Industries / Media, arts and recreation / Media and entertainment
Podcasting
Podcasting is the making and distributing of audio shows, and now often video ones, published as series of episodes that people download or stream through apps, most of them delivered by an RSS feed, a web file listing a show's episodes that apps check for new ones.
2003
In July 2003 the journalist and radio host Christopher Lydon interviewed the software developer Dave Winer at Harvard's Berkman Center for Internet & Society, and the two released Lydon's interviews on his blog through Winer's "RSS-with-enclosures feed that attached the audio files"; the Harvard Gazette calls the interview "what would become the very first podcast." Podnews, in an article by Eric Nuzum, says the conversation "usually settles" on Lydon's episode of 9 July 2003, but dates the first podcast feed to 20 January 2001 (see Hosting and distribution).
42
Milestones achieved across 6 domains.
14
To watch for: announced, not yet confirmed outside.
18
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Podcasting grew out of blogging software. In December 2000 Dave Winer added the "enclosure" to RSS, which let a feed carry an audio file, and in July 2003 Christopher Lydon's interviews at Harvard went out that way. Apple built podcasts into iTunes in June 2005 with a directory of more than 3,000 free shows, and that directory became the catalogue most apps drew on. Recording took longer to settle: hosts interviewed distant guests over Skype calls until browser tools from 2015 on recorded each guest on their own computer.
Where it stands
The catalogue is large and mostly idle. Listen Notes counted about 3.8 million podcasts on 19 November 2026, and one developer's count of the Podcast Index database of August 2026 found 421,928 of 4.7 million feeds still live. Edison Research finds that 58% of Americans aged 12 and over listened in the past month in 2026. The Reuters Institute's 2025 report says about a third of the people it surveyed in 48 markets used a podcast monthly. Video now carries much of the audience: YouTube says it has more than 1 billion monthly podcast viewers, and IAB and PwC count US podcast ad revenue at $2,862 million in 2025. Machine-made shows have arrived as well: on one day in April 2026 the Podcast Index rated 45.7% of new shows as possibly made by AI.
What is open
Much of the trade's checking is done in private or not at all. This research found no published controlled test of whether certified hosts count the same downloads alike, no count of how many old episode files still download, no blind listener test of an AI edit against a human editor's, and no public set of episodes with every ad marked. Speech-to-text software that times each word, such as whisper.cpp and ElevenLabs Scribe v3, AI language models and the open Podcast Index database put each of these jobs within reach of a small team.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Recording
Capturing the voices of a show, in one room or across a distance, and now often on camera as well.
7 · 2 · 3
Last achievedThe double-ender moves into iPadOS2025
Watch forStudio-quality vocals from AirPods
3
openEditing
Turning raw recordings into a finished episode: cutting, cleaning up the sound, mixing and setting loudness, and more and more using AI to make voices, translations or whole shows.
7 · 3 · 3
Last achievedThe AI show factory retreats2025
Watch forRaw tape to rough cut
3
openHosting and distribution
Storing a show's audio, and now video, files on a server and getting each new episode from the maker to every listening app.
7 · 2 · 3
Last achievedBig apps bypass the feed2024
Watch forOne video upload, two apps
3
openAdvertising and paid shows
How shows earn money: selling time in episodes to sponsors, getting those ads into the audio, and charging listeners or platforms for shows.
7 · 2 · 3
Last achievedOutsiders time the ads2025
Watch forCloned host voices read ads
3
openMeasuring the audience
Counting how many people hear or watch a show and finding out who they are, through surveys, server logs (the record a web server keeps of every file request) and the play data that apps collect.
7 · 3 · 3
Last achievedVideo joins the count2025
Watch forA play, not a download
3
openBeing 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 · 2 · 3
Last achievedYouTube becomes the search box2025
Watch forAsking Spotify for shows
3
open
Advances to build with
1 research note
Open in podcasting
3 of 18 open deliverables
Open · Recording
Blind test of speech cleaners
A public test of AI speech cleaners on real, deliberately bad podcast recordings. Record about 30 clips of your own voice or of consenting volunteers: a laptop microphone, an echoing kitchen, a phone on speaker, street noise, two people talking over each other. Run every clip through five or more tools, have strangers rate each result without knowing which tool made it, and separately count every word each tool changed.
Built with what exists
Adobe Enhance Speech, Descript, Auphonic, Audo Studio and iZotope RX as the tools under test; Microsoft's open-source P.808 Toolkit, with listeners recruited on Prolific or Amazon Mechanical Turk; Whisper Large v3 or ElevenLabs Scribe v3 (see the Note "Fewer word errors") to transcribe before and after. Check each tool's terms on publishing test results.
Open · Editing
Blind test of AI edits
Collect raw interviews, either your own recordings made with the guests' consent or Creative Commons tape. Have an AI editor and a human editor each cut every interview to the same length, set both versions to the same loudness, let listeners hear each pair blind and pick one, and log every cut each version made.
Built with what exists
Descript Underlord and Riverside's editor for the AI cuts; WhisperX, an open speech-to-text tool that times each word, to list what each version removed; ffmpeg's loudnorm filter to set both versions to −16 LUFS; a simple web page that plays each pair in random order.
Open · Hosting and distribution
Count the dead audio
A census of how many podcast episode files still download. Draw a random sample of feeds from the Podcast Index's weekly public database, grouped by the year of each show's newest episode from 2005 to 2026, pull each feed's stored episode list from the Podcast Index API, and send a light request (an HTTP HEAD request, or a request for a single byte) to every audio file's address. Publish the share that fail, by year and by host, with the code and raw results.
Built with what exists
The Podcast Index database file (SQLite, about 1.8 GB compressed, updated weekly); a free Podcast Index API key, whose episodes-by-feed call returns up to 1,000 episodes per feed; Python with httpx or aiohttp; DuckDB for analysis. Name the crawler in each request and keep request rates low, so hosts can leave it out of their download counts.
Where it sits
Broader
