You know which colleague always circles back to budget no matter what the meeting's about. You know whose “I'll try” means yes and whose “I'll try” means no. You know the friend who hedges every plan until the last possible moment. This isn't data you collected — it's a pattern your brain assembled from hearing them, many times, over a long while.
Around a fire, you came to know each person's voice not just by its sound but by its habits — the stories they always told, the way they made a promise, the subjects they kept returning to. Speaker Themes is that knowledge, surfaced from the conversations themselves: how each person in your circle actually talks.
What a person's themes reveal.
Speaker Themes reads across the many conversations you've had with a person and surfaces the patterns in how they communicate. Not the content of any single chat — the habits that only become visible across a lot of them. A person's themes gather a few specific things.
The topics they return to. What this person keeps bringing the conversation back to — the subject that's clearly on their mind, the concern that recurs whatever the official agenda. Often it's the thing they care about most, and seeing it named is genuinely useful.
How they hedge. The way they qualify, soften, or leave themselves an out. Whether they commit cleanly or wrap every agreement in conditions. Once you can see someone's hedging pattern, you read their “maybe” much more accurately.
The rhythm of how they commit. How this person makes a promise — fast and firm, slow and cautious, enthusiastic-then-quiet. The shape of their commitments across many conversations tells you a lot about how to read the next one.
None of this comes from one recording. It's the accumulation — the pattern that emerges only when you've heard someone across many conversations, which is exactly the view Bonfiyah has and your memory only approximates.
Why it takes many conversations.
A single conversation can't tell you how someone tends to talk — it can only show you one instance. Habits, by definition, only appear across repetition. That's why Speaker Themes is a feature a per-meeting tool structurally can't offer: if speaker identity dies with each recording, there's no way to gather a person's pattern across sessions, because there's no continuous “person” to gather it onto.
Bonfiyah can do it because the foundation is already there. The same person is recognized across your whole library by voice, so everything they say across every conversation accumulates onto one continuous identity. Speaker Themes is, in a sense, a reading of that accumulation: point the AI layer at everything one person has said over time, and the patterns in how they talk fall out. It's the same underlying memory that powers People Memory, asked a slightly different question — not “what's true about this person?” but “how does this person communicate?”
How it relates to People Memory.
People Memory and Speaker Themes are close kin, and the distinction is worth drawing. People Memory holds the substance of a person — their role, the projects they touch, the threads open between you, how their commitments tend to land. Speaker Themes is specifically about communication style — the recurring topics, the hedging, the rhythm of how they speak and commit.
Think of it this way: People Memory tells you who someone is and where things stand with them; Speaker Themes tells you how they talk. Both are read from the same accumulated picture, and they complement each other. People Memory is the dossier; Speaker Themes is the ear you'd develop for someone after knowing them a long time, made explicit. You'd want both for anyone who matters to your work or your life, and they're deliberately distinct so each stays sharp.
Where it earns its place.
Reading a negotiation. Knowing how the person across the table hedges, and how they tend to commit, changes how you interpret what they say in the moment. Their “let me think about it” means something specific, and the pattern tells you what.
Managing people well. The themes a report keeps returning to are often the things they most want addressed and least want to say outright. Seeing the recurring topic surfaced is a quiet way to attend to what's actually on someone's mind.
The people you're closest to. With a partner, a family member, an old friend, the patterns are the relationship. Seeing how someone you love tends to talk — what they return to, how they commit — isn't surveillance; it's the attentiveness of really listening, made durable.
What it isn't.
Speaker Themes isn't a personality test and it isn't a verdict on anyone. It doesn't diagnose people, score them, or tell you what they're “really like” underneath. It surfaces patterns in how someone talks, across conversations you actually had with them — observed habits, not judgments of character.
It also isn't built from anything but your own recordings, behind the same consent tooling that ships in every tier; it's a reading of conversations the person agreed to be part of, not a profile assembled from anywhere else. And it's a feature about understanding people more attentively, not about catching them out. The point is the kind of listening you'd do for someone you cared to understand — just held steady across more conversations than memory can. It's an ear for how each person talks. Honest about what it is, and nothing more.
Try it on the people you talk to most.
Because themes only emerge across repetition, the way to feel this is to record several conversations with the same few people over a stretch of time and let the Pro AI layer run. Then open Speaker Themes for one of them. By the time the pattern is built, you'll likely see something you sort of knew about how they talk — named, specific, and a little sharper than the intuition you were carrying.
Speaker Themes is one of the eighteen surfaces in the Pro AI suite, and it's free to try. You always learned a person by their habits, not just their voice. Speaker Themes hears those habits across every conversation, so the way each person talks becomes something you can actually see.
— Richard
