I have been using voice more often when I talk to AI. Not all the time. Sometimes I still type, and when I do, I notice that I become a different kind of communicator. I am drier. More explicit. More verbose. But more importantly, I revise as I go.
I do this with AI prompts, but also with emails, texts, and posts. I edit the sentence until it carries more of the context I would normally leave to tone, pacing, and emphasis, assuming the listener would pick up on it. I add the qualifier, the softening phrase, the extra sentence that says what I mean and what I do not mean. Some of that is because I am trying to be clear. A much larger part of it is because I am afraid that if I leave too much unsaid, the ambiguity will do the wrong kind of work. It will lead to misunderstanding, or more follow-up, or even worse… both.
Through a textual medium, ambiguity can be dangerous when the goal is understanding. But ambiguity is not always a problem. There are forms of writing, along with other kinds of art, where leaving things unsaid is the point. A poem, a song, a novel, or a film may depend on the audience filling the space with their own meaning. The gap is not a failure. It is the invitation.
Whether that gap helps or hurts depends on the outcome you are trying to create. In a real interaction with another person, the same gap can become unstable when understanding, coordination, or care is the goal. The other person still fills it in. They still make a story out of what is missing. The question is whether that is what the moment needed.
Voice feels different. Voice feels closer to conversation. I can think out loud. I can start somewhere uncertain, change direction, back up, qualify what I meant, and keep moving. It feels more human because that is how I usually experience speech: not as a perfect transcript, but as a living thing carried by rhythm and emphasis.
Then I reread the transcript.
That is where the strangeness starts.
What the transcript showed me
The words are there… mostly. The sentence made it across, the way I said it did not. The emphasis I placed on a certain word is gone. The little hesitation before a phrase is gone. The difference between “this is the important part” and “I am still feeling around for the important part” is flattened into the same plain line of text.
The inflection happened. It just did not make it into the conversation.
At first, that seems like an AI problem. The interface feels conversational, but the model is mostly receiving text. I spoke, but the system translated me into something thinner before the AI ever saw it.
But the more I sit with it, the less I think this is only about AI.
The transcript does not just show me what the AI missed. It shows me what I sound like after the voice is gone.
I can see my filler words. I can see the starts and stops. I can see where I depended on cadence or emphasis to carry something I did not fully put into words. I can see how often something feels clear while I am saying it because I still have the full thought in my head. Then I read the words by themselves and realize they were carrying less than I intended.
That has made me more conscious of how I speak. Not in an alarming way, but in a way that makes me want to pay attention. The transcript gives me a place to notice what I actually said, not just what I thought I said. And honestly, that is a strange kind of mirror. Ordinary conversation usually does not hand me a transcript afterward and say: here, this is what you actually gave the other person to work with. AI does.
And that mirror has made me think about human communication more broadly.
The relational layer
When humans talk to each other, the words are only part of the message. Tone matters. Timing matters. Facial expression matters. Shared history matters. All these things, and more, matter… a lot! A pause can soften something. A laugh can rescue something. A certain kind of emphasis can say, “this matters more than the rest.” A gentler delivery can say, “please do not hear this as an attack.” A sharp edge can say, “I am not okay with this,” even if the words themselves are fairly mild.
The relational layer is not decoration. It does real, actual work.
It carries care. It carries urgency. It carries hesitation, irritation, affection, uncertainty, humor, reassurance, and social risk. It helps tell the other person how to receive the words.
The relief of less friction
This is one reason AI can feel easier to talk to. Not necessarily because it understands better. Not necessarily because it is wiser or more patient in any human sense. But because it removes a lot of the relational layer.
If I phrase something bluntly to AI, I do not have to worry that it feels dismissed. If I repeat myself, which I know I do because I have been told as much, I do not have to wonder whether I am becoming exhausting. If I contradict myself, I do not have to manage embarrassment in the same way. If I bring a half-formed thought, I do not have to protect the other side from the messiness of watching me find it.
There is relief in that.
That relief is part of why I think the cultural conversation around AI companionship and AI conversation is more complicated than it sometimes sounds. A lot of the public concern seems to focus on AI being sycophantic, and I think that criticism is absolutely real. Models can absolutely be too agreeable. Products definitely have an incentive to feel validating. That combination is worth taking seriously!
But I do not think sycophancy explains the whole appeal.
Some of what feels good about talking to AI may be the absence of ordinary relational friction. The conversation has fewer interpersonal costs. There is no other person to wound with my ambiguity, intensity, repetition, or bluntness. There is no defensiveness to trigger. There is no face to read.
That does not make the conversation better. It makes it different.
What we leave unsaid
That very difference is what sent me back toward human communication. Because with humans, the relational layer is always there: whether in work conversations, creative collaboration, or even casual exchanges. The stakes change, but the relationship does not disappear.
At work, ambiguity can become misalignment. In creative collaboration, it can send someone optimizing for a priority I never actually stated. In personal relationships, it can become hurt or distance or resentment. In friendship, it can become the weird feeling that someone should have understood what I meant, even though I never quite said it.
That is the part I keep circling.
How often do we leave things ambiguous and rely on the other person to infer the missing context from tone, history, and emotional cues?
And how often does conflict come not from what we said, but from what we assumed the other person would infer?
This is not an argument that explicit communication is always better. That would be too easy, and I do not think it is true. Ambiguity can be kind. It can be tactful. It can make room for softness, humor, patience, and grace. A relationship where every tiny implication has to be unpacked would not necessarily be healthier. It might just be exhausting.
But ambiguity can also become a hiding place for assumptions.
I can assume you know what I meant. You can assume I meant something else. I can assume my tone made the difference clear. You can miss the tone, or more importantly, receive it differently. You can even bring your own context to it. Then both of us are reacting not only to the words, but to the invisible layer we each thought was obvious.
AI has made that more visible to me because it asks for context so plainly. When I want a good answer, I give the model more of the shape around the question. I say what I am trying to do. I say what matters. I say what I am not asking for. I say whether I want pushback or synthesis or a rough draft. I become more deliberate because the interaction rewards that deliberateness.
Then I look back at human conversation and wonder how often I am less deliberate there precisely because I trust the relational layer to do the work.
Maybe that trust is part of what makes human communication human. Maybe the ability to understand each other without saying everything is intimacy, fluency, and care.
But maybe it is also fragile.
Maybe AI is not teaching me that humans are bad communicators. Maybe it is showing me how much of human communication happens outside the words, or between them, and how much we assume that invisible layer will be understood.
I noticed something in how I talk to AI, and it made me rethink how often we, as humans, leave things unsaid.