The AMP Accords Are Here
Today the Alliance for Measurement in Podcasting (AMP) published its AMP Accords: three ratified proposals answering what a podcast is, how exposure should be measured, and how advertising should be attributed. We've read them closely. We endorse them. Here's a summary, our take, and nine things we'd add.
A Summary of the AMP Accords
Quick background on what the AMP Accords are and where they came from.
AMP was a twelve-member task force convened in July 2025, drawing from across the industry to include advertisers/agency input, as well as hosting platforms, publishers, attribution, and talent representation. Members met monthly for a year, participated as individual industry experts rather than company representatives, voted anonymously, and dissolved once the work was done. It was a small group built to move fast, which means most of the industry is weighing in now.
The AMP Accords propose three things.
They define a podcast. If it's on-demand and it works with your eyes closed, it's a podcast. Video counts, RSS counts, neither is required.
They measure podcast consumption instead of delivery. Four new metrics, built to sit alongside downloads rather than replace them:
- Play: at least 30 consecutive seconds consumed, audio or video
- Audience: unique people who played the episode
- Ad Impression: any portion of an ad actually played for a user
- Ad Audience: unique people who heard the ad
The difference between these four and a download is huge. It's comparing a file transfer to a human pressing play.
And they rebuild attribution on those consumed impressions. This means richer identifiers shared with listener consent, plus randomized holdout groups as a standard feature of podcast ad serving. That's the same incrementality testing advertisers already expect from Meta, Google, and Amazon.
All of this would roll out in phases. New metrics get reported alongside downloads starting now, no pricing changes until 2027, and play-based currency becomes the goal for the back half of 2027 and beyond.
An Important Caveat
Note that none of this is binding. These are suggestions, but suggestions worth taking seriously.
The AMP task force has dissolved and handed its work to the IAB. And just two days before AMP released its recommendations, the IAB Tech Lab opened its own Podcast Technical Measurement Guidelines v2.3 for public comment. Coincidence?
Either way, v2.3 is a proposed update to server-side download counting, now extended to cover video. Consumption isn't in it. So in the same week, one document doubled down on measuring delivery while another proposed measuring what people actually play.
Both are open for input right now, with v2.3 comments closing on August 19. What happens next is up to the market. Which is why we're writing this.
Below, we'll go deeper on each of the three, add context where it helps, and offer suggestions where we think the Accords can go even further.
What Is A Podcast?
pod·cast noun
An on-demand show rooted in the spoken word and fully intelligible as audio. Typically episodic and conversational, podcasts cover wide-ranging themes and formats. They are accessed as audio or video via open RSS feeds and other distribution methods.
Shorthand: If it works with your eyes closed, it's a podcast.
— The proposed definition of a podcast, as ratified by the AMP Task Force, May 2026
This "on-demand and eyes-closed listening" test for "what is a podcast?" will actually get used. Most industry definitions are complicated and muddy. This one is simple enough to settle an argument in the middle of a media plan. It isn't a perfect taxonomy; read literally, it sweeps in audiobooks and lecture archives (which, to be fair, could also be released via RSS). But this is the most accurate definition buyers and sellers can apply in real time, and it beats a technically pure, jargon-heavy one that nobody remembers.
Evolving Metrics
AMP's Four Proposed Exposure Metrics
Play. A user consumes at least 30 consecutive seconds of a podcast, whether audio or video.
Audience. The number of unique users who played a podcast at least once.
Ad Impression. Counted when any portion of an ad airs for a user.
Ad Audience. The number of unique users exposed to an ad impression at least once. Also known as reach.
— Ratified by the AMP Task Force, February 2026
AMP's transition plan matters as much as the proposed new metrics. It runs in four phases. First comes shadow reporting: plays get reported alongside downloads, nothing changes on price, and everyone just watches. Then education and recalibration, where sales teams get the context right to explain that it's the same audience, with better measurement. Then there's an opt-in window where publishers can start transacting on plays for new campaigns if they're ready. And if the market goes along, plays become the default currency in the back half of 2027, with third-party verification behind them.
The sequencing logic is solid. Running plays alongside downloads with no pricing changes gives everyone time to compare the two numbers, figure out how they differ, and adjust planning models before any rates adjust. There will be new framing and an adjusted mindset to selling some shows. The good news is that nobody gets ambushed by a smaller (but more credible) number, and nobody has to renegotiate a deal on data they've had for a week.
Anyone who lived through iOS 17 knows what happens when the industry skips that step.
Lessons Learned from iOS 17
That lesson isn't secondhand for us. The iOS 17-corrected "phantom impressions" (a term we coined for overserved downloads that never became listens) surfaced in part because our team at ADOPTER, Adam McNeil and Shane Estrada in particular, refused to accept "DAI just underperforms" as an answer. That pattern led us to audit individual ads and work with Podscribe on time-of-insertion verification, confirming that ads were actually inserted when episodes were released, and then to go further and request raw server logs.
Working sessions with Pete Birsinger, whose implementation guide now anchors the AMP Accord's exposure metrics, validated the underperformance by uncovering overserved impressions (sometimes 300+ to the same IP!). Cross-advertiser validation followed through Podscribe, and Sounds Profitable carried the findings forward with Apple, which fixed the issue with iOS 17. We were one of several threads that, pulled together, solved the problem. Nobody in that chain could have done it alone. We also realized quickly that any correction would cause short-term pain for the industry, but ultimately improve its long-term health. That's why I don't read this framework as aspirational. It formalizes how this industry looks inward and fixes what's broken beneath the surface, even during boom times.
What We'd Add to the AMP Accords
One more piece of important context before the list, since several of these touch the attribution proposal:
AMP's Proposed Three Attribution Pillars
Pillar One: Attribution should be consumption-based, not download-based, using the Ad Impression definition above.
Pillar Two: Platforms share chunks of consumption data, including identifiers like IP, mobile ad IDs, and hashed emails. Anything deemed PII (Personally Identifiable Information) passes only with user consent.
Pillar Three: Ad servers offer standardized, randomized user-level holdout groups for incrementality measurement.
For baked-in ads, where nothing can be randomly withheld, the proposed workaround uses a prior episode's audience as the control group, with overlapping listeners excluded.
— Ratified by the AMP Task Force, April 2026
On the Metrics
1. Ad completion is the next metric.
Counting an impression when "any portion" of an ad airs is good, but for a 60- to 120-second host-read endorsement, the difference between two seconds of exposure and full delivery is major. The chunk-level data in the guide carries most of what's needed to report ad completion. It's not perfect; completion still requires exact ad boundaries and rules for skips and scrubbing, and baked-in ads make even boundary identification hard. But the foundation is in the spec, and completion makes this much more substantive.
2. Don't let measurability sabotage the embedded host-read.
The framework leans toward dynamic insertion, understandably, since that's where ad servers, holdout groups, and per-campaign controls live. DAI is also still imperfect and subject to human error when implementing it. The prior-episode holdout method for baked-in ads deserves credit for existing at all, but comparing two different episodes' audiences doesn't account for variance based on episode topics, guests, release timing, and news cycles. It's a useful estimate, but it won't have the same accuracy as a DAI holdout. My concern is that if embedded inventory stays more difficult to measure than DAI, the market will drift further toward DAI. After more than a decade of buying for direct response brands, I can tell you that there's a world of difference in audience engagement between running in the first batch of impressions on a new episode and the second batch, and, for full catalog buys, the impression capping and repetitive ads still leave a lot to be desired. The embedded host endorsement remains the most trusted, best-performing ad format we place. These standards should elevate it, not orphan it.
On Adoption and Platforms
3. Player participation decides whether any of this happens.
AMP's Proposed Timeline
By December 31st, 2026, all major players will be working towards supporting at least one integration option. A "major player" is any podcast player accruing over 1% of all podcast plays.
— AMP Exposure Metrics Implementation Guide
That milestone needs more of an enforcement mechanism. Advertisers, agencies, and networks can give preferential planning weight to compliant inventory, which turns participation into a revenue question. We will need to reach the tipping point where the technical and privacy calculus makes sense, and the podcast players don't have the same incentives that networks do. To that end…
4. Closed platforms are further from compliance than they look, so let's be precise about the interim.
Start with the open ecosystem, where hosting platforms are closest. The HLS chunking pathway gets them most of the way there using infrastructure they already operate, but it isn't automatic. HLS files still get auto-downloaded (Apple Podcasts does it by default for the audio version), so anything fetched faster than realtime has to stay out of the play count. Those fetches don't get thrown away, though. They're still downloads. They're just not plays. And Podnews has documented that HLS requests only resolve to about 60-second granularity, short of the spec's 30-second chunks, which means full chunk fidelity will still depend on the players reporting from their end.
YouTube is a different story, and if we can't get their buy-in to offer up a lot more data, it's going to be a matter of going channel by channel. YouTube's channel insights sharing setting (off by default) and Apple Podcasts Connect's creator analytics are pathways to real data, but it's aggregated data, which includes audience composition, geography, watch time, and consumption spikes. Useful for planning, but it's not the event-level consumption logs AMP specifies, so it's not AMP compliance.
Genuine compliance from closed platforms will require new APIs or platform-run measurement products that aren't open (yet?).
In the meantime, there are two practical moves. First, make channel-level sharing of what exists a standard term of sponsorship onboarding and insertion orders, the way pixel placement became routine.
Second, build neutral hubs that aggregate those opt-in streams. Attribution providers with existing cross-channel dashboards, Podscribe among them (see my disclosure below), are positioned for that role. We can't build an ecosystem on screenshots of audience data.
On Pricing and Reporting
5. On podcast ad pricing, the math will change more than the money.
For the majority of undersold shows with open inventory, holding current podcast advertising CPMs while moving the denominator from downloads to plays is sensible, and it should deliver more dollars to underserved podcasts whose engagement doesn't match their impression counts. At the top of the market, however, where premium inventory commands the majority of brand and DR spend, I suspect rate cards evolve toward price-per-episode as the quoted figure, with plays and downloads/impressions published alongside it. The effective CPM becomes a derived number with a choice of denominator, not the price itself. There's also a scenario worth preparing for, where playthrough rates for tracked shows land well above the worst-case assumptions, and hit 75 or 80 percent and higher. If shadow reporting confirms anything close to that (market chatter and early insights suggest it might), the feared reckoning becomes a modest recalibration, and Phase 1 turns into evidence that engagement is stronger than the industry feared.
Either way, direct response advertisers can hold steady. If the episode price doesn't change and conversions don't change, CPA and ROAS won't change. The thing actually worth watching through the transition is whether guarantees, makegoods, and measurement windows stay stable while the reporting underneath them shifts.
6. Report addressable audience share, because an honest approximate baseline of addressable audience beats a modeled guess.
If we're going to standardize counting plays over impressions, geography belongs in the same reporting, since it's an audience attribute that advertisers and buyers reference frequently in decision-making. What brands and buyers need is simpler: what share of this show's plays happen in the market I'm actually selling in? Dynamic insertion already answers this through targeting; a U.S.-only DAI buy delivers what it says. Pixel-tracking mostly solves it for audio, even on episodic and baked-in inventory. But YouTube, where a channel's audience routinely skews far more international than the same show's RSS feed, leaves buyers guessing. The framework for connecting channel-level data carries what's needed. Standardizing this will give sellers a fairer basis for pricing geographic reach instead of watching it get rolled up wholesale, and removes one more modeling layer from video measurement.
On the Handoff
7. Equip the buy side, not just the sales side.
Phase 2 of the transition framework arms sales teams with talking points and pitch tools. The harder conversation happens on the other side of the table, when an agency has to explain to a CMO or CFO why reported measurement evolved while the audience, the content, and the results never changed. Agencies and Chief Audio Officers (CAOs) will be the shock absorbers of this transition, and they need the same toolkit aimed in the other direction: benchmark download-to-play deltas by category and platform, planning-model adjustments, and language that survives a thorough review. I'd encourage the standards bodies inheriting this work to treat buy-side education as a first-class deliverable, and to bring active buyers into the process.
8. Verification needs an owner.
"Third-party verification & audit" appears in H2 2027 as the finish line, but certification is what turns a proposal into a currency. The IAB handoff is the right move. An MRC-style audit path, with named criteria for what "AMP-compliant" means, should be scoped early rather than discovered late. Podscribe, given their involvement here, seems like a logical first "AMP-certified" third-party measurement partner.
9. Put privacy and data governance inside the standard, not beside it.
Under AMP's Proposal, The Data Players Would Share
For every 30 seconds of content played, the player sends an event in real time to the hosting platform or an analytics provider. Each event includes:
Required: A chunked timestamp range, timestamp when consumed, show identifier, episode identifier, and user-agent.
IP address: From consenting users. If the user opts out, the last digit (or octet) is zeroed (e.g. 1.2.3.0) so play stats can still be computed without any possible PII shared.
Mobile Ad ID (MAID) / GAID (Google's Advertising ID) and SHA256-hashed email: Optional, recommended for notified and opted-in users. The guide notes these two "will significantly improve advertiser confidence in attribution, planning tools, and help support growth of budgets."
— AMP Exposure Metrics Implementation Guide
Many would consider this addition critical before any handoff. The implementation guide asks players to share event-level listening logs alongside IPs, mobile ad IDs, and hashed emails. It places the consent work almost entirely on the players. Participation should flow from podcasters opting their shows in, with networks facilitating that opt-in at scale, the same way they operationalized prefixes and pixels. And listeners need a simple, working MAID-level opt-out that's honored everywhere downstream. The spec's truncated-IP fallback for opted-out users is the right instinct; it should be the guaranteed floor, not a suggestion.
Some history matters here. Podcast attribution settled at the current household IP level for a reason: simple matching, no individual identity, no profiling. It passed the privacy sniff test then, and it still does. Those of us who were buying when pixel attribution for podcasts was new remember how jumpy device-level identifiers made this industry, and that caution served us well. Nobody voiced that caution louder, or longer, than Todd Cochrane, who we lost last September. Writing this section, I remember the spirited privacy debates we had when Podsights was new. What AMP proposes, MAIDs/GAIDs and hashed emails riding alongside listening logs, is a genuine escalation in identifiability from where podcasting operates today. Better measurement justifies considering it, not rushing it. That step up in exposure has to be taken with the same care that kept this medium out of the privacy headlines other channels brought upon themselves.
The standard itself, not just the contracts around it, should specify who may receive this data, for which purposes, how long it can be retained, and whether it can ever be joined to broader identity graphs. It should also require reporting of consent coverage, because an audience figure built only from opted-in shows and listeners describes the people you could measure, and advertisers and buyers deserve to know how large that subset is before treating it as "the audience." To be clear, I very much want this data to exist. I want it governed well enough that regulators, platforms, listeners, and creators can all live with it, because podcasting has spent 20+ years earning a kind of listener trust that other channels burned through, and the measurement upgrade shouldn't be the thing that spends it.
Disclosure
One disclosure for the record: I serve in an advisory capacity at Podscribe, which is named in the document's adoption tier. The views here are our own, and they reflect what a decade of buying host-read endorsement campaigns for direct response brands tells us advertisers actually need.
A Big Thanks
The AMP task force deserves credit. Getting twelve people from competing corners of the industry to keep talking for a year and finish with three ratified proposals is much harder than it sounds. What kept it alive, as far as I can tell, was the constructive-counter rule: a no vote came with homework. You couldn't just object; you had to be ready to propose something better. That rule should outlive the task force. My sincere thanks to every member who gave their time, and to Dan Granger for convening the group and deciding the wait was over.
An Endorsement
ADOPTER Media supports the AMP Accords as an excellent first step, and we're prepared to participate in the next steps, wherever we can be useful. The document itself admits that nobody in the room got everything they wanted. Neither did I. But this is a workable start.
Glenn Rubenstein
Glenn Rubenstein is the Founder and CEO of ADOPTER Media and author of "Podcast Advertising Works," the first book on podcast advertising. Since 2007, he's helped brands engage online audiences through authentic advertising.
