How to read podcast downloads without lying to yourself

Most serious hosts and advertisers mean a download the way IAB Tech Lab defines it. The last clearly finished spec is Podcast Measurement v2.2. A v2.3 draft went out for comment in July 2026. Do not treat that draft as the certified rulebook yet.

In practice, a download is a real request for the episode file after junk has been filtered out, usually with about a minute of audio actually sent. Players often fetch a file in tiny chunks, including two-byte probes that are just checking whether the server works. Those are not listens. A log that counts every chunk as a play will inflate you.

The count comes from the server, not from the app confirming that someone sat through the episode. A person can download a file and never hit play. A person can listen while the file is still arriving. Both can still count. That is already enough to stop treating the number as attendance.

A host's unique audience is generally an estimate based on IP address and user-agent information within a stated time window. It is not a verified headcount or necessarily a persistent browser fingerprint. Shared networks and multiple devices can make that estimate differ from the number of people who actually listened. Check the provider's published method and reporting window.

Your host counts from its logs. Apple does not send you Apple's numbers. Spotify's in-app graph is Spotify's world. YouTube views are YouTube. None of these have to use the same filters.

Hosts that follow IAB rules throw out known bots, crawlers, and traffic from data centers. The sneakier fake traffic is supposed to look human, and IAB does not publish those filters. Hosts that skip this work will look bigger. Bigger is not better. It is noisier.

Buzzsprout is certified to IAB v2.2. If you typed in an all-time total from a previous host, that imported number is not certified. YouTube views sit in a separate field because YouTube tracks views, not downloads.

Transistor says it follows the same rules. It does not pay for the annual certification.

When you move hosts, the new dashboard starts at zero. The old all-time number stays behind. If a chart says 20,000 and 19,000 of those were a manual add, you do not have 20,000 listeners.

OP3, the Open Podcast Prefix Project, is a prefix you put in front of episode URLs. It measures in the open. No login. The stats page is public on purpose. OP3 is not IAB-certified. Anyone can see the same numbers, including a sponsor, a competitor, or a nosy listener.

The prefix sees the request. It does not always see how many bytes the host later served. Adding a prefix can also make apps treat the file as new and download it again, so the first days can spike.

Use OP3 as a second ruler, not as a secret. If OP3 and your host wildly disagree, one of them is counting junk, or the prefix is not on every file, or an old host is still serving episodes.

OP3's own site, including how it calculates a download, is the source. Do not put a prefix on a show if you are not willing to publish the numbers.

YouTube public views are not RSS downloads. Different platform, different fraud, different unit. Do not add them together.

A Spotify Follow is not an RSS subscriber count. That graph does not travel with the feed when you leave.

An email list is fine for a newsletter. It is not a download.

A spike the day you turned on a prefix or a redirect is often crawlers and apps refetching. Wait a week before you tell a sponsor the new average.

Pick one window and one source. Last 30 days from the host, IAB if they certify, plus OP3 if you run it. Say which.

Do not average a launch week with a dead month and call it typical. Do not show all-time if all-time includes a host import you cannot defend.

For a sponsor, downloads per episode in a recent window beats a lifetime vanity total. If the show is small, say the small number. Inflating it is how you lose the next deal.