Test chatbot accessibility and alternative contact routes

Launch control. Start on a normal content page without a mouse. The launch control needs an understandable accessible name and a visible keyboard focus indicator. After opening, users should understand which area is active and how to leave the conversation. A speech-bubble graphic alone does not explain the control's purpose to assistive technology.

Modal or panel. Decide whether the window is genuinely modal. A modal dialog blocks interaction with the background; a supplementary chat panel can leave the page usable. Apply the W3C dialog pattern only when it matches the behaviour. Marking a panel as modal while the keyboard and mouse still reach the background creates contradictory navigation.

Closing and reopening. Then test closing and reopening. Focus should return to a predictable location, and an existing draft should not disappear silently. Include navigation from a product page to the basket: the embedded chat must not obstruct the primary shopping task.

Status messages. New messages are dynamic content. Screen readers need to detect relevant changes without every generated fragment or intermediate status interrupting the current input. The W3C explanation of status messages helps make results and progress available without unnecessary focus changes.

Test these three reply situations in the chat separately:

• the user waits for an answer
• the user continues typing while an answer arrives
• the user reads an earlier part of the conversation

The interface should not unexpectedly jump to the latest message and destroy their position. Longer answers need useful paragraphs and meaningful link labels. A link called “here” is harder to identify outside its sentence than “Return conditions for Austria”.

Sending state. A sending indicator also needs a subsequent state. If transmission stalls, users need to know whether they can retry and whether the original message may already have arrived. A spinner without an explanation leaves that decision unresolved.

A customer uses a keyboard and screen reader. She opens the chat, asks for a return label and follows a link to her customer account. She signs in, selects an item and returns to the service conversation. This fictional case checks more than the widget: it includes authentication, error messages, form fields and confirmation of the submitted request.

The table follows the synthetic return request through six steps, with what to inspect and the expected outcome for each.

If the customer can operate the chat but cannot complete the return form, the task has failed. Assess the whole journey and identify the system responsible for each defect. This also prevents a widget supplier and website team from each assuming the other owns the inaccessible step.

Zoom and narrow screens. Increase text size and browser zoom, then test a narrow screen with its software keyboard open. Messages, close and send controls must remain reachable. Check focus visibility and contrast after applying brand colours; a dark widget theme does not establish accessibility.

Time limits. Time limits deserve a separate test. Security may require signing a user out, but drafts, warnings and reauthentication still need a comprehensible sequence. Someone who reads slowly or uses an alternative input method should not repeatedly lose their progress. Decide which information is deliberately cleared after a session expires and which input can be restored safely.

Long content. Also test messages that wrap onto several lines, long product names and large error text. A polished demonstration using short sentences can conceal controls that become unreachable in normal use.

Germany. Germany's BFSG covers specified products and services, including covered electronic commerce services. A chat forming part of such a purchasing or service process belongs in that process's assessment. Whether a particular offering is covered or exempt cannot be determined merely by the presence of a chatbot. The Federal Accessibility Agency explains electronic commerce services.

Austria and Switzerland. Austria has its own BaFG framework for specified products and services, applicable since 28 June 2025. The Austrian ministry explains its scope and exceptions. A German assessment therefore does not replace an Austrian one. Nor should EU legislation automatically be described as Swiss national law. Establish the operator, target market and service separately. Sources checked: 26 September 2026.

WCAG 2.2. Technically, WCAG 2.2 provides a structured assessment framework around perceivable, operable, understandable and robust content. Passing one checklist does not establish complete WCAG conformance or satisfaction of every applicable legal obligation.

Reproducible reports. Record browser, device, assistive technology, task, expected behaviour and observed result. “Screen reader does not work” is too vague. “The reply is not announced after submission, and focus remains in an unnamed input” identifies a fixable defect. Avoid real customer information in screenshots and issue reports.

Priorities and retests. Prioritise barriers to completing the main task over cosmetic inconsistencies. Retest each correction in the original journey and involve people who routinely use assistive technology. Automated tools can detect issues such as missing names or contrast failures; they cannot decide whether a conversation is understandable and practically usable.

Acceptance. Acceptance should establish an available service route, documented unresolved findings and an owner for future widget updates. An alternative phone number can help, but it does not automatically replace accessibility required of the digital service itself. Accessibility is an ongoing property of the complete customer-service process.

These pages cover the service overview and the related decisions for an accessible customer chatbot:

• AI chatbots for customer service
• Add a website chatbot with a defined service boundary
• Customer-service AI chatbots: plan the human handover
• Launch and operate a chatbot with clear release criteria

For a workflow review, start with the task, systems involved and typical exceptions. Discuss it with modernagents.