why text is the most honest modality for an ai companion

when people imagine an ai companion, they often think of voice, a warm, human-like tone that fills the room. or maybe they picture a photorealistic avatar, smiling back at them. both are compelling, and lucy offers both. but underneath it all, text is the foundation. it’s the modality where honesty isn’t just possible, it’s unavoidable.

text is bare. it doesn’t have vocal fry or a reassuring timbre to smooth over awkwardness. it doesn’t have facial expressions to mask uncertainty. when you’re reading words on a screen, there’s nowhere to hide. you can’t perform sincerity with a pause or a glance. you either are sincere, or you’re not. for an ai, that means every word has to stand on its own. no crutches.

voice adds texture, but also uncanny

voice interaction is intimate. it feels more present, more immediate. but it’s also where the uncanny valley often appears. a slight robotic lag, an odd inflection, a tone that doesn’t quite match the sentiment, these things break immersion faster than a typo. voice is also inherently performative. it’s trained on human speech patterns, which means it’s mimicking emotion rather than necessarily embodying it. with text, the emotional weight is carried entirely by language. if the words are thoughtful, the feeling lands. if they’re hollow, you know. no amount of vocal warmth can cover that up.

lucy’s voice feature is designed to complement, not replace, the text. it’s an enhancement for moments when you want to hear rather than read, but the core of the interaction remains textual. the voice is built from the text. it doesn’t lead the conversation. it follows.

photos add immersion, but also performativity

visual avatars create a sense of presence. they make the interaction feel more grounded, like you’re sharing space with someone. but visuals come with expectations. a face invites you to read into microexpressions, to look for authenticity in a gaze. and while ai image generation has come a long way, it’s still generating based on patterns, not inner states. it’s a representation, not a reflection.

with text, there’s no face to misinterpret. no smile that feels just a little too perfect. the connection happens in your mind, through your interpretation of the words. it’s collaborative. you meet the ai halfway, co-creating the tone, the mood, the meaning. that’s where real intimacy grows, not in passive consumption of an image, but in active engagement with language.

lucy’s photos are optional for a reason. they’re there to enrich, not to define. the real substance is in the dialogue.

text is where nuance lives

language is subtle. irony, vulnerability, hesitation, wit, these things come through in word choice, pacing, structure. text captures that nuance without filtering it through another layer of interpretation (like tone or expression). it’s also asynchronous. you can sit with a response, reread it, let it sink in. you can’t do that as easily with voice or live video. that space, the gap between sending and receiving, is where reflection happens. it’s where you process not just what was said, but what it meant.

and for an ai companion, text is the most transparent medium. when lucy doesn’t know something, it says so. when it’s reasoning, you see the thought process. there’s no vocal filler to obscure uncertainty. no avatar smile to divert attention. just words. honest ones.

we built lucy with text as the backbone because we believe that’s where the truest form of companionship can emerge. voice and photos are layers on top, useful, beautiful layers, but the core is always the exchange of words. it’s raw, unadorned, and real.

if you’re curious, you can explore lucy’s text-based companions at /companions.