How we counted

Where does the data come from?

• The Podcast Index public database, podcastindex_feeds.db, downloaded from https://public.podcastindex.org/podcastindex_feeds.db.tgz (linked from github.com/Podcastindex-org/database).
• Dump published 26 September 2026 (2026-09-26T23:14:32.000Z). 1,828,580,952 bytes compressed, a 5.1 GB SQLite file once unpacked, 4,731,573 feeds. SHA-256 of the download: 46d781c85f5680fc425a47f1d1ef19cb444a79748a74dd00c0d935b6b57f0477.
• One row per feed with its title, URL, host, generator, language, Apple Podcasts id and categories, episode count, oldest and newest episode dates, the date Podcast Index first saw it, the explicit flag and the newest episode’s file URL. There are no listening figures and no per-episode rows.

Licence and attribution

Podcast Index describes its index as open and “available for free, for any use”, and its database repository, which publishes the download, states that everything in it is under the MIT licence. We credit Podcast Index on every page of the census and on both tools: data from Podcast Index, the open, independent podcast index. We publish no feed URLs or episode data from the dump, only counts, and each show’s own title and numbers in the two tools.

Our aggregates (the numbers on these pages and the downloadable figures) are published under CC BY 4.0. Cite them as “The Podcast Census 2026, Slice (tryslicemedia.com), from Podcast Index data”.

How were duplicates, spam and test feeds cleaned?

In this order. Each step’s count is from the run that produced these pages.

• Duplicates: Podcast Index’s own duplicateOf mark; then feeds sharing an Apple Podcasts id or a podcast:guid, keeping the one with the newest episode (a show that moved host leaves its old feed behind).
• Clones: the same title, the same episode count and the same newest-episode time on different URLs. These are mirrors and template demo feeds; one is kept.
• Placeholders: empty titles, titles that are only digits or punctuation, “(feed disabled)”, and test names such as “Test”, “Untitled” or “Prueba”.
• Dates before 2000 or after the dump are treated as unknown, and those 31,728 shows are left out of date-based figures only.

What do active, finished, dead and video mean here?

• A podcast: a cleaned feed with at least one episode.
• Active (90 days): its newest episode is dated within the 90 days before the dump. Active (12 months): within 365 days.
• Finished: nothing in the 365 days before the dump, and its first episode is before 2026, so a show that started recently is never called finished. Used for the graveyard figures (n = 3,636,474).
• Lifespan: days from the oldest episode to the newest, for finished shows.
• Dead: the dump’s own dead flag is 0 on every row, because Podcast Index removes dead feeds before publishing it, so we do not use it. Separately, 78,864 feeds (1.8%) failed their last check (not a 2xx or 3xx answer); they are still counted, by their last known episodes.
• Video: the newest episode’s file ends in .mp4, .m4v, .mov, .webm or .mkv. The dump has no podcast:medium field and only the newest episode’s file, so this is “publishes video in its feed now”.
• Start year: the year of the oldest episode in the feed. First seen: the year Podcast Index added the feed; 2020 is its bulk import of the existing catalogue, so new-feed figures start in 2021.
• Cadence: (episodes − 1) ÷ days between first and last episode, × 30.
• Likely automated or bulk-published: at least 100 episodes, an average of 4 or more a day, over at least 30 days; or one author name on 100 or more shows first seen by Podcast Index in the same year. Aggregates only; no show is named.
• Language and country: the feed’s language tag (“en-US” is English, United States). Old codes are mapped (“in” to Indonesian). A country is only shown when the tag carries one.
• Host: the domain of the feed URL. Category: the first Apple Podcasts category in the feed.

How do the name checker and the stats checker work?

Both read a static index built from the same dump, split into small files by the first letters of each name. Full rows (title, episodes, first and last episode) cover the 1,025,297 shows active in the past year or with 50+ episodes; the other 3,365,744 names are kept as hashes, so an exact match is still found across all 4,391,041. A name is matched after lower-casing, removing accents and punctuation, and ignoring “the”, “a”, “podcast”, “show” and the same small words in Spanish, French, Italian and Portuguese. Similar names are trigram matches within the same first letters.

Percentile ranks compare a show with a table of 1,001 quantiles per measure, for all 4,391,041 podcasts with dates and for the 413,475 active ones, accurate to about a tenth of a percentile. Nothing is sent to our servers; an Apple Podcasts id is looked up with Apple’s public search API, from your browser.

What can this data not show?

• Listening. A show with one listener and a show with a million look the same here.
• Shows outside RSS. YouTube-only and Spotify-exclusive shows are not in the index; nor are their video files.
• The 1,000-episode window. Podcast Index reads at most a feed’s newest 1,000 items (13,827 feeds are at it). Their counts read “1,000+”, and their start dates are unknown and left out of start-year figures. We confirmed it: The Joe Rogan Experience, The Daily and Stuff You Should Know all show 1,000 against 2,700 to 3,000 in their live feeds.
• Trimmed feeds. Some hosts list only the latest 10, 20 or 100 episodes, so those shows count low and look younger.
• Episode counts are what is in the feed now. A show that deleted its back catalogue looks smaller than it was.
• “Likely automated” is a pattern, not a verdict. It includes legitimate audiobook archives, radio stations and broadcasters, and misses automated shows that publish slowly.
• The dump is a snapshot from 26 September 2026. Shows that started or stopped after it are not reflected.

Can I check the numbers?

Yes. The dump is public and free: download the same file from Podcast Index, check its SHA-256 against the one above, and apply the definitions on this page in the order given. Each cleaning step’s count is in the table, so you can compare as you go. Every figure on the census comes from that one file, and the downloadable figures give each number with its n (4,391,041 shows after cleaning).

We checked a sample by hand too. On 2 October 2026 we read the live feeds of twelve well-known shows (Radiolab, Hidden Brain, 99% Invisible, Crime Junkie, How I Built This, The Daily, The Joe Rogan Experience, Stuff You Should Know, The NPR Politics Podcast, Planet Money, TED Radio Hour and Up First). Five matched the dump to within the one or two episodes published since it, with the same first episode. Four sit at the 1,000-episode window (their live feeds list 1,750 to 2,994), which is why those shows read “1,000+” and are left out of start-year figures. The other three, NPR feeds, keep only their latest 300 to 500 episodes: the same count as the dump, with the window moved on by a week, the trimmed-feed case described above.