Guide
What the scores mean, and how to get better ones
Three numbers come back with every conversation. Here is what each one is counting, and the change that tends to move it.
Generative language
How much of your language opens a conversation up rather than closing it down.
Phrases that foreclose an option, assign blame, or state a conclusion as settled pull this down. Phrases that invite a view, name a possibility, or leave room for disagreement push it up. Anything counted against you is highlighted in the transcript, so you can go and look at the sentence instead of arguing with a number.
Try this: Say what you are unsure about out loud. Hedged, curious phrasing scores well because it genuinely leaves the other person somewhere to go.
Listening
Your share of the talking, calculated per speaker rather than across the room.
This is why speaker separation matters. In a four-person meeting, holding 40% of the airtime is very different from holding 40% in a one-on-one, and the score reflects which of those you were in.
Try this: The cheapest improvement is silence after you ask something. Most people fill their own question within two seconds.
Questions
How many real questions you asked, and how much room each left for an answer.
A closed question that can be answered yes or no counts, but counts for less than one that cannot. Statements with a question mark on the end are recognised for what they are, so a meeting that felt like an interrogation and a meeting that felt like a conversation do not score the same.
Try this: Open with “what” or “how”. “Why” tends to read as a challenge even when it is not meant as one.
Getting good input
A few things that change the quality of the analysis
Tag the speakers once
After the first analysis you can label each voice as you, a teammate, or someone external. Anicca remembers the voice, so you only do it once per person. Tagging a teammate credits the conversation to them too, which is what makes a team dashboard mean anything.
Record in stereo where you can
If your source has exactly two channels, Anicca treats left and right as two distinct speakers, which is markedly more accurate than separating voices from a single mixed track. Most conferencing tools will give you a two-channel recording if you ask for one.
Longer is usually better
Files up to four hours are fine. A short clip gives the analysis very little to work with, and ratios calculated over five minutes swing wildly on a single interruption.
Exclude the ones that would skew you
A session you mark as excluded still credits any teammates you tagged, but contributes nothing to your own dashboard. Use it for the conversations where you were deliberately doing all the talking, like a presentation or a briefing.
One thing worth saying plainly
A low score on a hard conversation is not a failure. Delivering bad news badly and delivering it well can look similar from the transcript, and the numbers are there to show you the shape of what happened, not to grade you on it. The useful question is whether the same shape keeps recurring.