BRENNEN

Industries / Media, arts and recreation / Media and entertainment / Podcasting

Advertising 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

Milestones achieved, from 2014 to 2025.

2

To watch for: announced, not yet confirmed outside.

3

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2014

    Networks sell host-read ads

    A host-read ad is one the host reads in their own words, often from the advertiser's talking points. By late 2014 the ad network Midroll sold sponsor spots for 175 shows, including WTF and StarTalk, its advertisers had grown from 101 in 2013 to 259, and its chief executive Adam Sachs said "all our ads are host-read." That year Serial, a spinoff of This American Life, had one sponsor, MailChimp, and Slate reported prices of about $15 to $45 per thousand listens on popular shows, citing the Wall Street Journal for a price of $25 to $40 per thousand paid by MailChimp for Serial.

    slate.com

    Credit

    Midroll (Adam Sachs); Serial and MailChimp

  2. 2016

    Ads stitched in at download

    Before this, ads "were recorded by the host and hard-baked into the audio file" and stayed there for good. In January 2016 Panoply, the Slate Group's audio arm, launched Megaphone, built on "a real-time ad insertion engine" that adds a current ad to the file at the moment each listener downloads it, a method called dynamic ad insertion; Gimlet Media was using Megaphone and Acast this way by February 2016. In an IAB and PwC survey of 20 large companies, dynamically inserted ads rose from 37% of ad revenue in 2015 to 56% in 2016. In January 2020 Spotify began placing ads as an episode streams in its own app, for its own original and exclusive shows, and in November 2020 it bought Megaphone for a reported $235 million.

    mediashift.orgadexchanger.comiab.comradioink.comtubefilter.com

    Credit

    Panoply (Megaphone); Acast; Spotify

  3. 2017

    The industry counts ad money

    In June 2017 the IAB, the online advertising trade body, published a podcast ad revenue study run by PwC: 20 of the largest podcast ad companies took $69 million in 2015 and $119 million in 2016, which PwC says is less than the whole market, and host-read ads made up 63% and then 60% of it. The same series puts US podcast ad revenue at $1,925 million for 2023, $2,433 million for 2024 and $2,862 million for 2025, counting audio only; its latest report says the definition "is rapidly expanding as video becomes the leading channel." In Australia, IAB Australia's figures, compiled by PwC, show a quarterly record of $40.0 million in podcast ads in the June 2026 quarter.

    iab.comiabhongkong.compodnews.net

    Credit

    IAB and PwC; IAB Australia and PwC Australia

  4. 2021

    Paid shows inside the apps

    Apple Podcasts Subscriptions went live on 15 June 2021 in more than 170 countries and regions, after slipping from May: creators set a monthly price from $0.49, pay $19.99 a year to join, and Apple keeps 30% of subscription revenue in the first year and 15% after, with perks such as listening without ads. Spotify opened paid podcasts through Anchor that April, taking no cut for two years and 5% after. On 2 January 2025 the Spotify Partner Program began paying video podcasters in the US, UK, Canada and Australia by how much their videos are watched, with no ads for Premium subscribers, and Spotify says it paid podcasters more than $100 million in the first quarter of 2025 from ads and the program together, its own figure.

    apple.comtechcrunch.comtechcrunch.comnewsroom.spotify.com

    Credit

    Apple; Spotify

  5. 2022

    Regulators rule on host reads

    On 10 August 2022 the UK's Advertising Standards Authority (ASA) upheld a complaint against a Huel ad that Steven Bartlett read on The Diary of a CEO: no words up front said it was an ad, the page-turn sounds that framed it were "brief and quiet," and his delivery was in "a similar style and tone to the editorial material," though the episode's written description named Huel as a sponsor. Guidance from the ASA and CAP, the body that writes the UK ad code, took effect on 16 August 2024 and strongly advises a short spoken label at the start, such as "paid-for advertisement" or "sponsored." In the US, the Federal Trade Commission revised its Endorsement Guides in 2023, and its guidance says: "If the representation is made audibly, the disclosure should be made at least audibly."

    LessonA trusted host's voice can make an ad sound like the show, so the label has to be heard at the start.

    asa.org.ukasa.org.ukftc.gov

    Credit

    ASA and CAP; US Federal Trade Commission

  6. 2023

    Growth slows, studios shrink

    IAB and PwC found US podcast ad revenue grew only 5% in 2023, to $1.9 billion, in "a challenging advertising climate that particularly affected mid-tier companies." In June 2023 Spotify cut 200 podcast jobs, about 2% of its staff, and merged the studios it had bought, Gimlet and Parcast, into Spotify Studios. Growth returned to 26% in 2024.

    LessonPodcast ad money rises and falls with the wider ad market, and mid-sized sellers feel a downturn first.

    iab.comtubefilter.comiabhongkong.com

    Credit

    IAB and PwC (the count); Spotify

  7. 2025

    Outsiders time the ads

    Ad load is the share of an episode's running time given to ads. Magellan AI, an ad-tracking firm, analysed 94,724 episodes of popular US podcasts and reported an average advertiser ad load of 8.34% in the third quarter of 2025, about 5 minutes an hour. Separately, a University of Hamburg master's thesis timed the ads by hand in 160 episodes of popular US shows and found 8.95% ads, and found 3.43% in 1,000 podcast-like YouTube videos, using labels from SponsorBlock, a tool whose users mark sponsor segments.

    magellan.aiinf.uni-hamburg.de

    Credit

    Magellan AI; Felix Wolf, Universität Hamburg

Watch for

  1. Cloned host voices read ads

    Spreaker, part of Triton Digital, sells host-read ads voiced by AI copies of hosts' voices, made with ekoz.ai: hosts opt in by giving a voice sample, AI writes each script from the advertiser's talking points, and the host approves the ad before it runs. The companies say it is "already delivering millions of impressions" but give no figures, and the announcement does not say whether listeners are told the voice is synthetic.

    Achieved whenan independent ad tracker or researcher finds AI-voiced host reads running on many shows and reports how many there are and how often listeners are told.

    podnews.net

    Claimed by

    Triton Digital (Spreaker) and ekoz.ai, July 2025

  2. Podcast ads bought by software

    Programmatic buying means advertisers' software bids for ad slots in automated auctions instead of dealing with a sales team. In July 2025 Spotify opened its own, licensed and partner-network podcast ads to this kind of buying, through a self-serve Ads Manager in 8 countries and an Ad Exchange in 12, and claimed about 170 million monthly podcast listeners and a 64% rise in advertisers using the exchange since April 2025.

    Achieved whenan outside count, such as the IAB and PwC revenue study, reports what share of podcast ad revenue is bought programmatically and shows it rising.

    podnews.net

    Claimed by

    Spotify, July 2025

Open

  1. 01

    Open test set of ads

    A public set of podcast episodes with every ad marked by start time, end time and type (host-read or inserted), each labelled by two people. Publish the timestamps and feed links, not the audio, and record the download date and country, because dynamic insertion serves different ads by time and place.

    The brief
    Credit · unclaimed
  2. 02

    UK ad label audit

    A checker that finds each host-read ad in a transcript and asks whether a clear label, such as "ad," "sponsored" or "paid-for advertisement," is spoken at its start, as the 2024 CAP guidance advises. Run it on one recent episode from each of the top 100 UK shows; the same rules could later test the FTC's "at least audibly" rule in the US.

    The brief
    Credit · unclaimed
  3. 03

    Test automatic ad break placement

    An independent test of hosting companies' automatic mid-roll placement, in which software picks where inserted ads will interrupt an episode. Upload the same episodes to each service and score every break it chooses as mid-word, mid-sentence, at a sentence end, or at a change of topic.

    The brief
    Credit · unclaimed

Open 01 · the brief

Open test set of ads

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public set of podcast episodes with every ad marked by start time, end time and type (host-read or inserted), each labelled by two people. Publish the timestamps and feed links, not the audio, and record the download date and country, because dynamic insertion serves different ads by time and place.

Why it is still open

Wolf's 2025 thesis says, "to our knowledge, a labeled dataset of advertisements in podcasts is not publicly available," and notes that Spotify's 100,000-episode research dataset has been withdrawn. This research found Wolf's own 160 episodes, with no public release stated; Spotify researchers' 2021 labels for "extraneous content," built on the withdrawn dataset; and SponsorBlock, which covers YouTube only and in Wolf's check had valid labels for 78% of 100 sampled. Magellan AI and Podscribe detect ads commercially and do not publish their data.

inf.uni-hamburg.dearxiv.org

Built with what exists

Public RSS feeds; whisper.cpp (see the Note "A first stable number for local AI") or ElevenLabs Scribe v3 (see the Note "Fewer word errors") for transcripts with word timings; Audacity or Label Studio for marking; Claude Fable 5.2 (see the Note "Three working devices") as a first detector to beat.

  1. First step

    Download two recent episodes from each of 10 shows from one logged location, label 5 of them twice, and measure how well the two labellers agree.

  2. Done looks like

    300 episodes from 60 shows, each labelled twice, with a score for how well the labellers agree and a published detector score to compare with the best F1 score of 0.80 in Wolf's thesis (F1 combines how many ads a detector finds with how many of its finds are right).

Open 02 · the brief

UK ad label audit

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A checker that finds each host-read ad in a transcript and asks whether a clear label, such as "ad," "sponsored" or "paid-for advertisement," is spoken at its start, as the 2024 CAP guidance advises. Run it on one recent episode from each of the top 100 UK shows; the same rules could later test the FTC's "at least audibly" rule in the US.

Why it is still open

The ASA's AI monitoring of 2025 checked more than 50,000 influencer posts on Instagram and TikTok and does not mention podcasts. A November 2026 review of five episodes looked only at written show notes and says this "is not a substitute for checking the audio." Podscribe offers ad checks privately to advertisers. This research found no published count of compliance since the guidance took effect.

asa.org.ukeverything-pr.comasa.org.uk

Built with what exists

Public RSS feeds downloaded from a UK connection; whisper.cpp or ElevenLabs Scribe v3; Claude Fable 5.2 to flag ad reads and their opening words; the CAP guidance as the rulebook. Keep the audio private, publish counts and short quotes only, and follow the terms of the chart and hosting sites.

  1. First step

    Transcribe 10 episodes from the current UK chart, mark the ad reads by hand, and write the labelling rules from the CAP guidance.

  2. Done looks like

    100 shows checked, the share of host reads with a spoken label at the start, and the checker's precision and recall (the share of its flags that are right, and the share of real cases it finds) against 200 ads labelled by hand, all with the method published.

Open 03 · the brief

Test automatic ad break placement

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An independent test of hosting companies' automatic mid-roll placement, in which software picks where inserted ads will interrupt an episode. Upload the same episodes to each service and score every break it chooses as mid-word, mid-sentence, at a sentence end, or at a change of topic.

Why it is still open

Megaphone's page on Smart Ad Locations gives no accuracy figures, does not let you approve placements during setup and warns of "inaccurate placement" on shows with ads already in the audio; Acast says it uses "silence detection" and also gives no figures. This research found other vendors' descriptions, from Podigee, Spreaker, StreamGuys, OCDevel and the Washington Post, but no independent tally.

support.megaphone.fmlearn.acast.com

Built with what exists

Your own recordings or Creative Commons episodes; accounts at hosting companies whose terms allow test uploads, such as Acast and Megaphone; whisper.cpp word times (see the Note "A first stable number for local AI"); a short script comparing break times with sentence ends.

  1. First step

    Prepare 5 episodes, mark their sentence ends from a transcript with word times, upload them to two services that place breaks automatically, and log where the breaks land.

  2. Done looks like

    30 episodes on at least 4 services, every automatic break scored, and a published table of the share placed mid-sentence, with timestamps.