00:00 Cold open
Short answer
Treat a podcast download as proof that an episode file reached a device, not that anyone pressed play. Under IAB Tech Lab Podcast Measurement v2.3, released in September 2026, a request counts as a download only after bots and pre-loads are filtered out, at least one minute of the episode was sent, and repeats from the same IP address and user agent inside one measurement window are merged into one.
Before you buy on a cost per thousand downloads, ask the seller three things: which IAB-certified host or vendor counts the numbers, which measurement window it uses, and whether you are billed on downloads or on ads delivered.
00:29 Segment 1
What does one podcast download count?
The guidelines define four numbers. They sound alike, and sellers mix them up in decks, so here is each one in a sentence, in the guidelines' own terms.
- Download
- "a unique episode request that results in an episode being delivered to the podcast consumer's device", complete or partial, after filtering (v2.3, section 6.1).
- Podcast consumer
- One unique IP address plus user agent in a stated time frame, such as a day, week or month. Version 2.3 replaced the word "listener" with "podcast consumer", because video podcasts are now in scope (v2.3, sections 1.3 and 6.2).
- Ad delivered
- An ad inside a valid download whose bytes were sent. A dynamically inserted ad counts only when "100% of the ad content (all bytes)" was downloaded (v2.3, section 6.3).
- Client-confirmed ad play
- An ad that made the player send a tracking beacon when it was played. The guidelines call it "the most accurate count for ad plays in a podcast", but many podcast apps and platforms do not provide these signals (v2.3, section 6.3).
None of the first three proves a play. A file played while it downloads and a file saved for later look the same in the server logs:
"This distinction can typically only be discerned by the player."
IAB Tech Lab, Podcast Measurement v2.3, section 3.2
01:29 Segment 2
How does a file request become a counted download?
A host starts from its server logs, where every request for the episode file is a line. Each request, or each group of range requests from one player, has to pass four questions from section 5.5 of v2.3 before it counts:
Did the request carry episode bytes?
NoHEAD requests move no data and "should be excluded"; 304 "not modified" answers "should not be counted". Two-byte range requests (bytes 0-1), which apps send to test the server, are ignored.
YesNext question.
Did it come from a real podcast app?
NoRequests from known bots, data centers and VPN traffic, user agents that say they are bots or are malformed, and duplicates sent by a paired Apple Watch are filtered out.
YesNext question.
Was at least one minute of the episode sent?
NoTreated as a pre-load or an accidental tap. The header plus "enough of the podcast content to play for 1 minute" must be downloaded; for an episode shorter than a minute, the whole file. Range requests from one player may be added together to reach that minute.
YesNext question.
Was this IP address and user agent already counted in this window?
YesMerged into the earlier download. The guidelines' example: one episode requested 10 times by 6 user agents behind one IP address in 24 hours is 6 downloads and 6 consumers. IPv6 addresses are cut to their first 64 bits first.
NoCounted.
One counted download
The host must document and self-audit this process "at least twice a year" (v2.3, section 5.5.6).
02:39 Segment 3
Why can two certified hosts report different numbers?
Because "one window" is not one rule. Version 2.3 admits that the 24-hour window in v2.2 could be read two ways and now names three windows a host may use, as long as it applies its choice consistently (v2.3, section 5.5.1). Here is the guidelines' own example: one device, one episode, four requests.
| Request from the same IP and user agent | Calendar day | First-touch rolling 24 h | Last-touch rolling 24 h |
|---|---|---|---|
| Monday 10:00 PM | 1 | 1 | 1 |
| Tuesday 3:00 AM | 1 | 0 | 0 |
| Tuesday 10:15 PM | 0 | 1 | 0 |
| Wednesday 11:15 PM | 1 | 1 | 1 |
| Downloads counted | 3 | 3 | 2 |
The same four requests give 3 downloads under two windows and 2 under the third. Add separate filters and separate log access points, and the guidelines expect certified vendors to differ: "Differences within ±5%" are "generally considered acceptable", as long as both reports cover the same dates, geography and shows (v2.3, section 5.6). A gap larger than that between the seller's report and your own vendor is worth a question.
03:33 Segment 4
Downloads, consumers, ads delivered or plays: which number is which?
Sellers quote whichever number is largest. This table shows what each one measures, so you know what you are paying for.
| Metric | What it counts | Shows that someone played it? |
|---|---|---|
| Download | Filtered episode requests, one per IP and user agent per window, with at least 1 minute sent | No: the file was sent |
| Podcast consumer | Unique IP and user agent pairs in a stated time frame. IP hopping on phones can count one person twice; recycled IPs can merge two | No: it estimates devices |
| Ad delivered | Ads inside valid downloads whose bytes were sent; all bytes for an inserted ad | No: the ad was sent |
| Client-confirmed ad play | Ads that fired a tracking beacon from the player when played, ideally with quartiles | Yes, where the app supports it |
| Spotify plays | Episodes "watched or listened to for at least 30 seconds on Spotify" (Spotify glossary) | Yes, on Spotify only, and not an IAB download |
Spotify's own dashboard also shows "Downloads", which it defines as downloads "via your show's RSS feed on any podcast platform" (Spotify for Creators). Ask which of the two numbers is in the deck.
04:31 Segment 5
Why does the "ad delivered" rule matter more every year?
Because most podcast ads are no longer recorded into the episode. They are inserted by an ad server when the file is requested, and an inserted ad only counts as delivered when every byte of it was downloaded. That share grew from 48% to 92% of US podcast ad revenue in three years:
- 2019 48% inserted
- 2020 67% inserted
- 2021 84% inserted
- 2022 92% inserted
The money is real: US podcast advertising revenue grew 17.6% in 2025, to $2.9 billion (IAB and PwC, Internet Advertising Revenue Report, full year 2025). The same 2023 study found mid-roll took 63% of 2022 revenue, pre-roll 31% and post-roll 6%. A mid-roll sits deeper in the file, and an ad counts as delivered only when the part of the file that holds it was downloaded, so it follows that a download that stops early can still count while the mid-roll inside it does not (v2.3, section 6.3).
05:23 Segment 6
When did the rules change?
- September 2016
IAB Tech Lab publishes the first Podcast Measurement Technical Guidelines, v1.0.
- December 2017
Version 2.0.
- February 2021
Version 2.1.
- October 11, 2023
Apple explains iOS 17's changes to automatic downloads (Apple Podcasts for Creators). Downloads pause when a listener "hasn't played a show they follow (latest five episodes for more than 15 days)". When they resume, Apple "will not download previous episodes", and episodes "older than seven days" are not downloaded automatically as new. Both changes mean fewer automatic downloads; Apple gave no figure for the effect.
- May 2024
Version 2.2. On October 22, 2026, all but one of the hosts and measurement vendors on IAB Tech Lab's compliant companies list were certified on this version.
- September 2026
Version 2.3: it states explicitly that video podcasts delivered as MP4 files or HLS through open RSS follow the same rules, "listener" becomes "podcast consumer", and the three measurement windows are spelled out (IAB Tech Lab version history).
06:07 Segment 7
What should you ask a podcast before you book?
- Who counts the numbers? Find the host or measurement vendor on IAB Tech Lab's compliant companies list, with the version (2.2 or 2.3) and the date it was certified.
- Is it certified for the metric you buy? If you pay per ad delivered, the vendor should be listed for "Ad Delivery", not only "Download".
- Which window? Calendar day, first-touch or last-touch rolling. Use the same one when you compare shows.
- Downloads of what, over what period? Per episode within 30 days, per month, or all time are different numbers. Ask for the period in writing.
- Downloads or ads delivered on the invoice? For mid-rolls in dynamically inserted breaks, ads delivered is the closer measure of what you got.
- Video counted the same way? Under v2.3, video files and HLS delivered through RSS follow the same rules. True streaming is outside the guidelines (section 2.3), so ask how those plays were counted.
- A trend from before October 2023? Since iOS 17, Apple Podcasts no longer downloads the backlog when paused downloads resume, and skips episodes older than seven days. Compare recent episodes with recent episodes.
06:57 Segment 8
Questions buyers ask
Is a podcast CPM priced per 1,000 downloads?
Usually. CPM buys were 94% of US podcast ad revenue in 2022, and the IAB and PwC 2023 study defines cost per thousand as a "Price determined based on verifiable podcast downloads". Still write the unit into the insertion order: downloads or ads delivered, counted by a named certified vendor.
Are video podcast downloads counted the same way?
Yes. Version 2.3 states explicitly that video episodes delivered as MP4 files or HLS through open RSS feeds follow the same filters, one-minute rule and windows as audio.
Why did a show's downloads drop in late 2023?
One known cause is Apple's iOS 17 change, explained on October 11, 2023: automatic downloads stay paused for followers who have not played the show's latest five episodes for more than 15 days, and when they resume, Apple no longer downloads the backlog or episodes older than seven days.
How big a gap between two reports is normal?
IAB Tech Lab treats differences within ±5% between certified vendors as generally acceptable, when both reports cover the same dates, geography and shows. Larger gaps usually come from different windows, filters or log access points.
07:49 Credits
Sources
- IAB Tech Lab, Podcast Technical Measurement Guidelines v2.3 (September 2026)
- IAB Tech Lab, Podcast Measurement Guidelines: version history
- IAB Tech Lab, compliant companies: podcast measurement
- Apple Podcasts for Creators, iOS 17 automatic download improvements (October 11, 2023)
- IAB and PwC, US Podcast Advertising Revenue Study 2023
- IAB and PwC, Internet Advertising Revenue Report, full year 2025 (April 2026)
- Spotify for Creators, analytics glossary
Post-roll
