Measure what you actually do

Record yourself in your named recurring conversation, transcribe it, and take three counts.

Aim for at least eight minutes of your own speech, which usually means recording a whole meeting. Get consent. Where recording is not permitted, write a contemporaneous account within ten minutes, marking every place you remember hesitating or losing the thread, and note in your record that this sample is weaker than a recording.

Transcribe it yourself, or turn off your tool's cleanup setting, or count fillers by ear against a cleaned transcript. This is the step that most often goes wrong: automatic transcription removes disfluencies by default and will show you a speaker who does not exist.

Then count. Disfluencies per hundred words, splitting fillers, repeats and restarts. Words before your point becomes recoverable, for each substantial contribution, marking the ones where it never does. And structure present or absent, as a binary call, for each contribution.

Takes about three hours: one to record, ninety minutes to transcribe eight minutes of speech, and thirty to count. Budget the transcription honestly, because it is four to eight times real time by hand.

It fails in two ways. Recording a meeting where you barely speak, which produces a sample too small to count. And cleaning the transcript, which removes the thing being measured. If your disfluency rate comes out below 2 per hundred words, check the transcript before you believe it.

Give eight minutes of the recording to your named reader and ask four questions.

Do not tell them what to listen for and do not describe the meeting first. Play it, or send it, and then ask: what was I saying, in one sentence. What did you not understand. Where did you lose the thread. And what did I sound like you were feeling.

Write the answers verbatim before you respond to any of them. Then write your own answers to the same four questions, separately, and compare.

Takes about forty minutes. It fails when the reader knows your subject, because they will reconstruct your meaning from their own knowledge and report that you were clear. Two exercises in this course specifically want somebody outside your field and this is the first.

The comparison is the output. Expect your fourth answer to be harsher than theirs: anxiety moves self-rated performance about twice as far as observer ratings across the interview studies, and you are now measuring your own version of that gap.