How to search podcast transcripts in ChatGPT

Find the moment someone said something on a podcast, who said it and when, without listening back through hours of audio. The Firecrawl plugin lets ChatGPT search podcast transcripts through Firecrawl Alexandria, using Particle's speaker-labelled transcripts.

• Tracking down a quote you half remember, with the show, speaker and timestamp.
• Following what podcasts said about a company, person or topic this week.
• Reading one guest's part of an episode without the rest.

Plugin: the Firecrawl plugin. Opens Firecrawl Alexandria to ChatGPT and charges each Particle call to your Firecrawl account.

Explore Firecrawl Alexandria data providers across categories

Replace the topic with your own and send it to ChatGPT:

@firecrawl Search podcast transcripts through Firecrawl Alexandria for
discussions of nuclear power for data centers from the last 30 days. Show
me the price of each tool before you run it.
 
Give me a short update:
- The five most relevant passages, each with the show, episode title,
  publish date, speaker, their role and the timestamp.
- One sentence on what each speaker argued.
- Which shows came back to the topic more than once.
 
Quote the transcript lines exactly as returned and don't paraphrase them
as quotes.

• Add Firecrawl to ChatGPT. Use the button below to install the plugin from the directory.
• Sign in to a Firecrawl account. Particle calls are billed there. Sign up here if needed.
• Clear Particle's terms once. An admin on your Firecrawl organization accepts them before the first search.

Podcasts are audio first, and most of what ChatGPT can reach is text about the audio. Four problems come from that.

1. The episode text isn't there to read. ChatGPT's search is sometimes shut out of a page by a paywall, a technical fault or robots.txt (OpenAI), and an episode page may carry show notes rather than a transcript. When the text is missing, the answer leans on whatever was written about the episode.
2. Listening is a separate, manual step. ChatGPT can transcribe audio you upload, but only on paid plans and workspaces, with files up to 512 MB and no video files (OpenAI). That works for one episode you already have. It doesn't help you find which episode to upload.
3. Platform transcripts have gaps. Apple doesn't show transcripts for episodes outside its catalog, supports a fixed list of languages, and hides them on explicit episodes when content restrictions are on (Apple). YouTube shows a full transcript only for videos that have captions (YouTube). Searching one platform can miss the show you want.
4. Made-up quotes. ChatGPT can produce fabricated quotes and citations (OpenAI). Speech-to-text has a version of the same problem, and a study of Whisper found roughly 1% of transcriptions contained whole phrases or sentences that weren't in the audio. A quote is only useful with a speaker and a timestamp you can check against the audio.

Searching podcasts from ChatGPT uses these parts:

• Firecrawl Alexandria: a library of structured data that AI agents query without going through web pages. It has over 100 official providers, paid under agreement whenever an agent uses their data, and four indexes Firecrawl builds: web, Research, Developer and Government. Firecrawl compared agents on 845 tasks where only the toolset varied, and the ones with Firecrawl Alexandria produced answers rated 21% higher than those relying on built-in web tools.
• Search by meaning or by words: describe a topic in your own words, or put an exact phrase in quotes, and narrow it by show, date range, language or segment type.
• People and companies: Particle resolves a name first, then finds every line where that person or company came up.
• Whole episodes: once a search finds an episode, ChatGPT can pull its full transcript or one speaker's lines.

Point the prompt at what you're researching. For a company or brand:

@firecrawl Find every podcast mention of Nvidia since Monday, newest first.
For each, give the show, the speaker, their role and the line where it came
up. Leave out ad reads, then tell me which shows mentioned it most.

For a person's interviews:

@firecrawl Find podcast episodes from the last six months where Sam Altman
was a guest. List the show, date and episode title, then pull his lines
about AI regulation from the most recent one, with timestamps.

Say who or what to look for, the time window and the fields you want back, and ChatGPT returns passages you can check rather than summaries you can't.

Search one show for a phrase

@firecrawl Search Pivot transcripts for "data centers" and show the matching
lines with the episode, date, speaker and timestamp.

Read one speaker's lines

@firecrawl Get the transcript of the latest episode of Pivot and show only
Scott Galloway's lines, with timestamps.

Export subtitles

@firecrawl Get the transcript of that episode in SRT format so I can use it
as subtitles.

See the tools before paying

@firecrawl Which tools can search podcast transcripts? List each one and
what it costs per call.

Browsing the catalog and checking prices costs nothing. ChatGPT spends Firecrawl credits only when it runs a tool, and your plan's credits are listed on Firecrawl pricing.

ChatGPT names the tool it ran and the credits spent.

• How to find podcast sponsors in ChatGPT: who advertises on a show, the offer and where the ad sits in the episode.
• Guest appearances from Particle: where a person has appeared as a guest, episode by episode. Particle offers 105 tools in Firecrawl Alexandria, and transcripts are one part of them.
• Earnings call transcripts from Fiscal.ai: what a public company's management said on its quarterly call, with speakers mapped.

• Shows Particle transcribes. Search covers the podcasts in Particle's index, so a small or private feed may not be there.
• Audio links aren't guaranteed. An episode record may include a web page or audio stream, but not every episode has one.
• Quote with care. Transcripts are machine-made, so check an important line against the audio at its timestamp before you publish it.