Particle Podcast Get Share Of Voice
particle_podcast_get_share_of_voiceSplit a person's, company's or entity's podcast mentions by who said them: the speakers' occupations (SOC 2018) or fields of research (ANZSRC 2020). Answers "who is talking about NVIDIA — investors, engineers, journalists?" in one call: each group's mention lines, share, distinct speakers and episodes over a window (default the last 30 days, at most 365), optionally bucketed by day, week or month. By default it counts guests' and panelists' lines at the detailed level (Chief Executives, Financial Analysts, Dietitians), each with the top-level group it sits under. speakers="all" adds hosts, who speak most mention lines and mostly code to media; level="major" gives SOC major groups or ANZSRC divisions only; podcast_slug or publisher_slug narrows to one show or network.
keyword_search splits the lines that say a word or quoted phrase instead of an entity's mentions (who talks about prior authorization?), counting a line only when its own speaker said every word in it. Its window is at most 90 days, and a word said in too many episodes is refused with a way to narrow it. This is the per-speaker answer that particle_podcast_get_episode_timeseries with keyword_search and guest_occupation cannot give: that counts matching segments in episodes with such a guest, whoever spoke them.
Shares are of the mentions spoken by someone with an expertise read, and a person listed under several codes counts in each, so shares can sum past 100%.
HOW the subject is spoken of splits the same mentions by speaker statement characteristics: by="sentiment" (or certainty, specificity, temporal_orientation) gives one group per label, breakdown splits each occupation's lines by an axis's labels (how investors vs engineers speak of NVIDIA), and speaker_statement_characteristics counts only lines said a certain way (who makes predictions about it). Label shares are of the scored mentions, the lines that carry labels; a group with fewer than 20 is marked low sample, so read it as examples. examples (1-3) quotes the lines that most clearly carry each group's or label's label, with speaker, episode and moment link. Entities and companies only, not keyword_search. To set two to five subjects' label shares side by side (NVIDIA vs AMD vs Intel), use particle_podcast_compare_share_of_voice.
Each group's code is a guest_occupation (or guest_field) value: pass it to particle_podcast_find_mentions with the same entity to read what that group said. For a keyword, particle_podcast_search_transcripts with the same keyword_search finds candidate passages: its guest filters keep segments with such a guest, so check each matching line's speaker, and host lines are not reached that way. Not for the mention lines themselves (particle_podcast_find_mentions) or for a brand's mention volume over time and the shows carrying it (particle_podcast_get_mention_trends).
Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| by | No | Group the speakers by 'occupation' (SOC 2018, default) or 'field' (ANZSRC 2020 fields of research), or group the counted lines by HOW they were said: 'sentiment', 'certainty', 'specificity' or 'temporal_orientation' gives one group per label (how negatively or positively NVIDIA is spoken of is by='sentiment'). Not with keyword_search. | |
| level | No | Group level: 'detailed' (default; detailed occupations or ANZSRC groups, each with the top-level group it sits under) or 'major' (SOC major groups or ANZSRC divisions only). | |
| limit | No | Groups to return, most mentioned first (1-50, default 10). | |
| examples | No | Also quote up to this many lines (0-3, default 0) for each group, or with breakdown for each of its labels: the quote, who said it, the episode and a timestamp linked to the moment. Strongest first: the lines that most clearly carry the label or read the way speaker_statement_characteristics asks; with neither, the newest. At most one per episode, from different speakers where it can. Not with keyword_search. | |
| interval | No | Add each group's mentions per day, week or month. | |
| speakers | No | Whose lines count: 'guests' (default; guests and panelists) or 'all' (hosts included). Hosts speak most mention lines and their own occupations dominate an 'all' split, so pass 'all' only when the hosts' voice is the question. | |
| breakdown | No | Also split each group's lines by their label on this axis: by='occupation' with breakdown='sentiment' is how each occupation spoke of the subject. A group's breakdown shares are of its scored mentions, and a group with fewer than 20 is marked low sample. Must differ from by. Not with keyword_search. | |
| entity_slug | No | Knowledge-graph entity slug from particle_entity_resolve for anything that isn't a person or company (e.g. 'germany'). | |
| include_ads | No | Count mentions inside ad reads too. Off by default: ad copy is not a speaker's voice. | |
| person_slug | No | Person slug from particle_person_resolve (e.g. 'sam-altman'). One of person_slug, company_slug, entity_slug or keyword_search is required. | |
| company_slug | No | Company slug or domain from particle_company_resolve (e.g. 'nvidia'). Resolves to the company's linked entity. | |
| podcast_slug | No | Only mentions inside this podcast (slug from particle_podcast_resolve). | |
| output_format | No | Output serialization. 'markdown' (default) returns the LLM-facing rendering. 'json' returns the structured payload as JSON text — use only for programmatic chaining where exact field extraction matters; the JSON shape is larger and noisier for an LLM to read. | |
| keyword_search | No | Split the lines that say these words instead of an entity's mentions: every word in one line, case-insensitive; wrap words in double quotes to require them adjacent and in order. Words count as spoken, so a word with two meanings counts both. Window at most 90 days. | |
| publisher_slug | No | Only mentions on podcasts of this publisher (network), by slug. | |
| published_after | No | Only episodes published on or after this date (YYYY-MM-DD). Defaults to 30 days before published_before; the window may span at most 365 days for an entity and 90 for a keyword. | |
| published_before | No | Only episodes published on or before this date (YYYY-MM-DD). Defaults to now. | |
| speaker_statement_characteristics | No | Count only mentions said a certain way: {"temporal_orientation": ["future"]} splits who makes predictions about the subject; {"sentiment": ["negative"]} who criticizes it. Labels listed for one axis match any of them; axes combine with AND, so add only the axes the request names. Not with keyword_search. |