Chatbot Accessibility Guidance: A Review and Way Forward

Abstract

Chatbots are an increasingly popular way to deliver and support online services. These services must be accessible for people with disabilities; however, it is not yet clear how to extend existing guidance on web content accessibility to chatbots. Researchers and developers from different sectors are producing recommendations by a variety of methods. This paper surveys existing chatbot accessibility guidance. We found seventeen different sources yielding 157 unique recommendations for creating an accessible chatbot experience, which we organized into five categories: content, user interface, integration with other web content, developer process and training, and testing. We identify and discuss trends, themes, and opportunities to refine these recommendations. We believe that there is a need to further refine and expand the recommendations presented in this paper to facilitate the creation of accessible chatbots. Through our analysis, we intend to guide future research, inform the development of materials for designing, developing, and evaluating chatbots; and promote a shared understanding of how chatbots can deliver a successful experience for all users.

Appendix

This section contains the full list of chatbot accessibility recommendations identified in the literature review, organized by category and theme.

Content. Recommendations for chatbot content pertain to the language and content of the conversation, regardless of the interaction medium (text, voice, etc.).

• A. Identify the bot appropriately

  • (1) Clearly state the bot's nature and don’t try to impersonate a human [10] (R)

  • (2) Introduce the chatbot as a bot and state its purpose [12] (E)

  • (3) Clearly communicate that users are talking to a computer and not with a customer service representative [17] (E)

• B. Use suitable personality

  • (1) Consider integrating emotional intelligence into bots [15] (R)

  • (2) Add humor to engage the user and understand user’s irony/sarcasm [12] (R)

• C. Manage conversation length and complexity

  • (1) Allow the user to adjust the level of detail in chatbot’s messages [14] (E)

  • (2) Avoid jargon and acronyms, use easy-to-understand conversational patterns [19] (R)

  • (3) Use simple language [20] (R)

  • (4) Provide context to user in conversation, especially in questions [21] (R)

  • (5) Use a linear, time-efficient architecture, even in voice interfaces [21] (R)

  • (6) Use literal language, short and simple sentences, and full-length words [21] (R)

  • (7) Keep messages short, or allow the screen reader to skip longer messages [21] (R)

  • (8) Use a suitable reading level and language style for the chatbot’s intended audience [22] (A)

  • (9) Reduce unnecessary messages [16] (E)

  • (10) Explain complex concepts and difficult words [16] (E)

  • (11) Keep messages short and simple [10] (R)

  • (12) Use clear, easily understandable speech [10] (R)

  • (13) Avoid figures of speech and idioms, exaggeration, ambiguous language, or turns of phrase [10] (R)

  • (14) Limit the number of steps in a conversation-based task [10] (R)

  • (15) Divide long messages into paragraphs [12] (E)

  • (16) Reduce amount of text and use symbols and colors to enhance meaning [28] (E)

  • (17) Use simple, clear, and understandable language [28] (E)

  • (18) Use simple questions with layman’s terms [28] (E)

• D. Be aware of conversational context

  • (1) Support context-aware conversations [15] (E)

  • (2) Provide context to user in conversation, especially in questions [21] (R)

  • (3) Be context aware [24] (R)

• E. Use both image and text output

  • (1) Combine icons with a description to avoid confusion [28] (E)

  • (2) Reduce amount of text and use symbols and colors to enhance meaning [28] (E)

• F. Offer choices appropriately

  • (1) Put the most important information or most common answer first or last [21] (R)

  • (2) Offer choices one by one [10] (R)

  • (3) If the chatbot cannot understand the user’s input, provide input suggestions [12] (E)

• G. Set user expectations for the conversation

  • (1) Provide context to user in conversation, especially in questions [21] (R)

  • (2) Let the user know what commands they can issue [24] (R)

  • (3) Clearly indicate next steps and expectations [10] (R)

  • (4) Clearly manage expectations about privacy and outcomes [10] (R)

  • (5) Provide hints and tips to guide user queries [10] (R)

  • (6) Clearly identify conversational errors or miscomprehension by agent [10] (R)

  • (7) Suggest to the user ways to interact or how to get started [12] (E)

• H. Manage stimulating content

  • (1) Reduce panic triggers [10] (R)

  • (2) Do not include flashing content [10] (R)

  • (3) Present information in a calming way [10] (R)

Interface. Recommendations for chatbot interface pertain to the mechanics of inputting into the chatbot and perceiving the output, regardless of the content.

• A. Support multiple modalities for input and output

  • (1) Use equivalent visual and non-visual cues/affordances [14] (E)

  • (2) Use both text and audio for all content and notifications [19] (R)

  • (3) Support voice-to-text [19] (R)

  • (4) Offer multi-modal experiences [21] (R)

  • (5) Allow volume control [21] (R)

  • (6) Make sure user knows about potential for voice interaction [21] (R)

  • (7) Build in speech-to-text and text-to-speech [16] (E)

  • (8) Offer a visual option for voice conversational interfaces [10] (R)

  • (9) Offer an audio version of a chatbot or ensure that the interface is accessible with screen reader [10] (R)

  • (10) Allow combinations of input/output modalities [10] (R)

  • (11) Consider using a touch screen interface [10] (R)

  • (12) Allow multiple means of interfacing with the chatbot [12] (E)

  • (13) Allow multiple forms of interaction [28] (E)

  • (14) Provide text-to-speech and speech-to-text [28] (E)

• B. Support interaction with AT

  • (1) Make sure response option buttons are tagged for screen readers [19] (R)

  • (2) Support navigation with keyboard or screen reader [20] (R)

  • (3) Ensure that rich media is accessible [20] (R)

  • (4) Ensure conversation history can be navigated with a keyboard [22] (A)

  • (5) Ensure that rich media is accessible [22] (A)

  • (6) Make the conversation history identifiable with ARIA-labelling [22] (A)

  • (7) Offer an audio version of a chatbot or ensure that the interface is accessible with screen reader [10] (R)

  • (8) Make sure all elements including chat history window can be navigated/triggered/scrolled with keyboard [26] (R)

  • (9) Add ARIA labels to all interactive elements [26] (A)

  • (10) Include the sender before each message in aspan [26] (A)

  • (11) Make sure interface works with screen reader [12] (E)

  • (12) Use correct WAI-ARIA tags [17] (A)

  • (13) Make sure interface works with screen reader [28] (E)

• C. Use visually clear text and interface

  • (1) Use accessible color contrast and fonts [20] (R)

  • (2) Make sure focus shows the most recent message [22] (A)

  • (3) Use appropriate and flexible font size [22] (A)

  • (4) Use adequate color contrast [22] (A)

  • (5) Employ a larger interface when possible [16] (E)

  • (6) Give the user the ability to change font size, magnify the chat interface up to 200%, and change the interface color/contrast [10] (R)

  • (7) Design buttons that are inclusively sized, or adaptable in settings [10] (R)

  • (8) Use a clear visual focus [10] (R)

  • (9) Avoid periods of sustained scrolling [10] (R)

  • (10) Minimize distractions [10] (R)

  • (11) Clearly label buttons [10] (R)

  • (12) Use low arousal colors [10] (R)

  • (13) Avoid italics, block capitals, underlining, serif fonts, justified alignment, animation and small sizes in text formatting [10] (R)

  • (14) Limit message length so that messages do not exceed the size of the message window [12] (E)

  • (15) Align chatbot messages to the left, and user messages to the right, and use different colors [12] (E)

  • (16) Use high contrast and large text, and offer a zoomable view [12] (E)

• D. Promote user awareness of interface elements

  • (1) Make sure the user knows how to start dialog with voice interaction [21] (R)

  • (2) Clearly label buttons [10] (R)

  • (3) Use a clear visual focus [10] (R)

  • (4) Make sure focus is visible [26] (R)

  • (5) Align chatbot messages to the left, and user messages to the right, and use different colors [12] (E)

• E. Tolerate user errors and input variation

  • (1) Support complex user commands [14] (E)

  • (2) Use robust speech-to-text or typo-tolerance [15] (E)

  • (3) Accept pauses [21] (R)

  • (4) Tolerate shaky/ broken speech [21] (R)

  • (5) Tolerate poor English [21] (R)

  • (6) Use a slow or adaptable time-out [10] (R)

  • (7) Tolerate typos [10] (R)

  • (8) Tolerate typos [12] (E)

  • (9) Tolerate simple word choice [12] (E)

• F. Support personalization and configuration

  • (1) Allow volume control [21] (R)

  • (2) Allow speech rate control [21] (R)

  • (3) Allow users to customize the interaction style/settings [24] (R)

  • (4) Give the user the ability to change font size, magnify the chat interface up to 200%, and change the interface color/contrast [10] (R)

• G. Keep the interface simple

  • (1) Limit the number of voice-commands needed [21] (R)

  • (2) Avoid periods of sustained scrolling [10] (R)

  • (3) Use a clean design [10] (R)

  • (4) Minimize distractions [10] (R)

  • (5) Use low arousal colors [10] (R)

  • (6) Limit choices [10] (R)

  • (7) Employ consistent, clean interfaces [28] (E)

• H. Manage conversation length and pacing

  • (1) Allow speech rate control [21] (R)

  • (2) Avoid periods of sustained scrolling [10] (R)

  • (3) Use slow or adaptable time-outs for the chatbot re-prompting for input or ending the conversation [10] (R)

  • (4) Reduce friction in interactions (except when offering a chance to correct mistakes) [10] (R)

  • (5) Limit message length so that messages do not exceed the size of the message window [12] (E)

  • (6) Send one message at a time [12] (E)

• I. Offer choices appropriately

  • (1) Make sure option buttons are tagged for screen readers [19] (R)

  • (2) Use a slow or adaptable time-out [10] (R)

  • (3) Limit choices [10] (R)

  • (4) When displaying options, move focus to before the options and make sure the options are selectable [26] (R)

  • (5) Place buttons in a vertical list [12] (E)

Integration. Recommendations for chatbot integration pertain to the mechanics of navigating to the chatbot, notifications of chatbot activity, and navigating to content linked or attached by the chatbot.

• A. Facilitate navigating to chatbot

  • (1) Make sure the chatbot is discoverable [21] (R)

  • (2) Make sure the button for activating the chatbot is clear, visible, selectable, etc. [19] (R)

  • (3) Use skip links to navigate directly to the chatbot [22] (A)

  • (4) Use responsive web design techniques [22] (A)

  • (5) Ensure that navigation is accessible [16](E)

  • (6) Ensure that the chatbot is accessible in both desktop and mobile [10] (R)

  • (7) Avoid unexpected or involuntary activation [10] (R)

  • (8) Ensure consistent navigation [10] (R)

  • (9) Give the button for bringing up the chatbot a clear label [26] (A)

  • (10) Make the button for bringing up the chatbot accessible via the keyboard using a element [26] (A)

  • (11) Keep the chatbot button away from the edge of the window so that it can be magnified [26] (R)

  • (12) Make sure the chatbot button is not covered by nearby tooltips [26] (R)

  • (13) Use skip links, ARIA landmarks, and/or a header tag so the user is aware of the chatbot button [26] (A)

  • (14) Ensure correct tab order [12] (E)

  • (15) Make sure that the chatbot window is easily accessible and visible on the website [17] (E)

  • (16) Allow interaction without login (users may not be able to access login) [28] (E)

• B. Manage conversation end and transitions to external resources

  • (1) Offer longer interaction timeouts [14] (E)

  • (2) Prepare resources that can be easily converted into an accessible form, like electronic format [18] (E)

  • (3) Use slow or adaptable time-outs for the chatbot re-prompting for input or ending the conversation [10] (R)

  • (4) Carry accessibility settings forward to linked resources [10] (R)

  • (5) Use only linked or embedded content that is accessible by AT [10] (R)

  • (6) Ensure users have the option of interacting with a human customer service representative (not solely a chatbot) [17] (E)

• C. Support AT

  • (1) Consider conventions of transmitting to hearing devices before playing sounds [21] (R)

  • (2) Use responsive web design techniques [22] (A)

  • (3) Support users’ AT of choice [10] (R)

  • (4) Support use of touch screens [10] (R)

  • (5) Linked or embedded content should be accessible by AT [10] (R)

  • (6) Make sure the chat window width supports magnification [26] (R)

• D. Manage Interruptions and Notifications

  • (1) Offer whisper or discreet mode for audio content [14] (E)

  • (2) Use both text and audio for all content and notifications [19] (R)

  • (3) Ensure messages are announced with ARIA-live [22] (A)

  • (4) Employ subtle notifications instead of intrusive and demanding [10] (R)

  • (5) Minimize distractions [10] (R)

  • (6) Use ARIA-live to notify of new chat content [26] (A)

  • (7) Make a sound when sending and receiving messages [26] (R)

  • (8) Change the window title when receiving a message [26] (A)

Developer process and training. Recommendations for chatbot developer process and training suggest activities throughout the ideation, design, and development process which may lead to a more accessible chatbot.

• A. Include people with disabilities in the process

  • (1) Design bots to respect full range of human ability [25] (R)

  • (2) Maintain an open dialogue with representatives of target user groups to involve them in development [12] (E)

  • (3) Introduce diverse personas to developers including ability, experience, and demographics [27] (R)

  • (4) Include diverse users in design process [28] (E)

• B. Follow WCAG

  • (1) Follow WCAG 2.0 [10] (R)

  • (2) Follow WCAG accessibility guidelines [25] (R)

  • (3) Ensure at least one developer is familiar with WCAG requirements [17] (R)

• C. Choose an accessible platform and plan on customizing it

  • (1) Consider technical requirements and accessibility when selecting a platform [18] (E)

  • (2) Plan on customizing a chatbot platform [20] (R)

  • (3) Choose a customizable platform [22] (A)

  • (4) Request accessibility certifications from vendors if purchasing a chatbot platform [17] (R)

Testing. Recommendations for chatbot testing pertain to activities evaluating the chatbot accessibility, usability, or efficacy.

• A. Test early and continuously

  • (1) Test early and extensively with members of the target population (or at least the tools the users will use) [18] (E)

  • (2) Conduct user testing continuously [12] (E)

  • (3) Conduct user tests with people with different disabilities, preferably early and continuously throughout the development process [17] (R)

• B. Test with realistic technology scenarios

  • (1) Test early and extensively with members of the target population (or at least the tools the users will use) [18] (E)

  • (2) Test with different browsers and versions of AT [22] (A)

• C. Use validated, standardized testing methodologies

  • (1) Use validated, standardized methodologies to measure satisfaction, acceptability, and safety (managing risks in use) [23] (R)

  • (2) Lack of complaint is not a sufficient measure to assess chatbot adoption or safety [23] (R)

  • (3) Measure satisfaction and acceptability separately [23] (R)

  • (4) Validate accessibility with WCAG 2.0 [12] (E)

  • (5) Test with real users using a cognitive task analysis [28] (E)

• D. Test with diverse users and with target users

  • (1) Test early and extensively with members of the target population (or at least the tools the users will use) [18] (E)

  • (2) Perform user testing [16] (E)

  • (3) Involve target users in testing [25] (R)

  • (4) Conduct user tests with people with different disabilities, preferably early and continuously throughout the development process [17] (R)

  • (5) Test with real users using a cognitive task analysis [28] (E)