Who said what, to whom, and what it changed.
A transcript records what was said. The Pulp conversational model follows what each message did next: who it reached, who it moved, and what changed.
How it reads a conversation
A conversation is a graph.
In our model, a conversation is a run of turns between people and the software acting for them. Every turn travels on a contract: who may speak, about what, within which limits, under whose moderation. Each reply answers something earlier. It backs it up, pushes back, or changes the subject. The model keeps the whole shape, from who answered whom to how far each move shifted the people listening. Its vocabulary has a name, the Pulp conversation ontology, and every term in it is defined on persuasion.wiki.
The words
Text, audio and video, kept in order: text by position, audio by time, video by frame. Claims, who and what gets mentioned, structure and fallacies are counted the same way every time.
The moves
How well each point was made, read on eight channels, from whether you trust the source to whether the timing is right. Every score comes with its reasons.
The people
Who is listening: who they are, what moves them, their state of mind and where they are. Each persona remembers, so the next message meets the person the last one left behind. The chief of staff at 8am and at 7pm are, persuasively speaking, two different people.
What it tells you
Four ways a message can go.
One message, one person, one moment, and four chances that always add up to one. They take it in and it holds up. They take it in and it doesn't. They take it in and reserve judgment. Or it never reaches them at all.
Measured against your goal
Every score is computed against a goal someone stated, so a number always answers a question: better at what, and for whom.
Scores with reasons
Each score carries the reasons it was given, so you can argue with it, check it, or change your mind about it.
Try one yourself
Pick a person and a place, say something, and watch what goes through their head.
Open the interactive demoHow we check it
Checked against itself, and against people.
Read twice, blind
In our poll data story, a second reading of 161 comment and question pairs, done blind, agreed with the first on 153. Every disagreement was taking a side against taking none. Not once did one side flip to the other.
Read the data storyReal people, every wave
The personas are checked against real people every wave, so a persona can't drift away from the people it stands for.
A range with every number
Every estimate comes with its interval, so you can see how sure we are before you act on it.
It learns from what happened
We're building a graph neural network on the shape of the conversation. It compares its readings with what actually happened next, and corrects itself.
Where you can use it
In our products, on your server, and in the open.
Thinkwell
Our workstation for people whose work runs on persuasion. It measured the poll data story in about 45 minutes.
Meet ThinkwellThe Pulp API
The same readings on your own server: Radiance scores a turn, TrueAura models who is listening, and Windfall rehearses the other side's next turn before you commit to yours.
Meet the APIpersuasion.wiki
Every term the model measures, defined in plain words, with how each one connects to the others.
Open persuasion.wikiHave a conversation you want read?
Send us a thread, a transcript or a debate. We'll tell you plainly what the model can see in it.