This review is part of a larger series of LinkedIn newsletters titled [The Human Side of AI: Cutting through the AI noise to show you how AI can be a powerful tool for your creativity, efficiency, and strategy].
Particle (particle.pro) is a podcast intelligence API. It transcribes and speaker-labels more than 100,000 podcasts, then exposes the results, plus a layer of metadata, through a REST API and a Model Context Protocol server. Worth clearing up first: this is not the Particle consumer news app. Same company, different product. particle.pro is a data platform built for developers and AI agents, and for a researcher it behaves like a queryable archive of everything said out loud on the podcast circuit.
For B2B research, that archive is a secondary-research and media-intelligence source. It answers a specific question well: who said what, on which show, and when.
Key strengths
Speaker-identified transcripts, and the clips to use them
Every indexed episode ships with full diarization, so a result is a labeled transcript, with the host and guests separated and advertisers marked, tied to a real episode and timestamp. New episodes are usually transcribed and enriched within minutes of publication, and coverage spans the entire Apple Podcasts Top 200 across 130+ verticals plus a long tail of independent shows. Each episode is also cut into topical clips (each with its own audio and transcript window, plus an engagement score) and into chapters, and transcripts come as speaker-attributed dialogue, plain text, or SRT subtitles, down to word-level timing. A clip can be embedded on a page through an HTML widget, which is a tidy way to drop a sourced quote into a deliverable or a post.
Four ways to search, and a graph you can walk
Search runs from one place three ways: exact keyword, semantic (natural language), and canonical entity, with a hybrid mode that blends keyword and semantic. Each hit returns a speaker-labeled window ranked by relevance. Entity search resolves nicknames, so “Zuck” finds Mark Zuckerberg, and each match arrives with enough context to tell one “Apple” from another. Underneath sits a knowledge graph where every returned slug (person, company, podcast, episode, publisher) is a valid input to the next call: a company record can return its competitors, products, and people, and a person record links out to external profiles. A researcher can resolve a company, list its people, pull their podcast appearances, then open the episode transcript and the entities discussed inside it. That traversal is what turns a single name into a competitive scan.
A data layer beyond the transcript
Particle structures more than the words. Sponsorship data maps every ad to a canonical brand, with leaderboards, trending sponsors, an advertising timeseries, and a co-occurrence view of which brands run on the same shows, usable for competitive-marketing and share-of-voice reads. Publishers and their shows carry political-bias analysis, which is more useful for narrative and framing studies than for buyer work. Rankings cover the full Top 200 across 130+ verticals, with historical charts and “movers” that flag what is climbing or falling, a fast gauge of where category attention is going. Guest intelligence, meaning rosters per show and appearance histories per person, plus trending guests and interview-format signals, answers where an executive has been showing up and who keeps booking them.
It drops into an agent, and it can watch a beat for you
The part that matters for a researcher who does not write code: Particle’s MCP server connects to Claude Desktop as a custom connector, installs into Cursor or VS Code in one click, and is reachable through Exa’s research agent. Once connected, the questions are plain language (“what did founders say about pricing on tech podcasts last week”), and the agent handles the calls. Alerts add a standing mode: track a keyword, brand, person, or topic and get pinged the moment a match airs, or in a daily or weekly digest, by email, Slack, or webhook, with the clip and transcript attached. That turns a lookup tool into ongoing monitoring, which is the mode most competitive-intel work needs. The raw API is there for teams with engineering support, but the agent path is what makes this usable at a research desk.
Where it fits for a researcher
The clearest fits for Cascade-style work:
- Competitive and executive monitoring. Standing alerts on competitor names, their executives, and category terms, so a podcast mention surfaces the same day with the quote attached.
- Share of voice and narrative framing. Semantic search across the corpus to see how a category or a positioning claim is being talked about, and by whom, over time.
- Message and buyer-language research. Pulling the words buyers, founders, and analysts use about a problem, straight from unscripted conversation rather than a survey box.
- Sourcing and evidence. Grabbing a timestamped, speaker-attributed quote, and an embeddable clip, to anchor a finding in a report or a client readout.
- Media and sponsorship landscape. Ad-intelligence and rankings data to map where attention and advertiser activity concentrate in a client’s category.
Where it could improve
The hard limit is scope: it does podcasts, and only podcasts. It will not analyze your own interview transcripts, design a survey, or synthesize a client’s win-loss set. It is a source to pull from, not a workbench to think in, and it sits alongside NotebookLM or Claude rather than replacing them. The corpus also skews toward founders, executives, analysts, and media voices, so it fits competitive and category work better than the voice of a specific B2B buyer, who is rarely on a podcast in the first place. There is also no point-and-click research dashboard, so the value depends on the agent connection or an engineer. The richest analytics (sponsorship and ad intelligence, plus political-bias views) run on initial credits only at the entry tier and open up on the $399 Business plan, while enterprise firehose access is contact-sales. One last caution: transcripts are machine-generated, so a quote headed into a client deliverable should be checked against the audio, and the entities a researcher tracks do pass to a third party, which is worth weighing on sensitive competitive work.
Ratings
| Dimension | Rating | Rationale |
|---|---|---|
| Usability | 3.5 / 5 | The MCP connector lets a non-coder query in plain language, but with no research GUI, value hinges on the agent path or an engineer. |
| Power | 4.5 / 5 | Within podcast intelligence it goes deep and fast: speaker-ID transcripts, four search modes, knowledge-graph traversal, and sponsor and bias metadata. |
| Flexibility | 3.0 / 5 | Flexible inside its domain (REST, MCP, firehose, several integrations and formats), but that domain is one slice of the research workflow. |
| Cost | 4.0 / 5 | A real free credit and a $29 entry with transparent per-call prices, and the best analytics live on the $399 tier. |
Conclusion
Particle is a narrow tool that does one thing well: it makes the spoken-word media record searchable and attributable, and fresh enough to act on the same week. For competitive and executive monitoring, and for sourcing quotes with a timestamp behind them, it earns a place as the podcast layer of a research stack. The judgment stays with the researcher: which mentions matter, and whether a machine transcript holds up against the audio before it reaches a client. Particle surfaces the quote. Deciding what it’s worth is still the researcher’s job.
Last updated: 7/22/2026.