
...

By Carylyne Chan
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
The AI companion market has largely been built inside the chat box. A user types, a model replies, and the relationship unfolds through words on a screen. For AI companies, chat was the obvious place to start. Text is cheap to generate, easy to personalize, easier to moderate than richer media, and familiar to anyone who has ever sent a message. It also matched the first great leap in generative AI: models that could produce fluent, responsive language at scale.
But to simulate genuine companionship, saying the right words isn’t enough. AI companions are trying to do more than produce the right sentence at the right time. They are trying to communicate care, affection, reassurance, attention, intimacy and presence.
The emotional architecture of a companion product begins with language, but emotional connection also depends on tone, timing, breath, hesitation and the feeling that care is arriving from somewhere beyond the screen.
Read More on Essays
For AI companions to reach their full potential, the category will need to build around the channels that can actually carry feeling.

...
Emotional Communication Needs More Than Words
AI companions are designed around the communication of feelings and attitudes. Unlike traditional AI assistance, they are not mainly explaining facts, summarizing documents or completing administrative tasks. They are trying to make a user feel wanted, understood, reassured or cared for.
That distinction is critical. Psychologist Albert Mehrabian found that when people communicate feelings and attitudes, words sometimes account for as little as 7 percent of perceived meaning, while vocal tone and facial expression carry much more of the emotional signal. When a person gives directions, explains a diagnosis or warns that a building is on fire, their words carry the essential meaning. But when trying to convey an emotion, delivery changes everything.
AI companions are a deeply emotional product. A companion telling a user “I’m here for you” is conveying more than information. The product is asking the user to believe that the statement carries attention, care and presence. Text can express those states, but it cannot fully deliver them. It cannot naturally convey the pause before a difficult sentence, the softness in the delivery, the breath before a user’s name or the restraint that makes reassurance feel sincere.
This is the central limitation of chat-first companionship. The industry has built deeply emotional products on the narrowest emotional channel.
Text Makes User Do Too Much Emotional Work
When a user reads “I’m here for you” on a screen, the sentence is processed through their own internal voice. They supply the pacing, tone, emphasis and emotional color themselves. The product provides the words, but the user completes part of the emotional experience.
That projection can be powerful. People have always formed attachments through written language. Letters, text messages and online relationships can carry real intimacy. A carefully written message can stay with someone for years. But AI companionship asks text to do something more demanding: create the feeling that care is arriving from outside the self.
That is where text often falls short. A user may understand the intended emotion and still feel a gap between the sentence and the experience the product is trying to create. The line reads correctly, but the feeling does not fully arrive.
This is why text-based companions can feel almost real but not quite. The language lands, but the presence still feels artificial.
The gap is particularly noticeable during emotionally charged moments. A user who feels lonely at night wants more than a supportive sentence. They want to hear someone slow down before answering. They may want their name spoken with care. They may want the subtle cue that the companion is not rushing, not reciting, not merely selecting the next plausible phrase. Text can describe attentiveness. Voice can perform it.
Voice Changes Emotional Register
A spoken sentence carries timing, breath, pitch, pace, warmth and restraint. These signals are often the parts of communication that make care feel real. The same words can feel perfunctory, intimate, protective or hollow, depending on how they are spoken.
Consider a companion saying a user’s name before a vulnerable sentence. In text, the name appears as another word in the line. In voice, it can change the entire interaction. A half-second pause before the name can create anticipation. A softer delivery can make the sentence feel private. A slight shift in pace can signal that the moment matters.
Voice also carries the involuntary. A slight roughness, a breath before a difficult word, a lower register when something feels private, a slower cadence when reassurance needs weight. These are the kinds of signals that make communication feel less composed and more alive. Text is always cleanly rendered. Voice can carry texture.
Users can forgive small imperfections in utility software. A calendar assistant can be blunt, and a search tool can be dry. A companion has a different burden. If the emotional delivery feels generic, the relationship weakens. If the voice sounds too polished, too flat or too eager, the user notices.
Voice raises the standard for the underlying writing as well. A hollow phrase that might pass on a screen becomes more obvious when spoken aloud. Overwritten intimacy sounds artificial. Generic reassurance sounds canned. The script has to be precise because voice amplifies whatever is already there. It can deepen a well-written line, but it can also expose a weak one.
For companion products, voice is not a minor UX improvement. It changes the emotional register of the relationship.
Audio Asks User to Be Present
Audio also changes how the user receives the interaction. A text message can be skimmed, reread, ignored or absorbed out of order. The user controls the tempo. However, voice unfolds in time. It asks the user to listen.
That temporal quality is part of what makes voice feel intimate. A spoken message happens in a specific moment. The pause, the breath, the pacing and the emphasis cannot be separated from the sentence. The user receives the line as an event, not only as content.
This is especially important for the moments that matter most in companionship: reassurance, vulnerability, affection, intimacy and repair. A companion apologizing after a difficult exchange, comforting a user after a bad day or expressing affection cannot rely only on semantic accuracy. The user is listening for whether the moment feels attended to. They are listening for care before they analyze the sentence.
Emotional audio has to be designed with pacing, pause architecture, pronunciation, vocal register and context in mind. A cheerful delivery can ruin a vulnerable line. A rushed delivery can make reassurance feel procedural. A user’s name, inserted awkwardly, can feel mechanical rather than intimate.
The small details define the experience.
Achieving Emotional Fidelity
We have already seen that people are willing to form relationships with synthetic characters and conversational agents. Millions of users have shown that companionship does not require a human on the other side of the interaction. The harder question is whether those relationships can feel emotionally credible over time.
Credibility will depend on whether companion products can move beyond fluent language and toward emotional fidelity. That means understanding that the right words are only the foundation. The product must also consider how those words arrive, how they sound, how long they take, what they withhold and what they reveal.
Text will remain important. It is flexible, accessible and useful for many interactions. But the deepest forms of companionship require signals that text cannot naturally provide. They require tone, breath, timing and the sense that the response is being spoken to the user rather than silently completed by the user’s own imagination.
The next stage of AI companionship will not be defined only by larger models or more sophisticated chat histories. It will be defined by whether these products can make emotional communication feel embodied enough to be believed.
Carylyne Chan is the managing partner of BlockSpaceForce, the company behind Murmur, a first-person voice platform for intimate audio fiction.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines