Skip to main content
Given one show, which other shows are like it? Particle API answers with a ranked, explainable list computed from what is actually in the audio and the metadata around it: the content of recent episodes (embedded transcripts), the show’s topic profile, the guests it books, the network it belongs to, and the advertisers it carries. Every result carries a calibrated score, a coarse band, and — on request — the specific signals that make the pair related, so a client can tell “same beat, same audience” from “same network, different subject” instead of trusting a bare number. Related sets are precomputed per show and refreshed as new episodes arrive, so the endpoint is a fast page read rather than a live search.
Response (truncated)
The podcast detail embeds the lean form on request, so one call answers “what else is like this” without a second round trip. The same include parameter also takes recommended_guests (the five guests the show could book next) and recommended_sponsors (premium-grade; the five advertisers active on shows like this one that do not buy it yet), the inline forms of /recommendations/guests and /recommendations/sponsors:
Response (truncated)
related is omitted when the show’s set has not been computed yet. The full list, pagination, and the basis live on GET /v1/podcasts/{id}/related.

​
Reading a result

Related shows are restricted to the source show’s language, and a show’s own list never contains itself. Only shows above a relatedness floor are listed, so a show with too few embedded episodes gets a short list or an empty one rather than fifty near-zero rows. Machine-generated and farmed feeds (synthetic voice and script, text read by a synthetic voice, syndication spam, brand impersonation) are never listed as related shows, and when a publisher carries the same show under several feeds — an “all shows” aggregate, a regional edition — only the best-ranked one appears. Sets recomputed since this rule shipped are stored already collapsed; for sets computed earlier the collapse happens when the list is read and is best-effort, so during a vector-store outage such a list can briefly show a republished feed. An empty first page returns 200 with a coverage note, not 404 — the show exists; the note says why nothing is listed.

​
MCP

The same data is available to agents as particle_podcast_list_related (with include: ["basis"]), and inline on particle_podcast_resolve via include: ["related"].

​
Choosing the right endpoint

​
How it is computed

For each show, candidates come from five independent arms — nearest shows by content centroid, shows sharing a guest, shows concentrated in its most specific topics, the show’s own publisher, and shows sharing a sponsor — and every candidate is scored on all signals at once, so a show with no embedded episodes still gets neighbours from its guests, topics and network. Candidates below the relatedness floor are not stored, screened feeds are dropped before scoring, and a publisher’s duplicate feeds of one show are collapsed before the top fifty are kept. Sets are recomputed when new episodes land, coalesced so a daily show recomputes a few times a day rather than once per episode. The ranker was validated against an external listener-behaviour reference (Apple’s “You Might Also Like” lists), a temporal shared-guest holdout, and a calibrated editorial judge; the measured recall and precision figures are in the changelog.