Conversational interfaces
A conversational interface is a back-and-forth with a model, by text or by voice. It is the most common shape for AI products right now, from chatbots and assistants to support widgets. Underneath the friendly surface it is a turn-taking loop, and each turn has to work for everyone.
Key points
Section titled “Key points”- A conversational interface is a turn-taking loop, and composing, sending, waiting, and receiving each have to work for everyone.
- Give the prompt field, send button, and secondary controls clear accessible names, since unlabeled controls are a common failure.
- Give each message a heading that names who said it, and keep landmarks to the page’s main regions rather than wrapping every message.
- Announce a finished reply once through a polite live region without dragging the user’s focus into it.
- Show the assistant’s state in text, not only with an animation, and do not let a long wait go silent.
- Tone, pacing, and explanation style carry cultural assumptions, so treat cultural fit as a build decision, not a late localisation layer.
The turn-taking loop
Section titled “The turn-taking loop”In one turn the user composes a message, sends it, waits, and then receives a reply. Accessibility means none of those steps breaks for a keyboard or screen reader user. The waiting and receiving steps are where conversational AI tends to fail, because the reply arrives on its own and often streams in.
Label the basics
Section titled “Label the basics”The prompt field and the send button need clear accessible names. This sounds obvious, and it is still missed. An audit of ChatGPT’s own interface found that the text area and the send button had no accessible label, so a screen reader user could not tell what they were for. Give every control in the composer a real name, and give the same care to secondary controls such as stop, regenerate, attach, and the rating buttons.
Make the message history navigable
Section titled “Make the message history navigable”A transcript is a series of turns, so give it a structure a screen reader user can move through. Headings do this best, because they are where most people start. In WebAIM’s screen reader survey, 71.6 percent of respondents said navigating by headings is their first approach to finding information on a long page.
Give each message a heading that names who said it. Eric Bailey recommends wrapping each message in an article with a heading such as “You said” or “Assistant said”, using the assistant’s real name. A sighted user can often tell the speakers apart by which side of the screen a message sits on. The heading gives everyone else that same cue. A visible heading helps the most people, so use one if the design has room. If it does not, hide the heading visually and keep it in the accessibility tree.
Set the heading levels to fit the page. Bailey uses h1 for the conversation title and h2 for each message. Headings inside a reply then need to start at h3, so adjust what the model returns rather than trusting its levels.
Do not turn the whole transcript into a single live region, because that makes every change announce at once. Skip the ARIA feed pattern too, even though it is also built from articles. It is designed for content that loads as you scroll, and Bailey warns that its focus management and keyboard expectations work against people who rely on the keyboard. Headings are easier to understand and far less work to maintain.
How you announce a new reply needs thought rather than a reflex reach for ARIA. As James Scholes points out, ARIA is for exposing information, not for managing behavior, and roles like log behave inconsistently across screen readers. Decide on purpose what a user should hear when a reply arrives, and build that, rather than hoping a role does it for you.
Keep the page shell simple
Section titled “Keep the page shell simple”This section is for chat that fills the page. A small pop-up widget has less structure to manage, although the rest of this page still applies to it.
A full-page chat needs only a few landmarks. Use nav for the list of past conversations, main for the messages, a named form for the composer, and aside for a panel of code or generated content. Do not add a landmark to each message. Bailey’s comparison of VoiceOver’s landmark list for a long session shows 3 entries without message landmarks and 23 with them. A list that long stops being a shortcut. That matters because many people rely on it. In the same WebAIM survey, 31.7 percent of respondents said they always or often use landmarks when a page has them.
Put a skip link first in the page and point it at the conversation title, so a keyboard user does not have to tab through every past conversation to reach the messages. Bailey also suggests skip links to the composer and to the profile and settings area.
A shortcut that moves focus between these regions helps in a long session. Bailey describes this as an older technique from Windows desktop apps, built around F6. Browsers already use F6 themselves, so Bailey suggests Control+F6 for the web. Browser shortcuts vary, so test whichever key you choose across the browsers and screen readers you support. Tell people the shortcut exists, and let them change it, as WCAG Operable describes.
Announce the reply without stealing focus
Section titled “Announce the reply without stealing focus”When the assistant replies, a screen reader user needs to know, but their focus should not be pulled into the new message while they are still typing or reading. The common approach is to wait until the reply has finished, announce it once through a polite live region, and leave focus where the user put it. This holds when the reply streams in token by token too, as Dynamic and streaming output explains.
Some content does not work when it is read out of a live region. A table’s cells sound odd without the structure around them, and a long code block is slow to sit through. Bailey suggests announcing a short description in its place, such as “settings.json, 738 lines” for code, or a table’s subject with its number of rows and columns. The content itself stays in the transcript with its real semantics, so the announcement is a preview of what is there. Bailey also warns that processing strings for live regions is brittle and takes more upkeep than teams expect, so take this on only if you can keep it working.
Show the assistant’s state
Section titled “Show the assistant’s state”People need to know when the assistant is working, when it has finished, and when it has stopped. Convey this in text through a status message, not only with an animated indicator that a screen reader will not read. A short status such as “Generating response” tells everyone where things stand.
When to announce it depends on what is on screen. If the interface shows loading text, announce the same text at the same moment. If it shows only an animation, wait a little, so a quick reply does not need a status at all. Bailey reports that about five seconds after the prompt is sent “seems to be the sweet spot”, based on inclusive user research Bailey evaluated. The data is not published, so treat five seconds as a starting value to test with your own users.
If a reply can take a while, one status update is not enough. Repeat it about every five seconds until the reply is ready, and announce the reply the moment it finishes rather than at the next cycle. Ask Mike Anything, a chat assistant built by designer Mike Mai, rotates through a few distinct phrases such as “Still processing your request…” and “Almost there…” every few seconds while it waits on the model. Each phrase is different text in the same live region, so it re-announces cleanly without repeating the exact words a screen reader user might assume had not changed. Keep the phrases true. Do not say the assistant is researching if it is not looking anything up.
Voice-first conversations
Section titled “Voice-first conversations”When the conversation is spoken, the same loop applies in a different modality. Give a visible and screen-reader-friendly transcript, let users read at their own pace, and do not assume everyone can hear the reply or speak the prompt. A voice interface still needs a keyboard and text path.
Audience: Engineer
Build the composer from real form controls with real labels, and give each message an article and a heading rather than putting the transcript in a live region. For new replies, use one pre-existing polite live region at the end of the page, wait until the reply has finished, and announce it once. Read Scott O’Hara on live regions and James Scholes on ARIA and experiential design before wiring it up. Keep focus in the composer when a reply arrives, and give a keyboard shortcut to jump to the latest reply for users who want it.
A minimal accessible shape looks like this.
<!-- First thing in the page --><a href="#conversation-title">Skip to conversation</a>
<nav aria-label="Conversations"><!-- past conversations --></nav>
<!-- One heading per message, and no landmark per message --><main class="transcript"> <h1 id="conversation-title">Live regions explained</h1> <article> <h2 tabindex="-1">You said</h2> <p>What is a live region?</p> </article> <article> <h2 tabindex="-1" data-last-message>Assistant said</h2> <div><!-- reply with real headings and lists, starting at h3 --></div> </article></main>
<!-- Composer with real controls and real labels --><form aria-label="New message"> <label for="prompt">Your message</label> <textarea id="prompt"></textarea> <button type="submit">Send</button> <button type="button" hidden>Stop generating</button></form>
<!-- One pre-existing polite region for status and new-reply text --><p class="visually-hidden" role="status"></p>A form only counts as a landmark once it has a name, which is why it has an aria-label here. The tabindex="-1" on each heading lets a shortcut move focus there without adding a tab stop. Point a jump to the latest reply at the heading marked data-last-message, and point a jump to the composer at the textarea rather than the form. Bailey’s version also repeats the heading text in an aria-label, as a workaround for screen reader bugs when a heading takes focus. Test whether the screen readers you support need that before you add it.
Use the region for the waiting status, then for the finished reply, so each one announces once. This sketch covers a reply with only an animation on screen. It uses the announce function from Live regions and announcements.
const phrases = ['Thinking', 'Still working on it', 'Still generating your reply'];let waitTimer;
function onSubmit() { let turn = 0; // Silent for a quick reply, then a new phrase every five seconds waitTimer = setInterval(() => { announce(phrases[turn++ % phrases.length]); }, 5000);}
function onReplyFinished(text) { clearInterval(waitTimer); // The reply never waits for the next cycle announce(text);}Pass a short description instead of the full text if the reply is a table or a long code block. Moving through phrases in order means the region never repeats the string it already holds. A plain element with role="status" is the simplest, reliably announced region, which is why it is used here rather than an output element. Live regions and announcements covers when to reach for each. Define the visually-hidden utility the usual way so the region is present for assistive technology but kept off screen.
.visually-hidden { position: absolute; width: 1px; height: 1px; margin: -1px; clip-path: inset(50%); overflow: hidden; white-space: nowrap;}Audience: Designer
Design the anatomy of a turn so each message is a clear, labeled unit with a visible author and a status. Specify the busy, done, and stopped states as real content, not just motion, so they survive into the accessibility tree. GitHub Primer’s Copilot accessibility principles are a good reference, especially the points about being transparent and predictable and about attributing authorship.
A simple way to make this buildable is to annotate each assistant state with the exact text it should announce, so nothing depends on the visual treatment alone.
| State | What it announces |
|---|---|
| Idle | Nothing |
| Waiting, with loading text on screen | The same text, at the same moment |
| Waiting, with only an animation | Nothing at first, then a short status about every five seconds |
| Done | The reply, or a short description if it is a table or long code block |
| Stopped | Response stopped |
| Error | Something went wrong, try again |
Keep a visible done state as well, such as Response ready, so sighted users get the same clear signal that the wait is over.
Audience: Self-advocate
With a screen reader, a chat interface can feel like talking into a void if the reply is not announced, or like being shouted at if every word is announced as it streams. The best ones tell you the assistant is working, let you know when the reply is ready, and let you read it at your own pace without dragging your focus around. If a tool does this badly, naming the screen reader, the browser, and what you expected to hear makes the report far more useful.
Audience: Accessibility Specialist
Test the conversation as a flow, not a snapshot. Walk a whole turn with the keyboard and a screen reader, including sending, waiting, and acting on the reply. Because the reply is generated, test several different ones, including a long answer, an error, and a refusal, since they often render through different code paths. Confirm the new-reply announcement is present and not overwhelming, and that focus stays put. Sit through a long wait to confirm the status repeats and stops as soon as the reply is announced. Compare the landmark list on a long session with a short one, because the two should match. If the product describes tables or code, check each description against the real content, because that wording comes from parsing generated output.
Cultural context
Section titled “Cultural context”A conversational interface carries cultural assumptions in its defaults. Most AI chatbots reflect Western mental models for tone, directness, and explanation style. Those defaults shape whether the product works for a given user long before anyone writes a line of adaptive code.
Research on the Gov.sg chatbot (N = 304) found a clear split in what users needed from the same interface. Users with high-context communication preferences responded primarily to social presence, the sense that the chatbot felt human and relationship-aware. Low-context users responded primarily to speed, accuracy, and task completion. Same interface, different things making it work.
How a chatbot explains itself matters as much as what it explains. Research across four linked studies found that high-context cultures consistently responded better to metaphorical or story-based explanations, while low-context cultures preferred step-by-step analytical formats. These are not preferences at the margins. They are consistent patterns.
Hofstede’s cultural dimensions give concrete interface decisions to make rather than theory to keep at a distance.
- Power distance shapes whether the assistant should sound authoritative and expert-led or more peer-like and collaborative.
- Individualism vs. collectivism shapes how to frame outcomes. When a product asks users to do something for others, collectivist framing often performs better regardless of the user’s own orientation, because it matches the behavioral logic of the action.
- Uncertainty avoidance shapes how much structure, confirmation, and reassurance to offer. Users less comfortable with ambiguity want predictable, structured responses. Too much hand-holding is patronising to others.
- Long-term orientation shapes whether to frame advice as urgent action or future planning. Urgency lands differently depending on cultural context.
The stakes change how much this matters. Cultural misalignment in a customer service chatbot is a friction problem. In a mental health support tool it can cause harm. A chatbot that confronts or corrects when a user is already vulnerable does not feel like support. It feels socially inappropriate. Research comparing everyday mental wellbeing chatbots across Sweden and Sri Lanka found that users from a culture with collectivist norms and a strong face-saving orientation significantly preferred chatbots that avoid challenging user views and give structured, predictable responses. Those needs do not reduce to a single Hofstede score. Cultural orientation and face culture interact in ways that require judgment rather than a formula.
Culture is a build decision. Tone, explanation style, pacing, and how the chatbot handles disagreement need to be embedded in the interaction logic from the start, not added as a localisation layer at the end.
Further reading
Section titled “Further reading”- Eric Bailey on making immersive LLM chat experiences more accessible.
- Intopia on labeling the ChatGPT interface.
- James Scholes on ARIA and experiential design.
- Ask Mike Anything, a live example of a single settled-reply announcement and rotating wait status.
- GitHub Primer’s Copilot accessibility principles.
- Luan Haoyue and Cho on user perceptions and context culture orientation in chatbot engagement.
- Naidoo and Chadha on culturally responsive AI chatbots, with the CRAIF-C framework tested across four studies.
- Arambewela on cultural context and trust in everyday mental health chatbots.