Learn AI /Responsible AI, Human Rights and Safety /Accessibility
Accessibility
- Starter
- Responsible AI, Human Rights and Safety
- Communications teams
- Human rights practitioners
Also called: a11y, disability inclusion.
Back to the milestone you came from
In five seconds
What it is
Designing tools, content and services so people with different disabilities can use them effectively, safely and with dignity.
In plain words
Make the tool usable by people with different disabilities, not just by the average user.
An everyday example
A town hall in Rome invites residents to comment on a new transport plan. At the entrance, there is a ramp as well as stairs, the agenda is available in large print, the speaker uses a microphone, and questions can be asked by voice or by writing on cards. More people can take part because the organisers changed the setup, not the people. The goal is not to give special treatment; it is to remove barriers that the original setup created.
A meeting room is a physical place, while many accessibility issues in AI are dynamic and depend on software behaviour. For example, an AI system can produce inaccessible outputs, such as poor captions or unlabeled images, even if the basic interface is well designed. The analogy also does not show that accessibility needs ongoing testing with real users, not just one-time setup.
Remember thisAccessibility means removing barriers so people with different disabilities can actually use the system and its outputs, not just be counted as potential users.
Understand it
Understand it
Why it matters
Accessibility affects who can participate, who is excluded and whose needs are overlooked. For a non-specialist, this matters because an AI tool can appear efficient while quietly blocking colleagues, applicants, service users or members of the public. That can lead to weaker evidence, unfair outcomes, legal and human rights risks, and avoidable safety problems. Accessibility also improves decision-making quality: when more people can use a system and check its outputs, errors are more likely to be found. In practice, it shifts the question from 'Can the tool run?' to 'Can people with different abilities actually use it and challenge it?'
Checking accessibility in an AI-supported service
After viewing this, a reader should be able to describe accessibility as a step-by-step design and review process, not a last-minute add-on.
01
1. Identify users and barriers
List who may use the service, including people with different disabilities, and note likely barriers such as unreadable text, audio-only information or time limits.
Draw a box with small icons for text, audio, keyboard and screen reader. Add a caption line: 'Who might be blocked and why?'
02
2. Design multiple ways to use it
Provide more than one way to perceive and operate the service, such as captions, keyboard navigation, clear language and text alternatives for images.
Draw a second box with parallel paths labelled 'see', 'hear', 'type', 'navigate'. Connect from step 1 with an arrow.
03
3. Test with assistive tools and people
Check the service with tools like screen readers and keyboard-only navigation, and if possible with disabled users, because design assumptions often miss real barriers.
Draw a third box with a magnifying glass over a web page and small labels 'screen reader', 'keyboard', 'user test'.
04
4. Fix both interface and AI output
Correct barriers in the layout and in AI-generated results, such as inaccurate captions, confusing summaries or images without useful descriptions.
Draw a fourth box split in two halves: 'interface' and 'AI output', each with a repair icon.
05
5. Monitor and improve
Keep reviewing complaints, errors and changing user needs, because accessibility can break when content, models or workflows change.
Draw a fifth box with a feedback loop arrow returning to step 1 and a small note icon for reports.
A left-to-right sequence of five boxes shows how an AI-supported service becomes more accessible. Read from identifying users and barriers, to designing choices, testing, fixing problems and reviewing outputs in real use. Each step names one action and one result.
On hover or focus, each step reveals one example barrier and one example fix. Keyboard users can move step by step with tab and arrow keys.
Describe this without the diagram
The diagram is an ordered process with five steps. First, identify who may use the service and what might block them, including barriers related to vision, hearing, mobility, cognition or time pressure. Second, design more than one way to access content and controls. Third, test with assistive technologies and, where possible, disabled users. Fourth, fix problems in both the interface and the AI outputs. Fifth, keep monitoring and improving because new content or model changes can create new barriers. The main point is that accessibility is continuous work across design, testing and operation.
With reduced motion: A static horizontal row of five numbered boxes with arrows between them shows the same sequence. The final box includes a dashed arrow back to the first box to show ongoing improvement without animation.
Accessibility means making sure a digital system can be used by people with different disabilities, including people with visual, hearing, mobility, speech, cognitive or psychosocial disabilities. In AI work, this applies both to the system itself and to what it produces. A chatbot, form, dashboard, image description tool or video summary may all need accessible design.
What goes in includes interface choices, content format, language, audio, colour, timing, navigation and any AI features such as speech recognition or image description. What happens is that designers test whether people can perceive the content, operate the controls, understand what is happening and get help when the system makes mistakes. What comes out should be a service more people can use without extra barriers.
Accessibility is not the same as general inclusion, although they overlap. Inclusion is broader and asks who is welcomed and able to participate. Accessibility is more specific: it asks whether a person can actually use the tool or service in practice. It is also not the same as convenience features for everyone, though some accessible design helps many users.
In AI, accessibility also means checking whether the model creates new barriers. For example, speech recognition may work less well for some accents or speech patterns, automatic captions may miss key words, and image generators may produce visuals without meaningful text alternatives. Because of that, human review and testing with disabled users remain important.
How “A public meeting with many ways to participate” maps onto the real thing
| In the story | In the AI system |
|---|---|
| Ramp, large-print agenda and microphone | Accessible design choices such as keyboard access, readable text and captions |
| Questions accepted in more than one way | Multiple input and output options, such as text, speech and screen-reader support |
| Organisers changing the room setup | Developers and organisations being responsible for reducing barriers |
A worked example
A monitoring officer in a country office is preparing an AI-assisted analysis of consultation comments for a monitoring report due for clearance at close of business.
Starting from: Comments include typed responses, voice recordings and scans of handwritten forms, processed only in an approved environment.
- The officer retains the original submissions and checks automated transcriptions against the recordings, since speech recognition can mishear words or omit meaning.
- The review interface is tested with keyboard-only navigation, a screen reader and high magnification. Form fields, headings and controls must have clear labels and a logical order.
- Testing finds that categories are distinguished only by colour and that some AI summaries omit details from brief or poorly transcribed comments.
- The officer adds text labels, corrects contrast and navigation problems, and sends uncertain transcriptions and summaries for human review.
- The monitoring report is circulated in an accessible format, with structured headings, meaningful link text and text alternatives for charts.
Result: The report can be reviewed using different assistive technologies, while the human checks reduce the risk that inaccessible inputs or faulty AI summaries silently remove participants’ concerns.
In a United Nations workplace
In response to an internal audit observation, a gender adviser supporting a UN Country Team working group revises the terms of reference for an AI-assisted consultation portal. Before the management response goes for clearance, the adviser requires keyboard operability, screen-reader compatibility, captions for audio, alternatives to image-based content and accessibility checks on AI-generated summaries and replies.
A technically functional portal can still exclude colleagues, national counterparts or consultation participants. Accessibility requirements must cover both the interface and the AI output, since a usable form does not help if its generated captions, summaries or recommendations are unclear, incomplete or incompatible with assistive technology.
Before you try thisDo not upload identifiable consultation submissions, accommodation requests or information about a person’s disability to a tool that has not been approved for that material. Exposure could create privacy, protection or discrimination risks. Accessibility testing and human review remain necessary even when a vendor claims compliance.
Where you may hear it
- At a vendor demo in a procurement meeting
- During clearance of a public-facing digital consultation
- When an audit observation is reviewed with the front office
- In acceptance testing before a portal is launched
- While preparing materials for a side event
Avoid confusion
Avoid confusion
Often confused with
Inclusion
Inclusion means designing and using AI so different people can access it, be represented in it, and avoid being unfairly left out or harmed.
Fairness
Designing and checking AI systems so they do not produce avoidable, unjust differences in how people or groups are treated or affected.
Transparency
Being open about where an AI system is used, what it does, what information it uses, and who remains responsible for its effects.
Explainability
Explainability means being able to show, in human-usable terms, why an AI system produced a particular output or pattern of outputs.
Common misconceptions
Common misconception: Accessibility only matters for a small group of users.
Correction: Many people benefit from accessible design, including people with permanent disabilities, temporary injuries, low bandwidth, noisy environments, ageing-related changes or limited digital skills. Even when the core purpose is disability access, the practical benefit often reaches much wider.
Common misconception: If an interface meets a checklist, the AI service is accessible.
Correction: Checklists help, but they do not guarantee that the whole service works. An AI system can still create inaccessible outputs, such as poor captions, confusing summaries or unlabeled images. Real testing with assistive technology and users is needed to catch those problems.
Common misconception: Accessibility is the same as inclusion.
Correction: They overlap, but they are not identical. Inclusion asks who is able to participate and feels welcomed. Accessibility focuses more narrowly on whether people can perceive, operate and understand the service in practice. A project may talk about inclusion while still leaving concrete barriers in place.
Quick check
Which example best shows accessibility in an AI-supported service?
The third option removes barriers in the interface and AI output. The first shows transparency, the second fairness, and the fourth explainability. Those practices matter, but they do not by themselves make a service accessible.
Go deeper
Go deeper
The simple version
Make the tool usable by people with different disabilities, not just by the average user.
The technical version
Accessibility in digital systems means designing and maintaining interfaces and outputs so users with disabilities can perceive information, operate controls, understand interactions and complete tasks with appropriate assistive technologies and alternative modes. In AI systems, this covers both the front-end experience and the model-produced content, such as captions, summaries, image descriptions, recommendations or conversational replies. Relevant checks may include keyboard operability, screen-reader compatibility, colour contrast, clear structure, text alternatives, caption quality, timing controls, error recovery and usability testing with disabled people. Accessibility is related to inclusion and non-discrimination, but it is not identical to either: it focuses on practical barriers in use.
Reviewed external resources
Primary sourceWeb Content Accessibility Guidelines (WCAG) 2.2 (opens in a new tab)W3C
This official W3C resource gives practical accessibility requirements for digital content and suits readers who need a recognised technical standard.
Intermediatelink checked 2026-07-31
Primary sourceRecommendation on the Ethics of Artificial Intelligence (opens in a new tab)UNESCO
This UNESCO recommendation places accessibility within AI ethics and human rights and suits readers looking for a policy-level international framework.
Intermediatelink checked 2026-07-31
Primary sourceB-Tech Project (opens in a new tab)UN Human Rights Office (OHCHR)
This OHCHR resource discusses business, technology and human rights responsibilities and suits readers interested in accountability and rights-based practice.
Intermediatelink checked 2026-07-31
EncyclopaediaAccessibility (opens in a new tab)Wikipedia
This Wikipedia page gives a broad overview of accessibility across contexts and suits readers who want a quick background summary.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Accessibility (opens in a new tab)YouTube search
This is a YouTube search results page for accessibility videos, so it suits readers who want to browse different explainers rather than read one source.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Accessibility (opens in a new tab)Reddit search
This is a Reddit search results page for accessibility discussions, so it suits readers who want informal community questions, examples and debates with caution about reliability.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
Accessibility means removing barriers so people with different disabilities can actually use the system and its outputs, not just be counted as potential users.