Skip to main content
How a company’s people speak on podcasts, together, and how that moves over time. It reports the share of their lines on each axis:
  • Sentiment: negative, neutral or positive.
  • Certainty: hedged, neutral or confident.
  • Specificity: vague, neutral or specific.
  • Temporal orientation: about the past, the present or the future.
Each axis is shown for the whole window (the last year by default) and per week or month, with how many of the people spoke. Use it for questions like “how confident do OpenAI’s people sound lately?” or “are Microsoft’s executives talking more about the future?”. The labels are the ones speaker statement characteristics puts on transcript lines. The people are the ones particle_company_get lists with include: ["people"]: those in a current role by default, and those in past roles too with include_past_roles. Shares are of their scored lines: lines they spoke, in any role, that carry characteristics. Lines too short to read carry none. low_sample marks a share with fewer than 20 scored lines behind it. For one person, use particle_person_get_statement_characteristics; every example names its speaker’s person_slug for it. Resolve the company first with particle_company_resolve.

​
Inputs

​
Output

A markdown document headed ## How <company>'s people speak, laid out like particle_person_get_statement_characteristics with one extra row:
  • **People:** says how many of its people said scored lines in the window.
  • The **Window:**, **Scored lines:** and **Characteristics version:** rows follow.
  • ### Overall lists each axis’s label shares.
  • ### By month (or ### By week) is a zero-filled table, with (partial) and (low sample) marks.
  • With include: ["examples"], each bullet under ### Strongest lines, <bucket start> names the speaker: **label** (axis): Name (person_slug …): "text" — Episode on Podcast (episode_slug …) <link>.
With output_format=json, the response is the REST body of GET /v1/companies/{id}/statement-characteristics:
  • company, current_only and people, and the person breakdown’s fields: scored_lines, episodes, low_sample, characteristics_version, characteristics.
  • buckets: each also carries people, and each example a speaker.
Reads are slower for the largest companies, about 3–4 s for one with ~800 people.