Why podcasts fail

Akshay Datt

host of Founder Thesis.

The podcast failure rate depends on which feeds are counted. Across all four million registered feeds, 47% have produced three or fewer episodes. Across a curated index of discoverable shows, about 99% reach episode three and about 60% reach a hundred. The gap is the denominator, not the medium.

What is the podcast failure rate?

Podcast failure rate is the share of podcast feeds that stop publishing, usually measured as the proportion that never pass a given episode count or that have published nothing inside a stated window. It has no single value because every published figure uses a different population and a different cut-off.

One number comes from an actual count of the whole catalogue: 47% of four million feeds had produced three or fewer episodes. That describes a catalogue in which registering an RSS feed costs nothing, so it fills with things that were never production commitments.

Two questions are tangled together. Why the catalogue looks like a graveyard is a selection artefact. Why a funded company show with a real budget line stops after fifteen episodes has a different cause.

Do 90% of podcasts fail by episode three?

No dataset produces that figure. The claim has a traceable path but no primary source, and the counted figure for the full catalogue is 47% with three or fewer episodes.

Forbes ran it on 6 January 2026: “According to a May 2025 report by Podnews, 90% stop after episode three, and another large chunk quit by episode 20.” The Podnews item is not a report. It is a first-person how-to headlined “How to outlast 99% of podcasts: four simple rules,” and it credits the episode-three figure to nobody in particular: “Many industry sources report that 90% of podcasts don’t make it past three episodes.”

Its companion claim is that of the tenth who get past episode three, 90% will not pass episode 20. That sentence carries the only hyperlink, and it points at a blog post by Riverside. As Riverside’s page stood in May 2025, its own source for the episode-three figure was a tweet from 2021. The episode-20 figure had no source at all. Follow the same link today and it redirects to a page where both numbers have been replaced by different ones.

Where does the seven-episode podfade number come from?

The seven-episode podfade number comes from nowhere traceable. The likeliest explanation is back-formation: podfade folklore, which dates to podcaster Scott Fletcher in February 2005 when he gave up two of his own shows, has long said podcasters quit around six weeks in, and six weeks of a weekly show lands between episode six and episode eight.

Buzzsprout’s podfade article says many podcasters do not get past their seventh episode “without experiencing a decrease in motivation,” a statement about morale rather than survival. The version that circulates has dropped the second half of that sentence.

Both folk numbers dramatise a phenomenon already dramatic at its measured value. Nobody checks a number that feels right.

Why do podcast survival statistics contradict each other?

Podcast survival statistics contradict each other because the denominators differ. Full-catalogue counts include every feed ever registered, including test feeds, spam feeds and single-episode experiments, which is exactly the population that produces the attrition. Curated indexes count shows that already cleared the bar of being catalogued as a discoverable programme.

Inside the curated index, the median show publishes across a span of 2.4 years.

What percentage of podcasts are active?

The share of active podcasts depends on the window, and the same catalogue at the same moment produces two different answers. In the four-million-feed catalogue, 12% had published in the previous 90 days, 8% in the previous 30, and under 4% had ten or more episodes plus a new episode in the week reviewed.

“Only X% of podcasts are active” is an empty sentence without the window attached.

Is the average podcast lifespan 174 days?

No. The 174-day figure is a median age, not an average podcast lifespan, and it was measured as the distance from a feed’s first episode to its most recent across shows indexable in November 2018. That is a snapshot across a population that includes shows launched the previous week and still running happily.

Two sentences later the same post reads: “The average podcast lifespan of 174 days is based on shows that were indexable in November 2018.” Median has become average, age has become lifespan, and the following section is headed “Why does a podcast’s lifespan matter?”

Tim Denning linked straight back to that post and inherited its wording: “The average lifespan of a podcast is 174 days before it dies an inevitable audio death.” The citation was intact. What travelled was the sentence the source had already got wrong about its own measurement.

Why do company podcasts stop?

Company podcasts stop for three reasons, in order of how much each explains: the commissioning budget was approved as a fixed unit and not renewed, the show belonged to one person and that person moved, and nobody agreed in advance what “working” meant.

1. The commissioning budget is a procurement event

Well-funded brand shows stop at the commissioning boundary. Where they stop:

• Open For Business (eBay, Gimlet Creative): 16 episodes, exactly eight in 2016 and eight in 2017, the last in April 2017.
• DTR (Tinder, Gimlet Creative): 15 episodes, December 2016 to December 2017.
• The Message / LifeAfter (GE Podcast Theater, Panoply): nine episodes in 2015, eleven in 2016, one in January 2017.
• Slack Variety Pack: 29 episodes, May 2015 to October 2016.

That is not a curve of dwindling enthusiasm. That is a budget line approved once, approved a second time and not renewed a third. A show commissioned as a season of eight ends at eight or sixteen because eight was the unit of purchase. The episode number where these shows stop describes a procurement cycle, not anybody’s stamina.

Nor do company shows always fade. eBay for Business published 312 episodes between August 2018 and September 2024, holding fifty a year for six straight years, then ended anyway, announced on air by its host with no reason given and no replacement. That show did not run out of steam. It was switched off.

2. The show belongs to a person, and the person moves

A company podcast tied to one person ends when that person moves. Sonal Chokshi built and ran the a16z podcast operation for nearly eight years and was its most frequent host. When a16z announced her move to a different team, the flagship show continued under a new host. The secondary show, 16 Minutes, did not: 17, 32 and 21 episodes across its first three years, then exactly one in 2022 and stop.

Behind The Tech, hosted by Microsoft CTO Kevin Scott, has published nothing since 18 February 2025: eighteen months of silence at one of the best-resourced companies on earth, with no ending announced.

Average CMO tenure at Fortune 500 companies was 4.3 years in 2024. A format that takes years to compound is frequently owned by a role that turns over in four.

The company shows still publishing after eight to twelve years have either changed host or never rested on one. Shopify Masters has run since 2014 under three hosts. REWORK changed host in October 2022 and kept its weekly slot, with 37signals’ founders as recurring voices rather than as the engine. The McKinsey Podcast is co-hosted by two staff editors rather than a principal.

Silence is not an ending. The accurate description of a quiet feed is the date of its last episode.

3. Nobody agreed in advance what “working” meant

A company podcast with no agreed definition of success gets cancelled against a number nobody chose. Steven Goldstein of Amplifi Media puts it directly: “success is impossible to attain unless it has been defined and is measurable.” Steve Pratt, co-founder of Pacific Content, is blunter: “If you need to drive short-term business results, a branded podcast is one of the worst tools you can use.”

Absent an agreed yardstick, a show gets scored at renewal against whatever number happens to be in the room.

Does a seasonal podcast count as failed?

No, and the unit-of-purchase logic that ends open-ended shows is what makes a seasonal podcast work. A gap that reads as death in a weekly feed is the format working as designed in a seasonal one.

Sequoia’s Crucible Moments, hosted by partner Roelof Botha, runs three seasons: 30 August to 30 November 2023, then 29 August 2024 to 23 January 2025, then from 23 October 2025. Both gaps ran about nine months. Nine silent months in an open-ended weekly show is the standard description of a dead feed. Here it is the format.

Inside Trader Joe’s was announced in May 2018 as a five-part series and is past 130 episodes across eight years. What a season changes for certain is how a gap reads.

What is a normal number of podcast downloads?

A normal number of podcast downloads is smaller than most people expect. The median episode gets 27 downloads in its first seven days. The top 25% get 96, the top 10% get 407, the top 5% get 1,003.

Measured over the first 30 days instead, the median is 124, the top-20% threshold is 1,000, the top-10% threshold is 2,700 and the adjusted mean is 1,170. Those medians cannot be compared. One window is seven days, the other thirty, and they are routinely quoted side by side as rival estimates of the same quantity.

Inside the 30-day set, the adjusted mean of 1,170 sits above its own top-20% threshold of 1,000. An episode performing at the average outperforms four episodes in five. That is why “average downloads” is the most misleading number in this business. A company show doing 200 downloads in its first week sits between the top-25% and top-10% thresholds while sounding like a rounding error in a quarterly review.

How many hours does a podcast episode take?

A podcast episode takes anywhere from around five hours end to end for a standard interview, eight to ten hours a week for a properly edited one, and 25 or more for a scripted narrative.

At REWORK, one in-house host-producer selects topics, records single-take, corrects the AI transcript and reviews video at 2x, with a third-party editor handling the cut, for a 30-minute weekly show.

The arithmetic:

• Six hours per episode across 52 weeks is about 312 hours a year.
• That is 15% of one full-time role on a 2,080-hour year.
• Fortnightly halves it to 156 hours.

Long-running company shows run 16 to 50 episodes a year, and nothing goes faster except a feed staffed by a full editorial department.

What to check before launching a podcast

Four things should be checked before a podcast launches, each checkable in advance.

Name the person on payroll whose job description contains the show. Not the person who is excited about it, the person whose performance review it appears in. If that is a senior executive and nobody else, the show’s lifespan is bounded by that person’s busiest quarter.

Design for the host leaving. Every company show still running after a decade has either changed host or was never built around one. That is easier when the title is not a person’s name and the premise is not one person’s contact list.

Cost one episode in hours, multiply by the year, then look at a real calendar. Six hours weekly is 312 hours. If that does not visibly fit somewhere, the show will stop in the quarter when everything else gets busy.

Agree the measurement window and the threshold before recording anything, and commit publicly to a specific number of episodes. Against an “average” that four episodes in five fall short of, nearly everything looks like failure. A show that ends where it said it would end is a completed project; a show that trails off is a visible one.

Discipline is not what separates them. The company shows that lasted do not look more disciplined than the ones that stopped, and several that stopped were better produced than several that continued. What separates them is that the weekly work was smaller than the slack in the calendar of the person doing it, and that somebody had written down in advance what the show was supposed to achieve. Sustainability here is mostly a subtraction problem.

Frequently asked questions