Make the assistant’s state understandable
An AI interface often moves between waiting for input, listening, processing and responding. A mascot can make those phases feel distinct, but the visual expression should reflect the actual application state. A friendly face must not imply understanding, certainty or task completion that the system cannot support.
Pair the expression with concise text such as “Listening” or “Preparing a response.” This helps visitors who do not see the character, prefer reduced motion or cannot easily interpret its expressions.
Separate listening from microphone permission
Listening should begin only after the user has enabled the relevant input and the application is receiving it. A mascot must not appear to listen merely because the page is open. Keep microphone controls visible and independent of the character.
If audio-level movement is in scope, normalize and smooth the measured level before passing it to the rig. Audio energy is different from lip sync: a mouth opening with loudness does not reproduce speech sounds. Make that distinction explicit in the proposal.
Give thinking a beginning and an end
A thinking state can communicate that a response is being prepared. It needs an exit when the result arrives, when the user cancels or when the request fails. A stalled request should not leave the character processing forever.
Streaming responses add another decision: does the mascot remain in a thinking pose until a full answer is ready, or transition when output begins? Choose behavior that matches the interface and keep it consistent across short and long responses.
Plan speaking and lip sync separately
Speaking can be a simple expressive state, amplitude-driven mouth motion or a more detailed viseme system that follows speech sounds. These options differ in rig complexity, data requirements and integration work. None is included automatically because a product is described as AI-powered.
Decide how playback pause, interruption and errors affect the character. If the visitor stops audio, the mouth should stop too. If a response is replaced midway, the state machine needs a clean route back to neutral before another speaking sequence begins.
Scope the complete assistant contract
Document the input source for each state and which state has priority. Listening, thinking and speaking should not be ambiguous overlapping flags unless the rig was explicitly designed to combine them. Include idle, unavailable and error states in the plan.
AI states, voice and lip sync belong to the advanced interaction path. Share your assistant architecture, audio provider behavior and desired visual vocabulary before production. The animation remains one part of the interface; it does not supply a model, voice service or microphone backend.