What counts as a podcast download?

A podcast download counts, under the IAB Tech Lab's Podcast Technical Measurement Guidelines, when a listener's app requests the file header plus enough of the audio to play for one minute, and that request survives a 24-hour window in which repeat requests from the same IP address and user agent collapse into one count. Episodes shorter than a minute need the whole file. Requests for zero to one byte, HEAD requests and 304 responses never count. Every number your host labels an IAB-certified download sits on that definition, which the IAB Tech Lab last revised on 29 September 2026.

Any podcast can be listed on PodAnswer for free, and a listed show gets its own page that can turn up when somebody searches the show's name, so if you have not claimed yours you can add it at https://podanswer.com/signup.

Section 5.4.3 and the minute it asks for

The operative language in the guidelines is short. Step 2 of section 5.4.3 requires "the header information plus enough of the podcast content to play for 1 minute." It is not a fixed byte count. Publishers work out how many bytes one minute represents for each episode from the file's bitrate plus the size of its ID3 tags, which means the threshold moves episode to episode. Where that size cannot be computed, or where the episode runs under a minute, the spec falls back to requiring a complete file download.

Response codes do most of the remaining work. A 200 response counts as a valid download. A 206 partial response counts only if it clears the one-minute bar, and 206s get deduplicated by IP and user agent because a listener skipping around in an episode generates a string of range requests that have to be reassembled before anything is counted. Akamai's code 000 on a 206 that ended early counts only if the request passes every other filter.

The minute itself matters less than podcasters assume. Podnews ran the arithmetic on real data and found that recalculating IAB figures at a 30-second threshold instead of 60 seconds moved the totals by roughly 1%. Almost everybody who starts an episode gets past a minute, or does not start it at all.

The window two compliant hosts can set differently

Deduplication runs over 24 hours, and the spec has been deliberately permissive about where those 24 hours begin. Version 2.2 said only that "this 24 hour window can be a calendar day (fixed) or a rolling 24 hour window," and required measurement companies to disclose which they use in a document of methodology. Version 2.3, released on 29 September 2026, replaced that vague wording with three named window types, calendar day, first touch and last touch, each with a worked example.

The practical consequence is one most hosts never spell out on the stats page: two providers can report different download totals for the same feed on the same day and both be fully compliant. If you have ever compared your host's dashboard against a prefix analytics service and found a gap you could not explain, the window is the first thing to ask about.

Six phones, one router

Uniqueness is decided by a pair, not by a person. Section 5.4.4 states that "a combination of IP Address + UA is used to identify unique listeners and downloads." The document's own worked example makes the behaviour concrete: a single file requested ten times by six different user agents behind one IP address counts as six listeners and six downloads. IPv6 addresses are truncated to their first 64 bits before matching. Providers may bring in extra signals such as user IDs or cookies, but they have to disclose it.

Read that rule in both directions and your numbers start making sense. A household running three phones on one router, each with a different app build, reads as three. One person who plays the same episode on the morning commute and again at the gym that evening on the same phone reads as one.

The traffic that never reaches your total

Before any counting happens, measurement providers are required to throw material out. The filter list covers IP ranges that cannot plausibly belong to a listener, meaning servers, AWS and other data centres, and VPNs; known bots; user agents that identify themselves as bots; malformed or bogus user agents; and duplicate requests generated by paired Apple Watch devices. Inclusion lists of IPs treated as safe "must be re-validated at least every 90 days." The public version of the guidelines states outright that its handling of sophisticated invalid traffic is not specified.

That filtering is the mechanism behind a complaint that turns up every few months from hosts who paid for growth. A few thousand requests arriving from a data centre range are not ambiguous to a filter, and they do not land in an IAB-certified total. We went through what the sellers in that market deliver in what you actually get when you buy podcast downloads.

Four platforms, four different verbs

None of the above governs the word "plays," which is where most of the confusion in podcast analytics now lives. The IAB guidelines define downloads. Each platform defines its own listening metric, and they do not agree.

Apple Podcasts describes Plays as "the total number of times people pressed play on your episode," and in its episode table counts plays "where the duration is greater than zero seconds." Apple's episode-level Listeners figure counts people who "played or watched more than zero seconds of the episode." Its Engaged Listeners metric sets a real bar, "at least 20 minutes or 40% of an episode." Apple also suppresses data entirely until at least five unique listeners have played something from your show in the selected window, which is why new shows see empty charts.

Spotify changed its rule in 2026. Since 11 June 2026, a Spotify play is the number of times an episode was watched or listened to for at least thirty seconds. YouTube counts a public view from the first frame of playback, a rule it confirmed on 24 August 2026, and reserves a separate creator-only engaged view for a viewer who stayed past the first frame and the initial seconds; YouTube has said directly that the widespread belief that this is a flat 30 seconds is "simply not the case" and that the duration varies by surface. The AMP Accords settle on 30 seconds of content played.

Podnews published a single episode's figures side by side, and they are worth memorising as a sanity check: 197 Apple downloads against 61 Apple listeners, those 61 listeners producing 459 Apple plays, and 32 Spotify downloads against 22 Spotify listeners. Podnews's own advice on the Apple play count is to ignore it, and its summary of the broader problem is that IAB download data is "not comparable to 'play' numbers" because downloads may happen automatically.

What changed in the 2026 rewrite

Version 2.3 did more than tidy the windowing language. It pulled video into scope, specifically video distributed through open RSS feeds or by server-side file delivery inside podcast apps, and retired the word "listener" in favour of "podcast consumer" to cover viewers as well. HLS is classed as a form of progressive download, and MP4 and HLS episodes fall under the same filters and definitions as audio unless the text says otherwise. There is no separate video threshold in the spec.

Individual hosts have gone further on their own. Buzzsprout's help documentation says an audio listen has to pass 60 seconds to register as an IAB-certified download, while a video counts from one second of playback and is logged as a video download rather than an audio one, and the page does not state whether those video downloads are IAB-certified.

The rewrite also added passages on the URL prefix redirect method and on what happens to measurement when an enclosure URL changes. If a move between providers is anywhere on your list, read those alongside what moves with you when you switch podcast hosts, because a mishandled redirect shows up in your stats before it shows up anywhere else.

The number to put in front of a buyer

Downloads remain the currency in sponsor conversations, for the straightforward reason that they are the only metric with a published standard, an audit trail and a version history behind them. Quote the download figure, name the window you measured over, and say which provider produced it. Our guide to getting podcast sponsors covers how buyers read those numbers, and how many podcast downloads is good puts them against the benchmarks for shows of different sizes.

Questions hosts ask about this

Does an automatic download count? Yes, if the app pulled enough of the file to play for a minute and the request cleared the filters. This is exactly why download totals run ahead of play totals, and why Podnews calls the two figures non-comparable.

Is a 30-second listen a download? Not under the IAB standard, which asks for one minute of content. It can still be a play on Spotify, whose threshold is thirty seconds, and it can be an Apple play, which triggers above zero seconds.

Why does Apple show far more downloads than listeners? Downloads count requests that survived the one-minute and dedup rules; Apple's Listeners counts people. The Podnews sample episode showed 197 downloads against 61 listeners, a ratio in the normal range rather than a fault.

Do three devices in one house count as three downloads? If they present different user agents, yes. The guidelines pair IP address with user agent, and the spec's worked example counts six user agents behind one IP as six listeners and six downloads.

Are video views downloads? Under version 2.3, video delivered over an open RSS feed or by server-side delivery in a podcast app is in scope and HLS is treated as progressive download. A YouTube view is a different measure under YouTube's own first-frame rule and does not enter an IAB download total.

Which 24 hours does my host use? Ask. Calendar day, first touch and last touch are all permitted under version 2.3, and your provider has to disclose its choice in its document of methodology.

Before you go

Listing a show on PodAnswer costs nothing and gives it a page that can surface in search for its own name. Hosts who want the question pages built out, one answer article at a time from their own transcripts, can see how that works on the page for podcasters and compare the Growth and Network plans on the pricing page.

Sources: IAB Tech Lab, Podcast Technical Measurement Guidelines (v2.3, 29 September 2026, and the v2.2 text); Apple Podcasts for Creators listener analytics documentation; Buzzsprout help documentation; Podnews.