GET /v1/podcasts/mentions returns every line of dialogue across the catalog where a resolved entity is named, grouped by episode and ordered by recency. Each item in data is one episode plus all of its mention windows. A mention window is a contiguous range of dialogue around a name, with is_mention flagged on the lines that actually contain it.
This is the right endpoint when the answer doesn’t depend on the wording of your query — only on who is being named.
Available to MCP agents as
particle_podcast_find_mentions.- Every line said about a company this week, newest first, with who said it and in which segment.
- What one show has been saying about one person over the past year.
- Where a person was named while appearing as a guest, as opposed to being discussed in their absence.
When to use Mentions vs Search
Examples
Response
data. windows[] is ordered by time within the episode. mention_count is the number of lines that triggered is_mention=true. mention_variants is the distinct strings observed for the entity (e.g., ["Sam Altman", "Altman"]) — useful for UI labels.
If an episode mentions the entity more than 50 times, the page surfaces the first 50 windows and sets truncated: true on the episode. The next-page cursor still advances to a different episode — windows never split across pages.
Filters
Empty results
An emptydata array means no transcribed episode matching your filters names the entity; it is not an error. Three things to check, in order: that the slug is the one GET /v1/entities/search?q= returns for the name, and that the result’s mentions.podcast_episodes is non-zero (a slug that resolves to nothing returns an empty 200 page, not an error; a wrong-but-valid one returns someone else’s mentions; a count of zero means no episode carries an entity tag for the subject, so the only lines left to find are the verbatim name matches this endpoint returns, marked matched_by: "name", for a person or company without a linked knowledge-graph entity); whether role, podcast_id, publisher_id, or the date window is doing the narrowing, by dropping them one at a time; and whether the name only comes up in ad reads, which include_ads=true brings back. For “how often, over time” rather than “which lines”, use GET /v1/podcasts/mentions/timeseries, which returns every bucket in the window in one response, up to 1,000 buckets, so pick an interval that fits the window.
Pagination
Standardlimit (1–100, default 25) + opaque cursor. The unit of pagination is the episode — a single episode never splits across pages, so a page may return fewer total dialogue lines than another with the same limit. Pass the cursor from the previous response back as ?cursor=… for the next page.
Related
- Episode search — find dialogue by topic or phrase.
- Transcripts → mentions in one episode — every entity mentioned in one episode (the inverse view).
- Podcast → mentions rollup — episode-level salience aggregated across one podcast.