4.7 million podcasts exist. 437,693 published an episode in the last 90 days

The number people quote is 4.7 million. It comes from the Podcast Index, whose public stats page reported a feedCountTotal of 4,728,565 when we read it on 2026-09-16 (stats.podcastindex.org). Listen Notes publishes a smaller one, 3,811,315, read the same day at listennotes.com/podcast-stats, because it excludes feeds with no episodes. Both are counts of feeds. A feed is the RSS file a podcast app reads, one file per show, and the index treats one feed as one show. A hosting platform creates the file when someone signs up, before any episode exists, and the index keeps it after the show stops publishing.

We took the Podcast Index feed dump dated 2026-08-16, 4,711,475 rows, and classified every row into exclusive buckets: a stub, which is a feed whose title matches a sign-up default and holds one episode or none, a feed with no episodes, a feed with exactly one, a feed with two or more and nothing in the last year, a feed with two or more that went quiet inside the year, and a feed with two or more that published in the last 90 days. Nothing was sampled and nothing was dropped.

We expected the live share to be low. Everyone who has looked at podcast data knows the long tail is enormous. We did not expect the dormant bucket to be the largest one in the index: 2,768,200 feeds, 58.75%, that got past a second episode and then published nothing in a year. Two other things surprised us. The sign-up stubs are almost one platform's: 66,914 of the 95,121 stubs sit on Spreaker, and 38,957 of those come from a single title template. Anchor, which hosts 1,906,462 feeds, more than three times Spreaker's 563,601, has no template at all. Its 494,322 feeds holding one episode or none carry titles people typed themselves.

Key findings

1. 437,693 of the 4,711,475 feeds in the Podcast Index dump of 2026-08-16, 9.29%, published an episode in the 90 days before that date. Source: Podcastquery, September 2026.
2. 2,768,200 feeds, 58.75%, hold two episodes or more and published nothing in the 365 days to 2026-08-16. That is the largest bucket in the index.
3. 892,068 feeds, 18.93%, hold exactly one episode. 76,139, 1.62%, hold none.
4. 95,121 feeds, 2.02%, carry a title a hosting platform wrote at sign-up and hold one episode or none.
5. 66,914 of those 95,121 stubs sit on Spreaker, which is 11.87% of Spreaker's 563,601 feeds. Anchor, at 1,906,462 feeds, has 3,340 stubs, 0.18% of its feeds.
6. 34,817 of the 356,901 feeds whose oldest episode falls in 2025 are stubs, 9.76%, and 34,451 of those are on Spreaker. The 2025 cohort is 69,618 feeds larger than the 2024 one; Spreaker's stub count grew by 22,878 across the same two cohorts, which is 32.86% of the growth.
7. One title template, "El podcast de ...", covers 33,563 of the 356,901 feeds that started in 2025, 9.40% of the cohort.
8. 2.63% of the 1,015,023 feeds whose oldest episode falls in 2020 published in the 90 days to 2026-08-16. For the 356,901 feeds that started in 2025 it is 24.88%.
9. Active share varies more by host than by anything else we cut on. On the same measure as the headline, a newest episode in the 90 days to 2026-08-16 over every feed a platform holds: 33.94% of Megaphone's 26,809 feeds, 31.69% of Transistor's 31,262, 5.27% of Anchor's 1,906,462 and 0.76% of Castbox's 51,149. Recount, 2026-09-16.

2,768,200 feeds hold two episodes or more and nothing from the last year

The buckets are tested in order and the first match wins, so they are exclusive and add to 4,711,475.

A stub is a feed whose title matches a sign-up default pattern and holds one episode or none: 95,121 feeds, 2.02%. An empty feed reports zero episodes and is not a stub: 76,139, 1.62%. A one-shot holds exactly one episode and is not a stub: 892,068, 18.93%. A dormant feed holds two episodes or more and has no episode dated inside the 365 days to 2026-08-16: 2,768,200, 58.75%. A fading feed holds two or more and has an episode inside 365 days but outside 90: 256,993, 5.45%. An active feed holds two or more and has an episode inside the 90 days to 2026-08-16: 422,214, 8.96%. The last bucket, unknown (Buzzsprout), holds the 200,740 Buzzsprout feeds, 4.26%, that are neither active nor fading.

Buzzsprout gets its own bucket because the dump carries no episode data for half its rows. 133,259 of its 270,067 rows report zero episodes, 49.34%, against 4.82% across the whole index, and not one of those 133,259 rows carries an episode date of any kind. We fetched four of them on 2026-09-16: one returned 404 and three returned HTTP 200 with no item in the file. Four feeds do not settle 133,259 rows, so we never class a Buzzsprout row as a stub, an empty feed or a one-shot, and we classify it on dates alone. Everything that is neither active nor fading prints as unknown. Buzzsprout rows stay in every denominator, which means every episode-based share above sits on a base that includes 270,067 feeds we cannot place.

The active bucket, 422,214, is smaller than the headline 437,693 because stubs, empty feeds and one-shots are tested first. The difference of 15,479 feeds splits into 11,786 one-shots and 3,693 stubs that published inside the window, and no feed in the empty bucket is active. Recount, 2026-09-16. 437,693 is the plain active-90 count over every row in the dump, and it is the figure on /how-many-podcasts.

The Podcast Index counts 4,728,565 feeds and Listen Notes counts 3,811,315, and both are counting feeds

Each published figure matches a definition we can print.

The Podcast Index figure of 4,728,565 counts every feed it holds, dead ones included, so it matches the first line. It sits 17,090 above our 4,711,475 because our dump is a month older. Listen Notes at 3,811,315 excludes feeds with no episodes, so it sits between "minus stubs and empty feeds" at 4,540,215 and "feeds with two episodes or more" at 3,506,878, nearer the second. Of the published numbers, Listen Notes is the closest to a count of feeds that have published more than once. The 4.1 million that turns up in podcast roundups is older than both: it sits on Buzzsprout's statistics post (buzzsprout.com/blog/podcast-statistics) under the heading "How many podcasts are there in 2022?", so it is a 2022 reading of the index, not a rounding of 4,728,565.

So the number to use depends on the question. If you want feeds that exist, 4,711,475 in this dump. If you want feeds that have published more than once, 3,506,878. If you want feeds a listener could subscribe to and expect something new, 437,693, or 314,204 for a 30-day window. When a story says "4.7 million podcasts" and then talks about what podcasters are doing, the number it wants is 437,693.

2.63% of the 1,015,023 feeds that started in 2020 published in the last 90 days

Start year here is the year of the oldest episode still in the feed, which is not the launch year. A show that trims its back catalogue moves into a later cohort and reads as younger than it is.

The 2020 cohort is the biggest one in the index at 1,015,023 feeds and the least active of the twelve at 2.63%. 687,014 of its feeds are dormant and 267,740 hold exactly one episode. Cohorts from 2015 to 2021 all sit between 2.63% and 4.50% active.

The stub share holds near half a percent from 2015 through 2023, then moves: 4.21% of the 2024 cohort, 9.76% of the 2025 cohort, 5.76% of the 2026 partial cohort. The 2026 row covers seven and a half months, so its 59.97% active share and its zero dormant count are artefacts of its age.

Megaphone has 33.62% of its feeds in the active bucket and Castbox has 0.76%

The platforms that charge per show sit at the top of the active column: Megaphone 33.62%, Transistor 30.50%, Simplecast 29.16%, Captivate 26.13%, Libsyn 24.52%. The free ones sit at the bottom: Anchor 5.16%, Spreaker 6.92%, SoundCloud 4.43%, iVoox 2.83%.

The active column is a bucket share, so it counts only feeds with two episodes or more. On the plain 90-day measure the headline uses, over every feed a platform holds, Megaphone is 33.94%, Transistor 31.69%, Anchor 5.27% and Castbox 0.76%. Recount, 2026-09-16.

Three platforms are mostly one-shots. The University of Cambridge feed domain is 19,827 one-shots out of 21,024 feeds and has no active feed at all, which is what a feed-per-lecture archive looks like. RSS.com has 103,385 one-shots out of its 178,058 feeds, and Anchor has 477,919 out of its 1,906,462. Castbox and Hubhopper are the two with both a high stub share and almost nothing active: Castbox 12.85% stubs and 0.76% active, Hubhopper 4.88% stubs and 0.72% active.

One Spreaker template covers 38,957 of its 130,812 one-or-none feeds, and Anchor has no template at all

We did not guess the sign-up defaults. For each of the 25 platforms with the most feeds, we took every title on a feed holding one episode or none and reduced it to shapes: the whole title when it runs eight words or shorter, its first one to four words with the rest replaced by *, and its last one to three words with the front replaced by *. Then we counted the shapes per platform. The full table, top 15 shapes for each of the 25 platforms, ships as stub_patterns.csv.

We kept a shape only where one platform holds nearly all of its matches, which is what a title written by a form looks like. We threw away shapes that are simply common things for a person to type, however large they got. That rule is what keeps Anchor out. Its most frequent shapes among one-or-none feeds are "the " at 30,048 and " podcast" at 29,125, which is what people write, and no Anchor-specific template shows up at any size. Spreaker reads differently: "el podcast de *" alone covers 38,957 of its 130,812 one-or-none feeds.

The kept list runs to 20 patterns. The largest are el podcast de * (39,222 feeds classed stub), *'s podcast (26,917), podcast de * (5,635), replays for * (2,409, all Castbox), *'s channel / canal / kanal (2,407, mostly Castbox), *'s show (2,192), test* / testing / prueba / teste / prova (1,840), the exact placeholder titles such as "my first podcast" (1,715), *'s first / second / third podcast (1,357, Hubhopper's default), and il podcast di * (1,142). A blank title field accounts for 9,255 more; its largest holders are apple.com at 769 and accessmedia.nz at 600, neither of them a consumer sign-up platform. The regexes are in classifier_patterns.csv and in METHOD.md.

The weakest pattern on the list is podcast de *, the one a Spanish speaker could plausibly type by hand. It accounts for 5,635 stubs, 5.92% of the bucket. Dropping it moves the stub total from 95,121 to 89,486.

32.86% of the 2025 cohort's growth over 2024 is Spreaker stubs

356,901 feeds have a 2025 start. 34,817 of them are stubs, 9.76%, and 34,451 of those stubs, 98.95%, sit on Spreaker. Spreaker's own 2025 cohort is 106,699 feeds, so 32.29% of what Spreaker gained in 2025 is a default-titled feed with one episode or none. Spreaker's stub count by cohort year runs 1,504 in 2023, 11,573 in 2024, 34,451 in 2025 and 10,792 in 2026 to 08-16. It grew by 22,878 feeds between the 2024 and 2025 cohorts, and the whole 2025 cohort grew by 69,618 feeds over the 2024 one, so 32.86% of the year's growth is Spreaker stubs. Spreaker passing Anchor in new feeds is the subject of an earlier report.

The template shows up in October 2024. Spreaker feeds by the year of their oldest episode, against the ones titled "El podcast de", summed from spreaker_el_podcast_de_monthly.csv:

The months around the step, every month of the second half of 2024, from the same CSV:

Over the 33 months from January 2022 to September 2024 Spreaker's monthly feed count ran between 1,699 in August 2023 and 4,033 in March 2022. It went to 10,069 in October 2024 and 12,372 in November. The share of those feeds carrying the template went from 12.47% in September 2024 to 37.04% in October. The full series, all 56 months from 2022-01 to 2026-08, is in spreaker_el_podcast_de_monthly.csv.

Across the whole index, 81,906 feeds carry a title starting "el podcast de" once the title is lowercased and its whitespace collapsed, 80,023 of them on Spreaker. An earlier pass printed 80,015 for the Spreaker share of that figure and we could not reproduce it; the raw host column and the normalised host both give 80,023, so the host aliases are not the cause. Recount, 2026-09-16. By start year the Spreaker ones run 1,839 in 2023, 12,733 in 2024, 33,563 in 2025 and 10,125 in 2026 to 08-16, which is 0.62%, 4.43%, 9.40% and 5.30% of each year's whole cohort.

Who is behind them. Spreaker's 66,914 stubs carry 62,614 distinct itunesAuthor values. 9,748 of the stubs carry a single-word author, and those 9,748 feeds hold 6,521 distinct single-word values. Recount, 2026-09-16. In spreaker_stub_authors.csv the first row is not a name: rank 1 is "(no itunesAuthor)" at 234 feeds, 0.35% of the bucket, and rank 11 is "." at 23. The other 28 rows are first names, none of them on more than 40 feeds: Ana 40, Maria 39, Daniel 32, Carlos 28, Diego 26. Among the 34,451 Spreaker stubs in the 2025 cohort, 76 carry no author, 0.22%. No institution shows up anywhere in the bucket. One school that appeared in our earlier scout note, "Formación Carpe Diem", turns out to have 209 feeds in the index and no stub among them: its feeds are titled after courses, one feed per course.

By language tag, 39,913 of Spreaker's stubs are Spanish, 59.65%, 9,922 English, 14.83%, and 6,044 Portuguese, 9.03%. Language is declared in the feed and we did not check it against the audio.

We fetched five of these feeds live on 2026-09-16 to see what a stub holds. All five are Spreaker feeds from the 2025 cohort titled "El podcast de ..." that hold one episode in the dump. All five returned HTTP 200 and all five held exactly one item, which confirms the dump's episode count.

Two of the five items carry Spreaker's own auto-numbered episode title, "Episodio N - ". The other three carry a title someone typed. So a stub holds one recording under a title the platform wrote, and what that says about the person who signed up is in the Limitations below.

The funnel: 4,711,475 rows in, 4,711,475 classified, 0 dropped

The future-date test is the one tools/count.py uses, a newest timestamp more than one day past 2026-08-16, and no row passes it. Inside the reference day itself, 2,545 rows carry a newest timestamp after midnight on 2026-08-16 and 141 more carry the midnight timestamp exactly. Recount, 2026-09-16.

A feed with no dates, no title or no host still lands in a bucket and stays in every denominator. The cohort table is narrower than the index only because 256,764 rows carry no oldest-episode date and 265,461 more start before 2015.

Two other counts from the same pass, for anyone checking the buckets against raw episode counts: 226,944 feeds report zero episodes, 4.82%, and 1,204,597 report one or none, 25.57%. Only 1,666 of the 226,944 zero-episode feeds carry an oldest-episode date at all, and none of those dates falls after 2020.

Method

Source: the Podcast Index feed dump dated 2026-08-16, table podcasts, 4,711,475 rows, read twice, once to mine title shapes and once to classify. Reference date 2026-08-16 for every window. Host normalisation comes from tools/count.py, so Spotify's creator domains fold into Anchor and the Buzzsprout feed domain folds into Buzzsprout, and every figure here matches the rest of the site.

Start year is the year of oldestItemPubdate, the oldest episode still in the feed. Active90, active365 and active30 use newestItemPubdate inside the 90, 365 or 30 days before 2026-08-16. A feed with no newest date is never active and stays in the denominator.

Buckets are tested in the order stub, empty, one-shot, dormant, fading, active, so they are exclusive and add to 4,711,475. Buzzsprout rows skip the first three tests because the dump carries no episode data for half its rows, and everything Buzzsprout that is neither active nor fading prints as unknown. The 20 title patterns kept as sign-up defaults, with their regexes and the count each matches, are listed in full in METHOD.md and classifier_patterns.csv. Cells under 5 print <5 and a share on a base under 100 is left blank.

The sanity fetches all ran on 2026-09-16 through a rotating proxy. The Podcast Index stats file (stats.podcastindex.org/daily_counts.json) returned feedCountTotal 4,728,565, episodeCountTotal 167,231,926, feedsWithNewEpisodes30days 337,826 and feedsWithNewEpisodes90days 472,076. Their 90-day figure is 34,383 above our 437,693, 7.9% higher, which is what a month of extra crawling against a static dump should produce. The Listen Notes stats page (listennotes.com/podcast-stats) read 3,811,315 podcasts and 193,936,838 episodes. The five Spreaker feeds listed above returned HTTP 200 with one item each, and their URLs are in METHOD.md too. Four Buzzsprout feeds that report zero episodes in the dump were fetched the same way, and their URLs are in METHOD.md. We also looked for Spreaker documentation of the default title and did not find any: support.spreaker.com refuses the proxy, and the live help centre returned nothing about a default show title across four search queries. The nearest documentary evidence is Spreaker's own show page at spreaker.com/podcast/el-podcast-de-spreaker--5234455, titled "El podcast de Spreaker" with an episode called "Episodio 7 - El podcast de Spreaker", the same template on the company's own account.

Limitations:

• A stub is a title, not a verdict about a person. Someone handed a default title who never recorded reads as a stub, and so does someone who recorded once and left the title alone. The live check found both.
• A one-shot may be a finished thing: a lecture, a sermon, an audiobook chapter. The University of Cambridge feed domain is 94.31% one-shots by design, and islamhouse.com behaves the same way.
• Start year is the oldest episode still in the feed, so a show that trims its archive moves into a later cohort. 256,764 rows carry no oldest-episode date and appear in no cohort row.
• Dormant absorbs the 256,537 rows with no newest-episode date.
• Buzzsprout, 270,067 feeds, carries no episode-based bucket, so every episode-based share here sits on a base that includes it.
• episodeCount is capped at 1,000 for recently refreshed feeds. Every cut here is at 0, 1 or 2 episodes, so no figure is affected.
• Language and itunesAuthor are declared by the publisher and unchecked.
• The Podcast Index is not every podcast. It keeps dead feeds, it carries feeds no app lists, and it misses feeds nobody has submitted.

More cuts of the same dump are on /statistics and /how-many-podcasts.

Data

Every file behind this report:

• buckets.csv, 7 rows, every feed in one of the seven buckets.
• buckets_by_cohort.csv, 12 rows, the same buckets by start year 2015 to 2026.
• buckets_by_host.csv, 20 rows, the same buckets for the 20 largest platforms.
• stub_patterns.csv, the top 15 title shapes among one-or-none feeds for each of the 25 largest platforms.
• classifier_patterns.csv, 20 rows, the patterns kept as sign-up defaults with the count each matches.
• cohort_2025_by_host.csv, 20 rows, buckets for feeds that started in 2025, by platform.
• spreaker_el_podcast_de_monthly.csv, Spreaker feeds titled "El podcast de ..." by month, 2022-01 to 2026-08.
• spreaker_stub_authors.csv, 30 rows, the author values behind Spreaker's stubs.
• spreaker_stub_languages.csv, 15 rows, language tags on Spreaker's stubs.
• how_many.csv, 7 rows, the count under each definition with the public figure it matches.
• README.md and METHOD.md.

The dataset is CC BY 4.0. Attribution: Podcastquery, https://podcastquery.com.