Industries / Media, arts and recreation / Media and entertainment / Podcasting
Being found
How a listener picks one show out of millions: directories that list shows, charts that rank them, search that matches words, and recommendations that suggest them.
7
Milestones achieved, from 2005 to 2025.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2005
Apple's directory becomes the phonebook
iTunes 4.9 added a free podcast directory, a list of shows people could search and subscribe to in one click, launched with "over 3,000 free audio programs," which Apple called "one of the largest Podcast directories in the world." Fifteen years later, the founders of a rival directory described Apple's list as "what everybody considers the default yellow pages" for podcasts.
LessonOne company's list became the catalogue most apps drew on, and worry about that "very centralised private entity" led to an open alternative in 2020.
Credit
Apple
2018
Charts gamed, then reweighted
In October 2018 little-known shows jumped to the top of Apple's podcast chart and Serial fell to No. 26; Dave Zohrob of the data firm Chartable reported "a highly unusual amount of manipulation," and Podnews editor James Cridland wrote that cheats sold chart boosts for as little as $5 because the chart was "weighted in favour of new subscriptions." Apple quietly corrected the rankings, Podnews says playback and completion appear to have been added from July 2019, and Apple now says listening, follows and completion rate count, and ratings and reviews do not.
LessonA ranking built on one cheap signal invites cheats, and even a blended chart measures only its own app.
9to5mac.comjames.cridland.netpodnews.netpodcasters.apple.com
Credit
Apple; Chartable (Dave Zohrob) and Podnews (James Cridland) for exposing it
2019
Search starts hearing the words
In August 2019 Google Search began showing playable episodes picked by "Google's understanding of what's being talked about on a podcast," not just titles, for English searches in the US; the same month Podnews found Apple Podcasts' test apps returning shows that said "Google Pixel" aloud. Google later pulled back: playable podcast results left Search in February 2023, replaced by a recommendations feature, "What to Podcast."
LessonA way of finding shows that lives inside one company's product can be withdrawn at any time.
Credit
Google (Zack Reneau-Wedeen, Google Podcasts); Apple
2020
Recommendations raise listening, narrow tastes
A randomized field experiment on Spotify compared two kinds of podcast suggestion, one built from each user's music listening and one showing shows popular with people like them. The personal version raised podcast streams by 28.90%, but each user's podcast listening became 11.51% less varied, while variety across all users rose 5.96%.
LessonTuning suggestions only for more listening narrows what each person hears.
Credit
David Holtz, Benjamin Carterette, Praveen Chandar, Zahra Nazari, Henriette Cramer and Sinan Aral
2020
An open catalogue for anyone
Adam Curry and the developer Dave Jones launched the Podcast Index as an alternative to Apple's list, an open directory whose "core, categorized index will always be available for free, for any use," as Podnews reported in September 2020. Open counts also made podfading visible, shows that stop publishing: in February 2023 the index held 3.8 million podcasts, but only 496,076 had posted in the previous 90 days. In October 2026 CastFox, a podcast search company, reported on its own index of 4.8 million feeds that 22.8% of shows have one episode, 52.8% have six or fewer, and 6.8% posted in the previous 90 days.
LessonThe size of a catalogue overstates choice, because most listed shows are dormant.
Credit
Adam Curry and Dave Jones
2024
Transcripts arrive in Apple Podcasts
With iOS 17.4 in March 2024 Apple began making transcripts, the full text of what was said, automatically for new episodes in English, French, German and Spanish in more than 170 countries and regions. Listeners can search for a word, tap the text to jump there, and, if deaf or hard of hearing, read an episode without playing it; iMore noted that before this a show had to post its own transcript on a web page for such listeners. With iOS 26.2, first reported in November 2025, Apple also made chapters, titled sections with start times, for almost every episode over ten minutes, and on-screen links to podcasts mentioned aloud, according to Podnews.
Credit
Apple
2025
YouTube becomes the search box
In May 2025 YouTube launched a weekly US chart of the top 100 podcasts, ranked by watch time and counting only playlists that creators mark as podcasts; Podnews notes that each big chart measures something different inside its own app: recent follows on Apple, unique streams on Spotify and total listening time on YouTube. In a survey of 4,141 US podcast listeners published in November 2026, Sounds Profitable and Signal Hill Insights found that 55% named YouTube as where they plan to find their next podcast, against 8% for AI assistants.
LessonChart ranks cannot be compared across apps.
Credit
YouTube (Google)
Watch for
Asking Spotify for shows
In April 2026 Spotify extended Prompted Playlist to podcasts: a listener types a request and gets a list of episodes, each with a note on why it was picked, in a test version in English for Premium users in seven countries. In May 2026 Spotify added the ability to ask questions about the episode playing, and it says "more than half of listeners who have used the feature have discovered a new show."
Achieved whena survey or study not run by Spotify reports how many users found a show this way and kept listening to it.
Claimed by
Spotify, April and May 2026
Chatbots as a way in
Spotify in ChatGPT launched in October 2025 in English in 145 countries, letting people ask a chatbot for episodes by topic, host or guest; it reached the UK and EU in December 2025, and since March 2026 episodes play without logging in. Sounds Profitable's November 2026 survey found that 30% of US podcast listeners had tried a chatbot for this and 18% still listen to a show found that way, but only 8% plan to use one for their next show, and it called chatbots "not a regular discovery method."
Achieved whenindependent surveys show chatbots as a regular way to find shows.
Claimed by
Spotify, October 2025
Open
01
Search inside openly licensed shows
A free public website that searches what is said in podcasts, not just their titles, covering only feeds that declare a Creative Commons or public-domain licence in the namespace's podcast:license tag. It returns the episode and the minute where the words occur and links to the original audio instead of hosting it, and it leaves out NoDerivatives licences or shows only short snippets for them.
The briefCredit · unclaimed02
A scorecard for podcast search
A repeatable, published test of whether podcast apps find a show when you type its exact title, and of which feed fields each app searches, including episode titles, descriptions, transcripts and chapters, rerun every quarter.
The briefCredit · unclaimed03
Chapters checked against hosts' own
A public accuracy test for automatic chapters. Take episodes whose hosts wrote their own chapters, generate chapters automatically, and score the start times and titles against the hosts' versions.
The briefCredit · unclaimed
Open 01 · the brief
Search inside openly licensed shows
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free public website that searches what is said in podcasts, not just their titles, covering only feeds that declare a Creative Commons or public-domain licence in the namespace's podcast:license tag. It returns the episode and the minute where the words occur and links to the original audio instead of hosting it, and it leaves out NoDerivatives licences or shows only short snippets for them.
Why it is still open
Searching full transcripts "vastly outperformed" searching titles and descriptions in the TREC 2020 Podcasts Track, a shared test scored by assessors at the US National Institute of Standards and Technology, but Spotify "is no longer granting access to new users" to that test collection. The University of Michigan's replacement, SPoRC, with 1.1 million transcripts, covers only May and June 2020, is limited to non-commercial research and has no public search. Live transcript search is sold commercially, by Podscan from $100 a month; Listen Notes says about 12% of new episodes have transcripts; and the American Archive of Public Broadcasting says only a small fraction of its more than 200,000 items have them. This research found no free search limited to openly licensed shows with a published accuracy test.
ar5iv.arxiv.orgarxiv.orgpodscan.fmlistennotes.com2025.code4lib.orggithub.com
Built with what exists
The Podcast Index API (free developer key) to find feeds, then a check of each feed for podcast:license; transcripts with word times from whisper.cpp (see the Note "A first stable number for local AI") or ElevenLabs Scribe v3 (see the Note "Fewer word errors"); SQLite FTS5 or Meilisearch for the index; a static website.
First step
Pull feeds through the Podcast Index API, check each feed for a
podcast:licensetag naming a Creative Commons licence or the public domain, and publish the count and list.Done looks like
More than 1,000 episodes from at least 50 openly licensed feeds indexed, and a test set of 200 known quotes, each picked by ear by someone who never saw the transcripts, with its true start time. Publish the share of quotes found in the top 10 results and the median error in seconds, with the test set and code.
Open 02 · the brief
A scorecard for podcast search
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A repeatable, published test of whether podcast apps find a show when you type its exact title, and of which feed fields each app searches, including episode titles, descriptions, transcripts and chapters, rerun every quarter.
Why it is still open
The closest work is Podnews's July 2021 test by Mark Steadman and James Cridland, who planted nonsense words in two of their own feeds and checked 14 apps; Apple searched only titles and author tags, and no app searched the namespace's tag for people. It was a one-off, it tested hidden fields rather than exact titles of many real shows, and it did not test transcripts, which the authors suggested as a later experiment. This research found only anecdotal rankings of apps, no repeated published test.
Built with what exists
The Podcast Index API to draw a random sample of active show titles; Apple's iTunes Search API and the Spotify Web API search, used within their terms; testing by hand in apps without public search; your own test feed with planted words in each field, including a podcast:transcript file and podcast:chapters; a spreadsheet. Check each service's terms before any scripted queries and keep the volume low.
First step
Draw 300 titles in groups (one-word titles, titles shared by several shows, titles not in English, titles with punctuation) and test the first 50 by hand in Apple Podcasts and Spotify.
Done looks like
300 real show titles tested in at least 8 apps, recording whether the right show comes first, in the top 5 or not at all, with results for planted words in at least 10 feed fields, the CSV and scorecard published, and a second run three months later showing what changed.
Open 03 · the brief
Chapters checked against hosts' own
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public accuracy test for automatic chapters. Take episodes whose hosts wrote their own chapters, generate chapters automatically, and score the start times and titles against the hosts' versions.
Why it is still open
Apple now makes chapters for almost every episode over ten minutes, and the Podnews guide to the feature, updated in April 2026, reports no published measure of their accuracy; Apple shows creators only how often chapters are used. The TREC 2020 Podcasts Track scored summaries, not chapters.
Built with what exists
Host-written chapters from feeds that carry podcast:chapters, found through the Podcast Index API; whisper.cpp for transcripts with word times; Claude Fable 5.2 (see the Note "Three working devices") or another language model to propose chapter starts and titles.
First step
List 100 episodes whose feeds carry
podcast:chaptersfiles and save the host chapters as the answer key.Done looks like
100 episodes from more than 20 shows, scored for at least two systems on the share of host chapter starts matched within 30 seconds, the number of extra chapters, and titles rated "same topic" or not by two people, with results published.