart shows ("mean scores were 71 and 62 points"). A result comes from your analysis ("the difference is significant, p = 0.002") and belongs in the long description or caption only if it is in your results. An interpretation is your conclusion ("the method improves learning") and belongs in the discussion. Examples of all three, plus good and weak alt texts for bar charts, micrographs, multi-panel figures, flowcharts and scatter plots, are in How do you write good alt text for a chart?

What do blind readers want from a chart description?

Two peer-reviewed studies give a clear answer. Lundgard and Satyanarayan sorted chart descriptions into four levels: how the chart is built (type, axes, colours), statistics (values, extremes, outliers), trends and patterns, and the author's interpretation. Blind readers rated statistics and trends as the most useful levels and the chart's construction and the author's interpretation as the least useful. 19 of the 30 blind participants (63%) did not want the author's subjective interpretation in the description^[18].

Jung and colleagues worked with 22 people with visual impairments. Participants wanted the chart type first, then a summary, then axes and ranges, then trends together with the range of values in which each trend appears. They preferred a real table to a table written out in the alt text, and they asked for data, not adjectives such as "a huge gap"^[17].

Worked example: one chart, three layers

The chart below uses invented example data. It shows a typical result figure: two groups, one variable, measured over time.

Layer 1: the alt text

Layer 2: the long description

This text sits in the visible document, directly below the figure or in an appendix, so sighted readers can use it too. If the figure had error bars, the long description would say what they show (for example, "error bars show 95% confidence intervals") and whether the intervals overlap.

Layer 3: the data table

The chart itself also follows two rules that help every reader: the two groups differ in marker shape (circles and squares) as well as colour, and each line carries its label at its end. Meaning never depends on colour alone.

How do you describe each type of figure?

The pattern stays the same for every figure: type, content, finding, pointer. The table lists what to add for common figure types in research documents; worked examples for several of them are in the short writing guide. The approach builds on two established sets of guidelines for describing STEM images: those of the National Center for Accessible Media (NCAM) at WGBH, based on surveys with describers and with blind and visually impaired students and scientists, and the Image Description Guidelines that NCAM developed with the DIAGRAM Center^{[19, 20]}.

Significance markers have no rule in any standard or publisher guide reviewed for this article. DTP Services Berlin recommends stating the result in words in the long description ("the difference is significant from week 3, p < 0.01") instead of describing asterisks, and only when the test is part of your results.

For formulas that remain images, the sources disagree. The German federal guidance recommends LaTeX code in the alt text, while Elsevier writes formulas out in spoken form^{[16, 55]}. Trained STEM screen reader users can follow LaTeX; other readers cannot. For a mixed audience, use the spoken form ("x squared plus y squared equals r squared"); for a specialist audience, LaTeX is an option.

In exams and exercise sheets, the alt text must not give away the answer. If a task asks which trend a chart shows, describe the values and leave the trend to the student^[16].

Where do you enter alt text in your software?

Write the alt text in the program where the document is made. Changes made directly in a PDF do not flow back into the Word, LaTeX or InDesign file, so they are lost when the PDF is exported again from that file.

Draft descriptions from R and Python plots

A plotting library knows the exact axes and values of a figure, so some tools write a first description from the plot object instead of from the image. In R, the VI() function of the BrailleR package turns a ggplot2 or base R graph into text: chart type, title, axes, labels and, for small plots, the individual values. Its documentation calls the output experimental and warns that some of it can be misleading^[34]. In Python, MatplotAlt writes alt text for Matplotlib figures in Jupyter notebooks, either with rules applied to the figure data or with an AI model. In its evaluation, the AI-written texts still contained factual errors, and they became more accurate when the model also received the rule-based text or the data table^[35]. For charts published on the web with Vega-Lite or Observable Plot, Olli turns the chart into a text structure that screen reader users can explore with the keyboard^[36].

These tools describe what a plot shows. The finding, the reason the figure is in the paper, still has to come from the author.

A minimal LaTeX example

Since November 2025 the LaTeX Project describes tagging as ready for production use with supported packages^[29]. The metadata line comes before \documentclass:

\DocumentMetadata{lang=en-GB, pdfversion=2.0, pdfstandard=ua-2, tagging=on}
\documentclass{article}
\usepackage{graphicx}
\title{Seedling growth under fertiliser treatment}
\begin{document}
\begin{figure}
  \centering
  \includegraphics[alt={Line chart of mean seedling height:
    with fertiliser, seedlings reached 11.6 cm after six weeks,
    against 6.9 cm in the control group. Values in Table 1.}]{height.pdf}
  \caption{Mean seedling height over six weeks.}
\end{figure}
\end{document}

Use a current TeX Live (2025 or later), preferably with LuaLaTeX, the engine the LaTeX Project recommends for new documents^[59]. The \title line matters: PDF/UA requires a document title in the metadata (Matterhorn checkpoint 06-003)^[23]. Check the PDF afterwards: packages that do not support tagging yet can break the structure, and the log file reports their status.

Long descriptions and data tables in the document

Write the long description as normal text directly below the figure or in an appendix, and start it with a clear label such as "Description of Figure 1:". The alt text then points to it. Build data tables as real tables, never as images: in Word, use a table with a header row (Table Design, Header Row) and set it to repeat on each page (Layout, Repeat Header Rows); in LaTeX, a tabular is tagged as a table, but every cell counts as a data cell until you declare the header row with \tagpdfsetup{table/header-rows={1}}. Put the command in the preamble and it applies to all tables; row headers in the first column are set the same way with table/header-columns={1}. In a longtable, the rows before \endhead are used as the header, and tagging a longtable currently needs LuaLaTeX^[37]. One header row and no merged cells make tables easiest to read with a screen reader.

The figure list: one table for all your figures

The simplest working method is a figure list kept next to the manuscript. De Gruyter collects alt texts with a template of this kind^[9]. Fill one row when you create each figure; the list then feeds the alt text fields, the appendix tables and, if a provider prepares the final PDF, the remediation.

Can AI write alt text for scientific figures?

AI can write a useful first draft. It cannot yet be trusted with the numbers, and a reader who cannot see the figure cannot check a description against the image. An accessible data table is what makes checking possible. Studies support both parts of that answer; most of them test chart understanding in general and only point towards how well AI writes alt text.

Word, PowerPoint and InDesign can generate alt text with AI. Microsoft's own documentation warns that it may be incomplete or imprecise and recommends human-written text where accuracy is critical^[42]. InDesign does it automatically when an image is placed if the option is switched on, which is the default for accounts with unlimited generative credits^[58]. In an example in a W3C editor's draft (not a W3C standard), Word described a bar chart only as "a graph with different coloured bars"^[43]. A description like that names the visual and misses the result.

Two further points apply to research. Unpublished manuscripts and data should not be uploaded to a cloud AI service without checking the rules of the university and the funder; FernUniversität in Hagen built its AI description prototype on local servers for this reason, and treats the output as a starting point that a person must approve^[44]. And medical, microscopy and chemical images are where AI errors look most convincing: a model that names the method correctly and the finding wrongly produces a description that sounds right^{[41, 45]}.

Who writes the alt text, and what does that cost?

Three kinds of work are often mixed up: writing the descriptions, preparing the data tables, and building the accessible PDF (tagging, placing and testing). Remediation providers that publish prices mostly charge per page for the third and put pages with complex figures into a higher tier. In the public price lists reviewed for this guide (European and US providers, November 2026), none showed a separate price for long descriptions, and none a lower rate for author-supplied alt text.

Whatever the pricing model, someone has to understand each figure before describing it. The 2026 study planned 10 to 12 minutes per figure for staff without subject knowledge^[4]:

Whether author-written descriptions also lower an invoice depends on the provider's pricing model. They reliably reduce questions, corrections and the risk of a scientifically wrong description. Whether a provider writes descriptions at all, and how, is worth agreeing before a project starts.

Standards and law: a reference

WCAG. Success Criterion 1.1.1 of WCAG 2.2 requires a text alternative that serves the same purpose for every informative image^[47]. For charts and diagrams, W3C documents a short alternative combined with a long description as a sufficient technique (G92); other ways of providing the long description are possible^{[11, 12]}.

EN 301 549. In the EU, WCAG reaches PDF documents through the European standard EN 301 549. The version currently cited in the EU Official Journal is v3.2.1, based on WCAG 2.1 Level AA^[48]. Version 4.1.1, based on WCAG 2.2, was published in September 2026 but, at the time of writing, is not yet cited in the Official Journal^[49]. Success Criterion 1.1.1 is the same in WCAG 2.1 and 2.2.

PDF/UA. PDF/UA-1 (ISO 14289-1) and PDF/UA-2 (ISO 14289-2, 2024) are the standards for accessible PDF^{[50, 51]}. The Matterhorn Protocol is the test model for PDF/UA-1: an informative graphic must be tagged as a Figure (checkpoint 13-001) with a text alternative (13-004), captions are tagged as Caption, decorative graphics are marked as artifacts, and an image fails when a more accessible form such as a real table or real text would be possible (13-007)^{[23, 52]}. Equations are tagged as Formula; in PDF/UA-1 each Formula needs alt text (17-002). PDF/UA-2 requires mathematical expressions to be represented in MathML within the Formula structure element. The same Formula can also carry alt text in spoken form: software that reads MathML uses the MathML, and other software falls back to the alt text^[22]. MathML is not yet read reliably by every PDF viewer and screen reader, so the spoken form is worth adding.

Law. Directive (EU) 2016/2102 requires public bodies to make their websites and the documents on them accessible, with exceptions, for example for certain office files published before 23 September 2018^[53]. Guidance published by Germany's federal monitoring body (BFIT-Bund) states that this covers universities and their teaching materials^[16]. The European Accessibility Act, Directive (EU) 2019/882, has applied since 28 June 2025 to specified products and services, including e-books; microenterprises providing services are exempt^[54]. It does not make every research PDF subject to the Act. De Gruyter names it as the reason for its accessibility requirements for book manuscripts^[9]. National rules differ; the German version of this guide covers BGG, BITV 2.0, the state laws and the BFSG.

DTP Services Berlin checks and documents technical conformance with PDF/UA and WCAG. Whether a document meets a legal obligation is for the publishing institution to assess.

Frequently asked questions

For supervisors and university accessibility teams

1. Make the figure list part of thesis and report submission.
2. Teach the three layers and the six steps in writing courses and doctoral training, with one worked example from the student's own field.
3. Require a tagged PDF export; Print to PDF creates no tags.
4. Read the alt text of a sample of figures yourself; a checker only confirms that alt text exists.
5. Agree on a local policy for AI use with unpublished data.
6. Add accessibility to the guidance for publishing on the university repository.

The short version

• The author knows what a figure is meant to show, so the author writes the content of the description, at the same time as the caption.
• Recommended for research figures: a short alt text, a long description in the visible text and, where exact values matter, a data table.
• Alt text order: figure type, what is measured, the main finding with numbers, where the details are.
• There is no 125-character rule. Keep the alt text to one to three sentences and put detail in the long description.
• AI can write a first draft. Every number needs checking, and unpublished data should not be uploaded without permission.
• Enter alt text in Word, LaTeX or InDesign, export a tagged PDF and check the result.

Sources

All sources were checked in November 2026. Each is marked as standard or law, research, or guidance from universities, publishers and specialist organisations. Preprints have not completed peer review. Two vendor documents (Microsoft, Adobe) are cited only for what they say about their own products.

1. ResearchPreprint Wang, L. L., Cachola, I., Bragg, J. et al. Improving the Accessibility of Scientific Documents: Current State, User Needs, and a System Solution to Enhance Scientific PDF Accessibility for Blind and Low Vision Users. 2021.
   https://arxiv.org/abs/2105.00076
2. Research Kumar, A. and Wang, L. L. Uncovering the New Accessibility Crisis in Scholarly PDFs. ASSETS 2024.
   https://arxiv.org/abs/2410.03022
3. Research Chintalapati, S. S., Bragg, J. and Wang, L. L. A Dataset of Alt Texts from HCI Publications: Analyses and Uses Towards Producing More Descriptive Alt Texts of Data Visualizations in Scientific Papers. ASSETS 2022.
   https://arxiv.org/abs/2209.13718
4. ResearchPreprint Raees, M., Iwamoto, Y., Papangelis, K., Heard, J. and Tigwell, G. W. How University Disability Services Professionals Write Image Descriptions for HCI Figures Using Generative AI. version 2, September 2026.
   https://arxiv.org/abs/2602.08937
5. Guidance University of Washington. Academic Course Content Action Team: Images report. 2025.
   https://www.washington.edu/accessibility/academic-course-content-action-team/images-report/
6. Guidance Taylor & Francis Author Services. An author's guide to writing good alt text.
   https://authorservices.taylorandfrancis.com/publishing-your-research/writing-your-paper/how-to-make-your-research-accessible/an-authors-guide-to-writing-good-alt-text/
7. Guidance AIP Publishing. Alt text author instructions. 2026.
   https://publishing.aip.org/wp-content/uploads/2026/03/Final_Alt-text_author_guidelines.pdf
8. Guidance ACM. Describing Figures for ACM Publications.
   https://authors.acm.org/proceedings/production-information/describing-figures
9. Guidance De Gruyter. Barrierefreie Manuskripterstellung. As of 4 April 2025.
   https://degruyter-live-craftcms-assets.s3.amazonaws.com/DGB_25/Documents/DeGruyter_Barrierefreie_Manuskripterstellung_v4.pdf
10. Guidance Springer Nature. Accessibility requirements for figures, images and supplementary materials.
    https://www.springernature.com/gp/policies/accessibility-figures-images
11. Standard / law W3C Web Accessibility Initiative. Technique G92: Providing long description for non-text content that serves the same purpose and presents the same information.
    https://www.w3.org/WAI/WCAG22/Techniques/general/G92
12. Guidance W3C Web Accessibility Initiative. Images Tutorial: Complex Images. Updated April 2026.
    https://www.w3.org/WAI/tutorials/images/complex
13. Guidance Springer Nature. How to write good alt text.
    https://www.springernature.com/gp/authors/publish-a-book/writing-your-manuscript/accessibility/how-to-write-good-alt-text
14. Guidance WebAIM (Utah State University). Alternative Text.
    https://webaim.org/techniques/alttext/
15. Guidance Elsevier. Journal Author Guide to Alt Text for Images. March 2026.
    https://supportcontent.elsevier.com/Support%20Hub/Journals/Journal_Author_Guide_Alt_Text.pdf
16. Guidance BFIT-Bund (Überwachungsstelle des Bundes für Barrierefreiheit von Informationstechnik). Barrierefreie Dokumente in Lernkontexten, version 2.2. 25 March 2026.
    https://handreichungen.bfit-bund.de/bf-dokumente-lernkontext/2.2/docs/barrierefreie-dokumente-in-lernkontexten.pdf
17. Research Jung, C., Mehta, S., Kulkarni, A., Zhao, Y. and Kim, Y.-S. Communicating Visualizations without Visuals: Investigation of Visualization Alternative Text for People with Visual Impairments. IEEE Transactions on Visualization and Computer Graphics 28(1), 2022 (IEEE VIS 2021).
    https://arxiv.org/abs/2108.03657
18. Research Lundgard, A. and Satyanarayan, A. Accessible Visualization via Natural Language Descriptions: A Four-Level Model of Semantic Content. IEEE Transactions on Visualization and Computer Graphics 28(1), 2022 (IEEE VIS 2021).
    https://vis.mit.edu/pubs/vis-text-model.pdf
19. Research Gould, B., O'Connell, T., Freed, G. Effective Practices for Description of Science Content within Digital Talking Books. WGBH National Center for Accessible Media. December 2008.
    https://www.wgbh.org/ncam-resources/2026-08-11/effective-practices-for-description-of-science-content-introduction
20. Guidance NCAM at WGBH with the DIAGRAM Center (Benetech). Image Description Guidelines. Archived by GBH.
    https://www.wgbh.org/ncam-resources/2026-08-19/image-description-guidelines
21. Guidance Penn State Accessibility. Flowcharts & Concept Maps.
    https://accessibility.psu.edu/images/flowcharts/
22. Guidance PDF Association. Best Practice Guide: Math in PDF, version 1.0. 2026.
    https://pdfa.org/download-area/publications/BPG-Math-in-PDF.pdf
23. Guidance PDF Association. Matterhorn Protocol checks for PDF/UA (checkpoints 06 Metadata, 13 Graphics and 17 Mathematical expressions). Tagged PDF School.
    https://taggedpdfschool.pdfa.org/api/db/md/matterhorn
24. Guidance University of Washington, Accessible Technology. PDFs from Microsoft Word.
    https://www.washington.edu/accesstech/documents/word-pdfs
25. Guidance WebAIM. Word and PowerPoint Alt Text Roundup. 31 October 2025.
    https://webaim.org/blog/word-and-powerpoint-alt-text-roundup/
26. Guidance Stanford University IT. Accessible Google Documents. May 2026.
    https://uit.stanford.edu/accessibility/guides/google-docs
27. Guidance Penn State Accessibility. InDesign: Add Image ALT Text.
    https://accessibility.psu.edu/pdf/indesign/indesignalttext
28. Guidance LaTeX Project. LaTeX News, Issue 41. June 2025.
    https://latex-project.org/news/latex2e-news/ltnews41.pdf
29. Guidance LaTeX Project. LaTeX News, Issue 42. November 2025.
    https://latex-project.org/news/latex2e-news/ltnews42.pdf
30. Guidance Typst. Changelog, version 0.14.0 (24 October 2025): tagged PDF by default and PDF/UA-1 export.
    https://typst.app/docs/changelog/0.14.0/
31. Guidance Quarto (Posit). Figures: alt text.
    https://quarto.org/docs/authoring/figures.html
32. Guidance Quarto (Posit). PDF Basics: PDF standards and accessibility.
    https://quarto.org/docs/output-formats/pdf-basics.html
33. Guidance Xie, Y. knitr: Options, chunk option fig.alt.
    https://yihui.org/knitr/options/
34. Guidance Godfrey, A. J. R., et al. BrailleR: VI(), extract information from a graphical object. R package documentation, CRAN.
    https://search.r-project.org/CRAN/refmans/BrailleR/html/VI.html
35. Research Nylund, K., Mankoff, J., Potluri, V. MatplotAlt: A Python Library for Adding Alt Text to Matplotlib Figures in Computational Notebooks. Computer Graphics Forum 44(3), 2025.
    https://doi.org/10.1111/cgf.70119
36. Research Blanco, M., Zong, J., Satyanarayan, A. Olli: An Extensible Visualization Library for Screen Reader Accessibility. IEEE VIS Posters, 2022.
    https://vis.mit.edu/pubs/olli
37. Guidance LaTeX Project (Mittelbach, F., Fischer, U.). The latex-lab-table package: tagging of tables. Version 0.85v, September 2026.
    https://mirrors.ctan.org/macros/latex/required/latex-lab/latex-lab-table.pdf
38. Research Hsu, T.-Y. et al. Do Large Multimodal Models Solve Caption Generation for Scientific Figures? Lessons Learned from SciCap Challenge 2023. Transactions of the Association for Computational Linguistics 14 (2026), 233-252.
    https://doi.org/10.1162/tacl.a.653
39. Research Wang, Z. et al. CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs. NeurIPS 2024 Datasets and Benchmarks Track.
    https://arxiv.org/abs/2406.18521
40. ResearchPreprint Eliza, I. J. and Rahman, M. D. (University of Utah). Chart-Supported or Model-Supplied? Examining MLLM-Generated Claims for Accessible Visualization. (workshop submission, IEEE VIS 2026), 2026.
    https://arxiv.org/abs/2607.25021
41. Research Brin, D., Sorin, V., Barash, Y. et al. Assessing GPT-4 multimodal performance in radiological image analysis. European Radiology 35 (2025).
    https://link.springer.com/article/10.1007/s00330-024-11035-5
42. Guidance Microsoft Support. Frequently asked questions about AI-generated alt text. Updated August 2026.
    https://support.microsoft.com/en-us/topic/frequently-asked-questions-about-ai-generated-alt-text-a667a4a6-5d18-45e6-aaf6-5370132f337a
43. Guidance W3C Accessible Platform Architectures Working Group. Accessibility of Machine Learning and Generative AI. Editor's Draft, 27 October 2026 (not a W3C standard).
    https://w3c.github.io/ai-accessibility/
44. Guidance FernUniversität in Hagen. Experimentieren mit KI und Bildbeschreibung (GAAD 2026). 7 May 2026.
    https://www.fernuni-hagen.de/barrierefrei/docs/gaad_2026_ki-bildbeschreibungen.pdf
45. ResearchPreprint Alampara, N. et al. Probing the limitations of multimodal language models for chemistry and materials research (MaCBench). 2025.
    https://arxiv.org/abs/2411.16955
46. Guidance University of Illinois Chicago, Graduate College. Document Accessibility for Theses and Dissertations.
    https://grad.uic.edu/academic-support/exams-defense/theses-dissertations-accessibility
47. Standard / law W3C Web Accessibility Initiative. Understanding Success Criterion 1.1.1: Non-text Content (WCAG 2.2).
    https://www.w3.org/WAI/WCAG22/Understanding/non-text-content.html
48. Standard / law ETSI. EN 301 549 V3.2.1 (2021-03): Accessibility requirements for ICT products and services. Version currently cited in the EU Official Journal; based on WCAG 2.1.
    https://www.etsi.org/deliver/etsi_en/301500_301599/301549/03.02.01_60/en_301549v030201p.pdf
49. Standard / law ETSI. EN 301 549 V4.1.1 (2026-09). Published September 2026, aligned with WCAG 2.2; not yet cited in the EU Official Journal at the time of writing.
    https://www.etsi.org/deliver/etsi_en/301500_301599/301549/04.01.01_60/
50. Standard / law ISO 14289-1:2014. Document management applications: Electronic document file format enhancement for accessibility, Part 1 (PDF/UA-1).
    https://www.iso.org/standard/64599.html
51. Standard / law ISO 14289-2:2024. Document management applications: Electronic document file format enhancement for accessibility, Part 2 (PDF/UA-2).
    https://www.iso.org/standard/82278.html
52. Guidance PDF Association. Figure structure element (PDF/UA-2 and Well-Tagged PDF). Tagged PDF School.
    https://taggedpdfschool.pdfa.org/tags/Figure
53. Standard / law Directive (EU) 2016/2102 on the accessibility of the websites and mobile applications of public sector bodies. EUR-Lex.
    https://eur-lex.europa.eu/eli/dir/2016/2102/oj
54. Standard / law Directive (EU) 2019/882 on the accessibility requirements for products and services (European Accessibility Act). EUR-Lex.
    https://eur-lex.europa.eu/eli/dir/2019/882/oj
55. Guidance Elsevier. Accessibility Compendium for society and partner publishers, version 2.0. February 2026.
    https://supportcontent.elsevier.com/Support%20Hub/Journals/Accessibility_Compendium_Society_Partner_Publishers_v2.0.pdf
56. Guidance Typst. Accessibility guide: accessibility checks on PDF export.
    https://typst.app/docs/guides/accessibility/
57. Guidance Quarto (Posit). Typst Basics: PDF tagging.
    https://quarto.org/docs/output-formats/typst.html
58. Guidance Adobe. InDesign User Guide: Generate Alt Text for images.
    https://helpx.adobe.com/indesign/desktop/add-graphics-and-media/add-edit-graphics/generate-alt-text-for-images.html
59. Guidance LaTeX Project. Tagging Project: Prototype usage instructions. June 2024.
    https://tagging-project.latex-project.org/documentation/prototype-usage-instructions-2024-06-01

Further reading