How to search podcast transcripts in Claude

Locate the exact moment a topic came up on a podcast, along with the speaker and the timestamp, instead of scrubbing through hours of audio. After you add the Firecrawl connector, Claude is able to search podcast transcripts in Firecrawl Alexandria, drawing on Particle's transcripts with a named speaker on every line.

• Pinning down a quote you only partly recall, with the show, the person speaking and the time it was said.
• Keeping up with what podcasters said this week about a brand, an executive or a subject.
• Pulling out just one guest's contributions from a long episode.

Connector: the Firecrawl connector. Through it Claude reaches Particle's podcast tools, and the credits for each lookup come from the Firecrawl team you approved.

Explore Firecrawl Alexandria data providers across categories

Swap in the subject you're researching and send the prompt below:

Use Firecrawl to search podcast transcripts in Firecrawl Alexandria for
conversations about nuclear power for data centers over the past 30 days.
Tell me what each tool costs before you call it.
 
Then summarize:
- The five strongest passages, each with show name, episode title, date,
  speaker, speaker role and timestamp.
- A one-sentence take on each speaker's argument.
- Any show that returned to the subject in more than one episode.
 
Reproduce transcript lines word for word as the tool returns them, and
don't present your own paraphrase as a quote.

• Add the connector. Select Connect on the Firecrawl connector listing. A browser tab asks you to log in to Firecrawl (or make a free account), choose the team to bill and grant access. You won't need an API key or any MCP setup, and Firecrawl in Claude covers the details.
• Switch it on per chat. Click + in the message box, choose Connectors and toggle Firecrawl.
• Clear Particle's terms. Particle is among the providers whose terms an organization accepts before the first call. A Firecrawl admin does it once from the dashboard.

Podcasts are spoken, and the web mostly holds descriptions of episodes written after the fact. That leaves Claude with four gaps.

1. Training data ends somewhere. Anthropic lists a knowledge cutoff for each Claude model and warns that models "may not be aware of events or information that occurred after their respective cutoff dates" (Anthropic). Last week's interview isn't something Claude can recall unaided.
2. Web results describe episodes instead of transcribing them. Claude's web search grounds answers in live pages with citations (Anthropic), yet an episode page usually offers show notes, and Anthropic adds that cited links occasionally don't work. The answer then rests on someone's summary of the conversation.
3. Platform transcripts are patchy. Apple doesn't offer transcripts for episodes outside its catalog, covers a limited set of languages, and explicit episodes lose their transcript once content restrictions are switched on (Apple). YouTube only displays a full transcript when a video has captions (YouTube). Relying on one platform can skip the show you need.
4. Quotes can be invented. Anthropic says Claude "can occasionally produce responses that are incorrect or misleading" and that it can display convincing quotes with no factual grounding (Anthropic). Machine transcription has its own version of this: research on Whisper found about 1% of transcriptions included entire phrases or sentences absent from the audio. A quote is worth repeating only when it comes with a speaker and timestamp you can verify.

Four pieces come together when Claude searches podcasts:

• Firecrawl Alexandria: structured data that AI agents query directly rather than scraping pages. Firecrawl licenses more than 100 official providers into it and pays them whenever agents draw on their data, and alongside them sit Firecrawl's web, Research, Developer and Government indexes. Firecrawl ran 845 tasks with blind AI judging, holding the model and prompts fixed, and answers from agents with Firecrawl Alexandria were rated 21% higher for quality than answers from agents relying on built-in web tools.
• Topic or exact words: describe what you're after in ordinary language or quote a precise phrase, then narrow by show, dates, language or segment type.
• Named people and companies: Particle identifies who or what a name refers to, then collects each line where they were discussed.
• Full episodes on demand: after a search turns up an episode, Claude can fetch the entire transcript or just one speaker's part.

Aim the prompt at your research target. To follow a company or brand:

Use Firecrawl to find podcast mentions of Nvidia since Monday, latest
first. For each one, list the show, who spoke, their role and the line
itself. Skip ad reads and tell me which shows brought Nvidia up most.

To follow one person's interviews:

With Firecrawl, find podcast episodes from the past six months that had Sam
Altman as a guest. Give the show, date and episode title for each, then pull
his remarks on AI regulation from the newest one, with timestamps.

Name the person or subject, a time range and the details you want returned, and Claude hands back passages with sources attached instead of loose summaries.

Find a phrase on one show

Use Firecrawl to search Pivot transcripts for "data centers" and list each
matching line with episode, date, speaker and timestamp.

Isolate one speaker

With Firecrawl, get the transcript of the newest Pivot episode and keep
only Scott Galloway's lines, with timestamps.

Get subtitles

Use Firecrawl to fetch that episode's transcript as SRT so I can use it for
subtitles.

Check prices first

Use Firecrawl to list the tools that search podcast transcripts and the
credit price of each call.

Exploring the catalog and looking up prices is free. Credits are only used when Claude runs a tool, and plan allowances appear on Firecrawl pricing.

Each reply ends with the tool Claude used and its credit cost.

• How to find podcast sponsors in Claude: which brands advertise on a show, what they offered and where the ad ran.
• Guest appearances from Particle: each episode where someone appeared as a guest. Transcripts are only part of Particle's 105 tools in Firecrawl Alexandria.
• Fiscal.ai earnings calls: the remarks a listed company's executives made on the quarterly call, with each speaker identified.

• Particle's index sets the scope. Only podcasts Particle transcribes are searchable, so niche or private feeds may be missing.
• Not every episode links to audio. Records can include an episode web page or audio stream, though some lack one.
• Verify before quoting. The transcripts are machine-generated, so listen at the timestamp before publishing an important line.