Podcasters and 500-episode back catalogues: making them searchable

You make a 500-episode back catalogue searchable by transcribing every episode, indexing the transcript passages with their timestamps, and putting a question-answering layer on top of the index. In SageTube that is one action — import the podcast’s YouTube channel into an Expert — followed by an automated pipeline: each episode is transcribed (captions when they exist, audio transcription when they don’t), split into timestamped passages, and embedded for retrieval. From then on, a question like “what has this show said about zone 2 training?” is answered across all 500 episodes at once, with citations pointing at the exact episode and minute. The alternative is listening: at 90 minutes per episode, 500 episodes is 750 hours — three months of full-time work weeks, for one show.

How big is a podcast back catalogue, really?#

The big interview shows — The Tim Ferriss Show, the Lex Fridman Podcast, Huberman Lab — have each published hundreds of hour-plus episodes, and they keep publishing weekly. A catalogue like that is not an archive in any usable sense; it is a pile. Nobody, including the host, can locate a specific statement in it by memory or by scrubbing a timeline.

We can see the same shape in our own data. As of 2026-07-25, 38 channels in SageTube’s production index hold 100 or more videos, and four of them are podcast-scale — averaging 30+ minutes per episode. The largest of those holds 1,222 episodes averaging 51.6 minutes: about 1,051 hours of talking on a single channel. That is the catalogue-size reality the average-channel numbers in our channel-archiving pillar understate — podcasts sit in the heavy tail.

The consequence of that size is covered at length in the pillar: the host cannot cite their own episode 113, fans re-ask questions that were answered thoroughly in 2022, and YouTube’s search finds titles, not statements. What this post adds is the concrete walkthrough: what indexing a catalogue like that actually involves and costs.

What does it take to index a full catalogue?#

One import, one source. A source in SageTube is a channel, not a video. The plan’s source limit gates how many channels you can import (app/Services/StripeBilling.php:111–119), so a 500-episode podcast is a single source. Before anything is charged, the import screen shows a cost estimate computed from the videos’ total runtime and your current wallet balance (app/Services/ChannelCostEstimator.php:15–42), so the decision is made with the number in front of you.

Two transcription paths, priced separately. Every episode takes the fast path first: if YouTube has usable captions, they are fetched directly — cheap and quick. Only when captions are missing does the pipeline extract the audio track and send it to a speech-to-text API (app/Jobs/ProcessTranscript.php:79–82). The audio path runs AssemblyAI as the primary provider with OpenAI as the automatic fallback (app/Services/TranscribeService.php:52–53). Pricing follows the path taken: $0.01 per minute for captioned episodes, $0.05 per minute for the audio path (config/subscription.php:64–65).

The worked example. 500 episodes × 90 minutes = 45,000 minutes. If every episode has captions, indexing costs $450. If none do, $2,250. In production the real mix is roughly 64% caption-path to 36% audio-path minutes — which is why our public estimates use a blended $0.025/minute rather than the optimistic caption-only rate (config/subscription.php:71–77) — so a realistic figure for a big interview show is about $1,100, once. After that, the catalogue is an index, not a bill: asking a question costs $0.02 per standard query (config/subscription.php:67), and no episode is ever re-transcribed to answer one.

What questions pay off across hundreds of episodes?#

The value of a catalogue-wide index is in questions no single episode answers:

• “What has every guest said about hiring first employees?” — an answer synthesized from a dozen interviews years apart.
• “Which episodes discuss sleep and caffeine timing, and do they agree?” — the cross-episode comparison a fan would never assemble by hand.
• “What was the book recommendation in the episode with the poker player?” — the classic half-remembered moment, found in seconds.

Every part of every answer carries a citation into its source: retrieval works over transcript passages that keep their start timestamp, and the answer cites the episode and the second it draws from (app/Services/ExpertAnswerService.php:561, 577–583). Clicking the citation lands you at that moment in the video — the answer is checkable, not just plausible. We ran the same watch-time-versus-one-question comparison on a lecture catalogue in the Karpathy post; the arithmetic only gets more lopsided at podcast scale.

How does the index keep up with weekly episodes?#

A podcast that publishes weekly would outdate a frozen index immediately. Live Knowledge Pulse attaches a sync schedule — daily, weekly, or monthly — to each channel in an Expert, and the scheduled sweep picks up and indexes whatever was published since the last run (app/Services/PulseSyncService.php:45–53). A weekly sync matched to the show’s release day means Thursday’s episode is answerable by the weekend, with no manual re-import.

Where should a podcaster start?#

Small and free. The Free tier includes one knowledge base, up to three channel sources, and a $5.00 starting balance (config/subscription.php:52–57) — enough to index roughly 500 captioned minutes, say the ten most question-generating episodes as a playlist, and see whether the answers hold up before committing to the full catalogue.

The step after that is the audience-facing one: publish the Expert as a public page. SageTube issues a share token and the catalogue becomes reachable at https://sagetube.ai/s/<token> (routes/web.php:365), where any visitor can ask questions without an account (app/Http/Controllers/SharedExpertController.php:17–19) — the “what was that supplement again?” comment answered by the archive itself, with a timestamp, at any hour. You can browse live public Experts on the Explore page to see the visitor’s side, or start with your own show.

Sources#

• The Tim Ferriss Show — episode archive
• Lex Fridman Podcast — episode list
• Huberman Lab — all episodes

Every product claim in this post is tied to the SageTube codebase as of July 2026: source-equals-channel gating (app/Services/StripeBilling.php:111), the pre-import cost estimate (app/Services/ChannelCostEstimator.php:15), the caption-first fast path and audio fallback (app/Jobs/ProcessTranscript.php:79), the AssemblyAI-primary/OpenAI-fallback transcription providers (app/Services/TranscribeService.php:52), per-minute indexing rates and the blended display rate (config/subscription.php:64–65, :71–77), per-query chat pricing (config/subscription.php:67), timestamped citation passages (app/Services/ExpertAnswerService.php:561), Pulse sync frequencies (app/Services/PulseSyncService.php:45), Free-tier limits (config/subscription.php:52–57), and the public Expert page (routes/web.php:365, app/Http/Controllers/SharedExpertController.php:17). Index statistics — 38 channels with 100+ videos, four podcast-scale channels, and the largest at 1,222 episodes (~1,051 hours, 51.6-minute average) — were queried from the production database on 2026-07-25.