What was released, and when
IAB Tech Lab is the advertising industry's technical standards body. Its podcast measurement guidelines are what hosting companies, ad networks and measurement firms are certified against, and they are the definition behind almost every download figure a podcast quotes. We explained how that definition works, and why a download cannot show that anyone listened, in how big a podcast audience is and why download numbers mislead. That guide was written against version 2.2. This one covers what changed in 2.3.
The draft was released for public comment on 21 July 2026, with comments due by 19 August. On 29 September Brad Pipkin, Tech Lab's director of product for advanced TV, announced the final version, saying the comment period produced no comments that required changes. The document itself was posted to a public GitHub repository on 28 September. Its list of contributing member companies has 84 names, including Spotify, Google, iHeartMedia, SiriusXM Media, NPR, The New York Times Company, Triton Digital, Podtrac and Nielsen.
Tech Lab lists six changes in the version notes: the explicit coverage of video; the new term “podcast consumer”; a description of URL-prefix measurement; guidance on enclosure URLs; a section on invalid traffic and platform anti-fraud responsibilities; and an explanation of three measurement windows.
What “video counts” actually means
The headline change is a scope clarification rather than a new metric. Version 2.3 says the guidelines apply to audio and video episodes “when distributed via open RSS feeds or podcast applications that rely on server-side file delivery”. A video episode delivered as an MP4 file, or as an HLS stream through podcast distribution, is subject to “the same measurement principles, filtering requirements, and metric definitions”. The document treats HLS as a form of progressive download, so the same logs and the same one-minute rule apply.
In practice, that means a video episode fetched by a podcast app from a show's RSS feed can be counted as a download in the same way as an audio file. It does not mean that a view of the same episode on YouTube becomes a download. YouTube is not a server-log RSS delivery, and it publishes its own view counts under its own definitions. Podnews reported on 21 September that three podcast ad agencies had said they would buy YouTube inventory only on YouTube's “engaged views” measure, after what it described as a change in August to how YouTube counts a public play. We have not verified that change against YouTube's own documentation; we report it as Podnews described it.
True streaming, in which an app keeps a live connection and can confirm what was played, is explicitly out of scope. The guidelines say they expect to address it in future versions.
Three windows, three answers
Every download count has to decide when a second request from the same device is a new download and when it is a duplicate. Version 2.2 referred to a “rolling 24 hr window”; 2.3 says that wording was unclear and that there are two ways to read “rolling”. It now describes three approaches. A calendar-day window resets at a fixed time each day. A “first touch” rolling window starts at the first qualifying request and lasts 24 hours. A “last touch” window extends every time another request arrives.
The guidelines include a worked example: the same episode requested from the same IP address and user agent at 10:00 pm on Monday, 3:00 am on Tuesday, 10:15 pm on Tuesday and 11:15 pm on Wednesday.
Valid downloads counted under each measurement window in IAB Tech Lab's worked example
Source: IAB Tech Lab, Podcast Technical Measurement Guidelines v2.3, section 5.5.1, September 2026.
The calendar window counts Monday 10:00 pm, Tuesday 3:00 am and Wednesday 11:15 pm. The first-touch window counts Monday 10:00 pm, Tuesday 10:15 pm and Wednesday 11:15 pm. The last-touch window counts only Monday 10:00 pm and Wednesday 11:15 pm, because each request inside 24 hours resets the clock. The bars are scaled to the four requests made.
The guidelines do not pick one. They say the choice “provides flexibility” and that implementers must apply their chosen logic consistently. For a reader of a podcast's numbers, the consequence is that two certified providers can count the same audience differently for reasons that have nothing to do with fraud or error. The example is deliberately small; how much the windows diverge across a real show depends on how often its audience re-requests episodes, which the guidelines do not estimate.
Discrepancies, prefixes and invalid traffic
Version 2.3 gives a figure for tolerable disagreement: when several third-party vendors measure the same show and agree on date range, geography and show selection, differences “within ±5%” are generally considered acceptable, and larger gaps may warrant investigation.
It also explains one common source of larger gaps. Many analytics and ad-attribution services work by adding a prefix to the episode's download address, so the request passes through their server first. That server sees the request but not the bytes, and some apps fetch the rest of the file directly from the host, so the prefix service cannot always confirm the one-minute threshold. Changing the address in the feed, by adding, swapping or removing a prefix, can make some apps treat an old episode as new and download it again. The guidelines' advice is not to change an episode's GUID when moving hosts and, if possible, not to change its title or publication date either.
The section on invalid traffic separates general invalid traffic, such as known bots and data centres, from sophisticated invalid traffic designed to look human. Tech Lab says the public document deliberately does not give specific guidance on the latter, so that fraudsters cannot adjust to it. It notes one practical tell: campaigns run through fabricated traffic can look promising on reach and produce “little to no results” on conversions. Apple Watch downloads remain filtered out as likely duplicates of the paired phone, even though the guidelines acknowledge that some are now independent downloads.
What version 2.3 still does not define
The guidelines define a download, a podcast consumer (a unique combination of IP address and user agent within a stated time frame), an ad delivered and a client-confirmed ad play. They do not define a play, a listen, a completion or a view. Tech Lab's announcement is direct about this: “A download is not the same thing as a confirmed play, and an ad delivered through server-side measurement is not always the same thing as a client-confirmed ad exposure.”
Those questions are the agenda for version 3.0. The working group lists them: how to measure consumption when playback signals exist, what should count as client-confirmed playback, listen-through or ad completion, and how to handle streaming, video, HLS, skipping, scrubbing and ad position. Tech Lab plans to release 3.0 for public comment in the middle of 2027.
Not everyone in the trade agrees on the route. Podnews reported on 17 September that a guide released by the IAB, Tech Lab's sister organisation, at its podcast upfront said the solution was “not to build a new framework from scratch” and that the practical challenge was knowing which existing standard applies where. We read that guide only as Podnews quoted it.
What to ask before trusting a podcast number
The guidelines are written for measurement providers, not for advertisers or guests. What follows is our reading of what they mean for a company deciding whether a show is worth paying for or appearing on.
Ask which unit you are being shown. A download figure from a certified host and a YouTube view figure are different measurements, made by different parties under different rules. Adding them together produces a number nobody has defined. The same distinction runs through our explainer on what an impression actually means. Ask which window, and who certified it. Version 2.3 makes the window a legitimate source of difference. A show that can name its host or measurement provider, and whether that provider appears on Tech Lab's list of compliant companies, is giving more information than one that quotes a round number. Treat small differences as noise. If two reports for the same show and period agree within 5 per cent, the guidelines regard that as normal. Larger gaps are worth a question, starting with whether the feed's episode addresses changed. For a guest, downloads are context, not outcome. A download count says how often an episode was fetched, not whether the segment with a company's executive was heard. Fit between the show's audience and a company's buyers matters more, as we discuss in what a podcast guest appearance costs, and how the ad market itself is sized is in why two IAB studies disagree about podcast ad growth.Methodology and limitations
This guide is based on the full text of IAB Tech Lab's Podcast Technical Measurement Guidelines v2.3 as published in the Tech Lab GitHub repository (file created 28 September 2026, last edited the same day), read on 5 October 2026; Tech Lab's announcement of the final version, dated 29 September 2026; and its press release of 21 July 2026 opening the comment period. The worked example and the chart are taken directly from section 5.5.1 of the guidelines; we did not apply the windows to any real log data. The YouTube “engaged views” development and the IAB guide of September 2026 are reported secondhand from Podnews, and we did not read the IAB guide or YouTube's documentation on the change. Four limitations apply. The guidelines are voluntary and describe what certified companies should do, not what every podcast does. We did not compare 2.3 line by line with 2.2 beyond the changes Tech Lab itself lists and the wording the new text says it replaced. We have not checked which companies are currently certified. And the effect of each window on a real show's numbers is not reported by Tech Lab and was not measured by us.
Conclusion
Version 2.3 does three useful things. It settles that a video episode delivered through a podcast feed is measured like an audio one, it admits that compliant providers use different de-duplication windows, and it gives buyers a tolerance, plus or minus 5 per cent, for disagreement between vendors. It does not bring podcast measurement any closer to knowing whether someone watched or listened; a download is still a request for at least a minute of file. For a company, the practical change is a better set of questions to ask: which unit, which window, which certified provider. The bigger change, a definition of playback, is a 2027 document.
Frequently Asked Questions
Do video podcast episodes count as downloads now?
When they are delivered through an open RSS feed or a podcast app that relies on server-side file delivery, yes. IAB Tech Lab's guidelines v2.3, finalised on 29 September 2026, say they apply to both audio and video episodes, including MP4 files and HLS streams, under the same one-minute rule. Views on YouTube are not downloads and are counted by YouTube's own rules.
What counts as a podcast download?
A request in which enough of the episode to play one minute of content was downloaded, after filtering out bots, pre-loads, two-byte range checks and duplicates from the same IP address and user agent within the provider's measurement window. Episodes shorter than a minute must be downloaded in full.
Why do two podcast analytics tools show different numbers?
Partly because certified providers may use different de-duplication windows: a calendar day, a first-touch rolling 24 hours or a last-touch window. In Tech Lab's example the same four requests count as three, three or two downloads. Prefix-based tools also see requests but not bytes. Tech Lab treats differences within 5 per cent as generally acceptable.
Does a podcast download mean someone listened?
No. The guidelines measure what a server delivered, not what was played. Tech Lab says a download is not the same thing as a confirmed play. A version 3.0, planned for public comment in mid-2027, is meant to address playback, listen-through and ad completion.
What is a client-confirmed ad play?
An ad play reported by a tracking beacon from the listener's device when the file is played, ideally with quartile markers. The guidelines call it the most accurate count, but most podcast apps do not send these signals to publishers.
What should a company ask before buying a podcast ad?
Which unit the numbers are in (downloads, consumers or platform views), which measurement window the provider uses, whether the provider is certified under IAB Tech Lab's compliance programme, and over what time frame unique consumers were counted. Those answers make two shows' numbers comparable; a round figure on its own does not.
Sources
- standards body IAB Tech Lab, Podcast Technical Measurement Guidelines, Version 2.3, GitHub, September 2026. Scope including video, measurement windows and worked example, prefix measurement, enclosure URLs, invalid traffic, ±5 per cent discrepancies, metric definitions. Read 5 October 2026.
- standards body Brad Pipkin, IAB Tech Lab, Clarifying Today's Podcast Measurement and Preparing for What Comes Next, 29 September 2026. Final release, no changes from comments, version 3.0 timetable and questions. Read 5 October 2026.
- standards body IAB Tech Lab, IAB Tech Lab Releases Podcast Technical Measurement Guidelines v2.3 for Public Comment, press release, 21 July 2026. Comment deadline of 19 August 2026. Read 5 October 2026.
- trade press Podnews, Agencies agree to only buy on “engaged views”, 21 September 2026, and IAB: a new framework is “not needed” for measurement, 17 September 2026. Secondhand account of YouTube view counting and of the IAB guide. Read 5 October 2026.