The State of Podcasting, August 2026

How many podcasts exist, when they were actually made, and how few of them are still publishing — counted from one weekly Snapshot of the Podcast Index.

Pinned to2026-08-17

frozen

Published 2026-08-23 · v1, no errataEvery figure below is the 2026-08-17 Vintage and will not change. Corrections arrive as errata, never as edits.

What we found

4,711,475feeds the index currently carries — the whole set, not a sample.

5.5%of every feed ever listed published in the last 90 days. The rest are dormant or dead.

2020remains podcasting's peak year for new shows — and 2025 is the first year since to turn back up.

The index carries 4,711,475 feeds #

That is every RSS feed Podcast Index still listed on 2026-08-17. It is not every feed that has ever existed — feeds the index has marked dead leave the database entirely, and there are 3,245,744 of those — and it is not a count of shows. A show with two feeds counts twice; a feed the index has never discovered does not count at all.

We publish the raw count rather than a de-duplicated “shows” estimate because de-duplication needs a judgement call about what counts as the same show, and a number carrying a hidden judgement is not one anybody should cite.

These counts — carried, ever listed, actively publishing — are re-stamped weekly on the live tool: How many podcasts are there? →

Podcasting’s biggest year was 2020 #

Dating each feed by its oldest surviving episode, 1,015,023 feeds published their first episode in 2020 — more than the previous six years combined. The years after shrank steadily, until 2025 broke the slide: 356,901 new feeds, a quarter more than 2024.

We date a feed by oldestItemPubdate, its oldest surviving episode — not by createdOn, which records when Podcast Index first ingested the feed. Using createdOn would put a 2013 show in 2020 simply because that is when the index found it, and would turn the index’s own growth into a claim about podcasting’s.

The trade cuts the other way too: a show that deletes its back catalogue dates late, and 259,595 feeds (5.5%) carry no usable episode date at all. Those sit in the last column rather than being dropped.

One feed in eighteen is still publishing #

436,068 feeds published an episode in the 90 days before this Vintage. Podcast Index currently carries 4,711,475 feeds and has marked a further 3,245,744 dead, so of every feed it has ever listed, 5.5% is still going.

Active5.5%436,068an episode in the 90 days before the Vintage

Dormant53.7%4,275,407still carried by the index, but silent

Dead40.8%3,245,744on the dead-feeds list — IDs only

Dead feeds are counted from a different source. The weekly database ships only living feeds, so the dead are a companion list of bare feed IDs — no title, no language, no host. We can say how many there are; we cannot say what they were. Everything else in this report describes the 4,711,475 living feeds only.

The 2020 cohort barely survived

Splitting the living feeds by the year each one started shows why so much of the index is quiet: the boom years produced enormous cohorts that almost entirely stopped. Of the 1,015,023 feeds that began in 2020, 2.6% published anything in the 90 days before this Vintage.

Read as a survival curve, not a forecast: a 2026 cohort has had months to go quiet, a 2020 cohort has had six years. The curve is U-shaped — the oldest cohorts survive at higher rates because a 2005 feed still listed at all is a survivor by definition — and the interesting number is how deep the boom-era floor sits.

Two companies host half the index #

Attributed through PodHood’s published Host mapping, Spotify carries 40.4% of all feeds and Spreaker 12.0% — 52.4% between them. The concentration is a legacy of the boom: Anchor, now Spotify, is where most of the 2020 cohort was made.

Spotify1,902,97040.4%

Self-hosted / other604,36312.8%

Spreaker563,80212.0%

Buzzsprout260,9075.5%

SoundCloud191,4864.1%

iVoox184,7423.9%

RSS.com177,5123.8%

Podbean106,2902.3%

These numbers are frozenThis table is the 2026-08-17 Vintage and stays that way. For the current split — re-stamped every week — the Hosting market share tool is the live surface.Open the live tool →

More than half the index is in English #

2,609,127 feeds (55.4%) declare English, and its grip on active podcasting is tighter still: 65.3% of every feed that published in the last 90 days. The activity column retells the boom story — of Indonesian’s 101,575 feeds, almost all made in the Anchor era, 245 still publish — while Japanese runs the healthiest catalogue of the fifteen, with 22.4% active.

Englishen2,609,12755.4%10.9% act.

Spanishes696,55714.8%4.1% act.

Portuguesept265,1095.6%3.0% act.

Frenchfr123,2982.6%15.3% act.

Germande120,3012.6%21.1% act.

Indonesianin101,5752.2%0.2% act.

Italianit82,1171.7%9.4% act.

Japaneseja49,5971.1%22.4% act.

Traditional Chinesezh-hant48,5671.0%14.7% act.

Russianru47,5741.0%7.1% act.

Simplified Chinesezh-hans40,6560.9%13.8% act.

Dutchnl34,9070.7%12.7% act.

Indonesianid32,4120.7%0.4% act.

Arabicar30,8110.7%4.3% act.

Hindihi30,5840.6%1.3% act.

These numbers are frozenThe fifteen biggest of this Vintage’s languages, as declared — Chinese by script, Indonesian under both its tags. Every declared language, re-stamped weekly, is on the live tool.Open the live tool →

A further 115,875 feeds (2.5%) declare no usable language tag and are excluded from the ranking rather than guessed. Languages are as declared by each feed — no content analysis.

The boom was the least-English era

Slicing each first-episode cohort by language turns the census into a timeline. English fell from 67.4% of the 2018 cohort to 47.0% of 2020’s while Portuguese spiked to 10.7% and Indonesian, a rounding error before 2018, to 6.1% — the Anchor wave, made outside English and collapsed with it. The quieter post-boom story: French, German and Japanese each hold a larger share of every post-2021 cohort than of any boom year.

The post-boom risers, on their own scale

French, German, Italian and Japanese never boomed — their whole story lives under 5%, invisible on the axis above, so it gets its own.

Each line is a language’s share of that year’s dated cohort; the eight biggest are drawn across the two panels — mind the different axes — and all fifteen ranked languages are in the CSV below. Undated feeds have no place on a time axis and are excluded here.

The “90% quit by three episodes” folklore fails the census #

Counting the episodes currently in every living feed, 38.9% have published three episodes or fewer — not 90% — and 23.7% have published more than twenty, which no definition of a “top 1%” can absorb. Quitting early is real; the folklore’s numbers are not.

38.9%stopped at three or fewer

1,831,341 living feeds — the share the folklore puts at 90%.

23.7%made it past episode 20

1,114,590 living feeds — nearly one in four, not a top 1%.

63.8%the worst-case ceiling

Even attributing every dead feed to the quit-by-three group, of everything ever listed.

The census covers living feeds; the 3,245,744 dead are bare IDs with no episode counts, and dead feeds skew short — so the living-feeds figure understates the quit-early share, and the ceiling above is the honest upper bound at this Vintage. The full ten-bucket distribution, sliceable by language, is the live podfade census, recomputed on this report’s monthly cycle.

Month over month, starting next month #

This is the inaugural monthly edition, built from the first weekly Vintage in the archive — so there is no earlier edition to compare against, and this report makes no month-over-month claims. From the September 2026 edition onward, this section reports the month’s movement: how much the index grew, how many feeds were observed for the first time, how the active share shifted, and which hosts and languages gained or lost share, in percentage points.

Every delta in a future edition is computed against the previous edition’s frozen numbers — this page’s numbers — not against a “latest” endpoint. Two frozen editions make one frozen comparison, so a cited change can never drift after publication.

What this report cannot tell you

Nothing about listening

The Podcast Index dataset is feed metadata. It carries no downloads, no plays and no audience — any number here is about supply, never demand.

Feeds, not shows

A show published on two feeds counts twice. We do not de-duplicate, because doing so would bury an editorial judgement inside a number people cite.

Discovery bias

A feed the index has never found is invisible to us. Coverage is best for anything with an Apple listing and weakest for private and regional hosting.

One snapshot, not a trend

Everything here is derived from a single Vintage. Week-over-week flow metrics need consecutive Vintages, which begin accruing 2026-08-17.

ErrataNo corrections issued. Any correction will be appended here with its date; the figures above will not be edited.

Full methodology →

Cite this report

PodHood Insights, "The State of Podcasting, August 2026", published 2026-08-23, Vintage 2026-08-17. Retrieved from podhood.com/insights/reports/state-of-podcasting-2026-08

The download is the exact table behind every figure in this report, carrying the same Vintage and publication date.