Industries / Finance and business services / Advertising and marketing / Graphic and web design
Accessibility
Accessibility in design is making brands, documents and websites usable by people with disabilities: readable color contrast and type, written descriptions of images, and PDFs and pages that screen readers and keyboards can work through.
7
Milestones achieved, from 1999 to 2025.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1999
Guidelines ask for alt text
WCAG 1.0, the Web Content Accessibility Guidelines, a checklist for making web pages usable by disabled people, became a W3C Recommendation on 5 May 1999. Its checkpoint 1.1 asked for "a text equivalent for every non-text element," the short written description of an image now called alt text, and checkpoint 2.2 asked for "sufficient contrast" between text and background without giving a number.
Credit
W3C Web Accessibility Initiative; editors Wendy Chisholm and Gregg Vanderheiden (Trace R&D Center, University of Wisconsin-Madison) and Ian Jacobs (W3C)
2008
Contrast gets a number
WCAG 2.0, published on 11 December 2008, set a floor anyone can measure: text needs a contrast ratio of at least 4.5 to 1 against its background, and large text at least 3 to 1. In October 2012 the same text became the international standard ISO/IEC 40500. When WebAIM, an accessibility group that publishes a yearly survey, scanned one million home pages in February 2026, it found text below this ratio on 83.9% of them, the most common failure, and detected failures of the guidelines on 95.9%.
LessonA rule designers can measure is easy to check, and 18 years later most home pages still break it.
Credit
W3C; ISO/IEC JTC 1 for the international standard; WebAIM for the measurement
2012
A standard for accessible PDFs
ISO 14289-1, known as PDF/UA (for universal accessibility), was published in July 2012. It sets out how to use tags, hidden labels that tell a screen reader (software that reads the screen aloud) which text is a heading, list, table or picture and in what order to read it, which gave layout programs a fixed target for the PDFs they export.
Credit
International Organization for Standardization (ISO). The 2012 edition has since been withdrawn and replaced by a revision
2023
A draft contrast formula pulled
A proposed new way of scoring contrast, called APCA, was taken out of the draft of WCAG 3 published on 24 July 2023 after it did not gain the working group's support, and the draft of 10 September 2026 still says "The contrast algorithm used in WCAG 3 is yet to be determined."
LessonColors chosen by a draft formula have no standard behind them; the accessibility consultant Adrian Roselli advises designers to "choose colors that also conform to WCAG2 contrast requirements.".
Credit
W3C Accessibility Guidelines Working Group
2024
US public bodies get deadlines
On 24 April 2024 the US Department of Justice published a rule under the Americans with Disabilities Act requiring the websites and mobile apps of state and local governments to meet WCAG 2.1 at Level AA. On 20 April 2026, four days before the first deadline, it moved the dates back a year, to 26 April 2027 for governments serving 50,000 people or more and 26 April 2028 for smaller ones.
LessonThe department said it had "overestimated the capabilities (whether staffing or technology)" of public bodies, and that generative AI "does not yet reliably automate the remediation of inaccessible content at scale.".
Credit
US Department of Justice
2025
Overlay claims ruled deceptive
accessiBe sold accessWidget, an AI plug-in added on top of a finished website, saying it "can make any website compliant with WCAG." The US Federal Trade Commission alleged that these claims were false, misleading or unsupported and that the company passed off paid reviews as independent ones; on 22 April 2025 it approved, by 3 votes to 0, a final order requiring a $1 million payment and barring such claims without evidence.
LessonA tool laid over a finished site cannot be sold as making it meet the guidelines without proof; the FTC's Samuel Levine said, "Overstating a product's AI or other capabilities without adequate evidence is deceptive.".
Credit
US Federal Trade Commission
2025
EU law reaches private services
The European Accessibility Act, Directive (EU) 2019/882 of 17 April 2019, has applied since 28 June 2025 to services sold to consumers, including online shops, consumer banking and e-books, so the websites and documents designed for them must be accessible. Microenterprises that provide services, those with fewer than 10 staff and no more than €2 million in annual turnover or balance sheet, are exempt.
Credit
European Parliament and Council of the European Union
Watch for
AI alt text in InDesign
InDesign 21.2 added automatic alt text for placed images, in Adobe's words "AI generates descriptive alt text directly on image frames," in English only and with no accuracy figure. Outside checks so far point the other way. Fondazione LIA, an Italian foundation for accessible publishing, tested a photo, a histogram and a diagram in May 2026, found errors in all three and said "we do not consider it ready for use in accessible digital publishing," and a user on Adobe's forum reported a plain brown background described as a nude figure.
Achieved whenan independent tester publishes results on a set of real layout images (logos, icons, charts, photos) in which most descriptions are judged correct and useful and decorative images are left undescribed.
Claimed by
Adobe, January 2026
Automatic tagging at 96.5%
Nutrient, which sells a tool that adds PDF tags automatically, reports that in a March 2026 test its output reached "a 96.5 percent PDF/UA conformance rate with zero critical failures" across 3,157 files checked with veraPDF, a free validator, at 322 files per minute. The test is the vendor's own, and the files are neither described nor published.
Achieved whenan independent party reruns the test on a published set of files and also checks, by hand or with screen-reader users, that headings and reading order are right and not only that the machine checks pass.
Claimed by
Nutrient, March 2026
AI audits on the canvas
Figma says its built-in AI agent, launched in beta on 20 May 2026, lets designers "Audit contrast ratios, text legibility, touch targets, and color blindness issues with AI—directly in Figma Design," so that problems are caught before anything is built. The page gives no accuracy figures and says "manual review is still valuable."
Achieved whenan independent test on design files with known, planted problems reports the share found and the number of false alarms.
Claimed by
Figma, 2026, on an undated product page that refers to the beta
Open
01
Tagging tests on designed documents
A public test set of designed documents, such as annual reports, brochures, menus and newsletters made in InDesign, Affinity and Canva, each with a hand-made reference of its correct headings and reading order. Run every common tagging route on the set and score each against the reference.
The briefCredit · unclaimed02
Alt text for layout images
A public test of AI alt text on the images designers actually place, scored by blind readers. The classes are logos, icons, decorative backgrounds, illustrations, charts, infographics and photos, and the test also checks whether purely decorative images are skipped instead of described.
The briefCredit · unclaimed03
Blind test of palette fixes
A test of whether tools that fix low contrast keep a brand's look. Take real brand palettes that fail the WCAG 2 contrast ratio, fix each with several tools and with a human designer, and have working designers pick, without labels, which fix looks most like the original.
The briefCredit · unclaimed
Open 01 · the brief
Tagging tests on designed documents
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test set of designed documents, such as annual reports, brochures, menus and newsletters made in InDesign, Affinity and Canva, each with a hand-made reference of its correct headings and reading order. Run every common tagging route on the set and score each against the reference.
Why it is still open
The one expert-checked public set has 125 PDFs "focused primarily on scholarly articles," and its authors say it leaves out brochures and reports with complex layouts. A CHI 2025 study of Acrobat's automatic tagging used one three-page conference-paper template and still found the wrong reading order on the first page in 17 of 19 PDFs. The one benchmark covering designed layouts comes from a vendor, RemeDocs, that ranks its own product and publishes no files. The US Justice Department said in April 2026 that generative AI "does not yet reliably automate the remediation of inaccessible content at scale."
Built with what exists
veraPDF, an open-source PDF/UA validator; PAC, the free PDF Accessibility Checker; the free NVDA screen reader; as routes under test, Adobe's PDF Accessibility Auto-Tag API, Acrobat Pro's automatic tagging, Nutrient's tagging, and the tagged export built into InDesign, Affinity and Canva.
First step
Build 10 test documents from the team's own content in the three programs, write down the right heading levels and reading order for each page, export tagged PDFs, and run veraPDF and PAC for a baseline.
Done looks like
A public repository with at least 60 documents, 20 from each layout program, made by the team or used with written permission; a reference tag outline for each; output from at least 4 tagging routes; a table of veraPDF and PAC failures, the share of headings at the right level and the share of pages read in the right order; and results from at least 5 paid, consenting screen-reader users on a 10-document subset.
Open 02 · the brief
Alt text for layout images
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of AI alt text on the images designers actually place, scored by blind readers. The classes are logos, icons, decorative backgrounds, illustrations, charts, infographics and photos, and the test also checks whether purely decorative images are skipped instead of described.
Why it is still open
The only outside test of InDesign's generator found covers three images. Research benchmarks cover charts, such as Alt4Blind with 5,000 chart images, or check model descriptions of charts automatically, as a 2026 study of 102 visualizations did. Logos, icons and decorative graphics are not covered.
Built with what exists
InDesign 21.2 or later as the generator under test; general AI models that read images, such as Gemini or Claude, for comparison; NVDA to hear the descriptions in exported PDFs; ElevenLabs Scribe v3 to transcribe testers' recorded spoken comments (see the Note "Fewer word errors").
First step
Export 100 images from the team's own layouts, place them in InDesign with automatic alt text switched on, and log each result beside a description written by a person.
Done looks like
At least 350 images, 50 per class, all the team's own or openly licensed; alt text for each from at least 3 generators plus a reference written by a person; at least 5 paid, consenting blind screen-reader users rate each description from 1 to 5 for accuracy and usefulness without knowing its source; a published table, per class and generator, of the share rated accurate and the share of decorative images wrongly described.
Open 03 · the brief
Blind test of palette fixes
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of whether tools that fix low contrast keep a brand's look. Take real brand palettes that fail the WCAG 2 contrast ratio, fix each with several tools and with a human designer, and have working designers pick, without labels, which fix looks most like the original.
Why it is still open
The tools exist, but none has been judged by people on real brand palettes. The cm-colors papers (December 2025) evaluate on "10,000 procedurally generated color pairs" and 10,000 web color pairs using computed color-difference scores, with no human study. Adobe's open-source Leonardo builds color scales to target contrast ratios but does not describe moving an existing palette to the nearest passing colors.
Built with what exists
cm-colors, an open-source Python library; Adobe Leonardo; Contrast-Finder; the WCAG 2 contrast formula; any online survey tool that shuffles the order of choices.
First step
Collect 30 palettes from openly posted brand guidelines, record their color codes, and work out which pairs of text and background colors fall below 4.5 to 1.
Done looks like
30 published brand palettes, each with at least one failing pair of text and background colors; 4 fixed versions of each, all checked at 4.5 to 1 or better for body text; at least 40 designers each judging at least 20 unlabeled, shuffled trials; each method's win rate with 95% confidence intervals and its average color change; all data released.