Methodology

How every number is made.

Attribution you can't interrogate is attribution you shouldn't trust. This page explains exactly how PodAttr counts, matches, and reports — including what our numbers can't tell you. Argue with it, test it, replicate it.

What we count How matching works Attribution windows Honest limits What we collect The worst case, named Incident covenant Standards

What we count: eligible downloads

When your podcast feed carries the PodAttr prefix, each episode request routes through p.podattr.com, is logged, and is redirected to your audio file — adding a single redirect hop, the same mechanism used by Podtrac, Spotify's prefix, Swap.fm, and other measurement services. If you sponsor-read for any large show, your feed almost certainly carries two or three of these already, stacked and coexisting.

A logged request becomes an eligible download only after filtering:

Invalid traffic removed. Requests from known bots, crawlers, monitors, and clients presenting no user agent are logged but excluded from every reported figure.
Household de-duplication at match time. Podcast apps routinely make multiple requests per episode (streaming ranges, retries, prefetch). Matching operates on households, not raw request counts, so app chattiness cannot inflate attribution.

A download is a server-side event. It tells us an episode was delivered to a household's device. It cannot confirm a human pressed play, listened to a specific minute, or heard the ad — no server-side podcast measurement can, and vendors who imply otherwise are overclaiming. We say "download," and we mean exactly that.

How matching works: households, not people

PodAttr never identifies individuals. Both sides of the match are reduced to a household key: the network portion of the connecting IP address (a /24 for IPv4, a /48 for IPv6) passed through a salted one-way hash. The raw IP is never stored — not on the download side, not on the conversion side.

When a household that downloaded your episode later appears in your sponsor's conversion events (via the sponsor's pixel), within the campaign's attribution window, that conversion is attributed to your show. Where several qualifying downloads exist, we use last-touch: the most recent eligible download before the conversion. Each conversion event is attributed at most once.

Attribution windows

The default lookback window is 30 days from download to conversion, adjustable per campaign. We chose 30 days because host-read podcast ads act more like recommendations than clickable banners: listeners convert hours to weeks later, on a different device, often via search. Every report states the window used, and time-from-download-to-conversion is shown on the report itself so you and your sponsor can see the distribution rather than take the window on faith.

Honest limits: where household matching over- and under-counts

Household-level matching is the accepted industry approach for podcast attribution, and it is directional by nature. We'd rather you hear its error sources from us:

Shared networks over-count. Dorms, offices, and public Wi-Fi put many people behind one household key. A colleague of a listener who converts can be matched. This inflates counts in dense networks.
Carrier networks under-count. Mobile carriers rotate and share addresses (CGNAT), and privacy relays mask them. A listener who downloads on cellular and converts on home Wi-Fi may never match. This loses real conversions.
Downloads aren't listens. Auto-downloaded episodes that were never played still count as exposure. This is a structural limit of server-side measurement, not a PodAttr choice.
Correlation, not proof of causation. A matched conversion is strong evidence, not a controlled experiment. Some matched buyers would have purchased anyway. Incrementality testing (exposed vs. holdout) is the rigorous answer to causation and is not something PodAttr currently performs.
Post-conversion events are invisible to us. Refunds, cancellations, and duplicate orders are not netted out unless your sponsor's integration reports them.

Our position: these errors partially offset, the numbers are honest and directionally reliable, and they are categorically better evidence than promo-code redemptions — which industry studies presented at Podcast Movement have shown to be inaccurate at best, since most converting listeners never type the code. We label our reports accordingly, and we will never present a matched count as a laboratory result.

What we collect — exactly

From a podcast download request: the salted household hash, the episode URL, the user-agent string (needed for invalid-traffic filtering), and a timestamp. From a sponsor's pixel: the salted household hash, the event type (visit, signup, purchase), an optional order value, the page path, and a timestamp. That is the complete list.

No raw IP addresses are ever stored, on either side.
No cookies are required for matching. No advertising identifiers, no fingerprinting, no cross-site profiles.
No names, emails, or customer details from your sponsor's checkout — only pseudonymous conversion events.
Data is processed and stored on Cloudflare's network (primary region: Western North America).
Podcasters can export every row of their data as CSV, and removing the prefix ends collection instantly. See removal.

Full details: privacy policy · terms.

The worst case, named

You should know precisely what happens if PodAttr fails, because your feed routes through us:

If our logging fails — database trouble, a bug, a bad deploy — your audio still plays. The redirect is designed to fail open: the prefix answers with the redirect first and logs asynchronously, and every logging error is swallowed rather than surfaced to the listener. You lose telemetry for the outage window; your listeners lose nothing.
The only scenario that stops downloads is our edge network failing to answer at all. The prefix runs on Cloudflare's edge — the same infrastructure fronting a substantial share of the web — and our uptime is public, permanently, at our status page.
And if we ever disappear — the fate of this industry's last independent tool — our continuity commitment applies: 12 months' minimum notice, your data exportable, and the prefix worker open-source so your tracking can outlive us entirely.

The incident covenant

If an incident ever affects episode downloads, we commit in writing to:

Acknowledge it on the public status page within 30 minutes of detection.
Publish a plain-language post-mortem within 48 hours.
Include one-click removal instructions in the incident alert itself — your fastest exit, offered at our worst moment.

Standards alignment

PodAttr's counting is designed against the IAB Tech Lab Podcast Technical Measurement Guidelines v2.3 (draft, released July 21, 2026), which specifically addresses URL-prefix measurement, enclosure URL changes, invalid traffic, and measurement window approaches. We use the draft's framework — eligible downloads, invalid-traffic filtering, defined measurement windows — and are participating in its public comment period. This describes design alignment with a draft standard, not a completed certification; we'll state it here plainly if and when that changes.

Questions, disagreements, or corrections to this methodology are genuinely welcome: ceo@podattr.com. Last updated July 23, 2026.