Audio & Podcast Advertising Standards.
The audio execution layer: VAST 4.x audio delivery, the DAAST legacy, the IAB Podcast Measurement Technical Guidelines, dynamic ad insertion, downloads and listener estimates, brand suitability, attribution — and what all of it means when agents do the buying.
This is a deep dive inside the six-layer standards system, not a seventh pillar. Audio transacts on shared rails — VAST has carried audio since version 4.1 — but podcasting runs on a distribution model the web never had: files downloaded through open RSS feeds, measured from server logs, with no standard playback telemetry. This page maps what is actually standardized, what is guidance, and what is honestly a modeled estimate.
In podcasting, delivery is not the same as listening — and listening is not the same as outcome.
Fast read
- What it is
- A deep-dive guide to the standards that deliver and measure audio and podcast advertising — and to the gap between what server logs can prove and what buyers want to know.
- What it covers
- VAST 4.x audio delivery and the DAAST legacy, the IAB Podcast Measurement Technical Guidelines v2.2, dynamic ad insertion, downloads and listener estimates, the voluntary compliance program, streaming-versus-podcast differences, brand suitability, and attribution.
- What it is not
- Not a seventh standards pillar — a deep dive inside the six-layer system. And not a claim that any download count, pixel, or lift study proves listening or outcomes on its own.
- Why it matters
- Agentic buying can plan an audio flight in seconds. What the channel can actually prove — a qualified download, a filtered log line, a modeled listener — is decided by this layer.
- Best for
- AdTech, MarTech, media, agency, publisher, podcast, and brand leaders building or evaluating audio and podcast execution in agentic workflows.
- Best next read
- Measurement & Media Quality, Research & Measurement Science, and the Audio playbook.
Where audio fits in the standards stack.
Audio is a child of the six-layer system with its own center of gravity. Delivery rides the Core AdTech rails — VAST 4.1 and above carries audio via the adType attribute, and DAAST, audio’s own former template, is officially deprecated with implementers directed to VAST. Measurement is where audio diverges: the trust layer’s client-side tooling does not transfer to RSS distribution, so the central artifact of this page is a server-log standard — the Podcast Measurement Technical Guidelines. Privacy runs on request-level signals rather than consent strings in the feed. This page is the close-up of that intersection.
Why audio and podcast are different.
Every audio-specific guideline exists because a web assumption broke at the feed. Eight realities define the channel.
- 01
Delivery is a file, not a session
RSS podcast delivery is an enclosure download: a client requests a file and the server watches bytes leave. The episode may be heard hours later, in part, or never — and the serving infrastructure cannot tell the difference.
- 02
A download is not a listen
There is no standard playback telemetry in the RSS path. The measurement guidelines define delivery events precisely so the industry can count something honest — not because the count means listening.
- 03
The substrate is a server log
Podcast measurement per the IAB Tech Lab guidelines is server-log-based: downloads, listeners, and ad delivery derived from filtered request logs. The web’s client-side measurement rails are not the default here.
- 04
The listener is an estimate
A “listener” under the guidelines is the unique combination of IP address and user agent, with IPv6 truncated to the first 64 bits. That is a statistical proxy for a device — not an identified person.
- 05
Ads arrive two ways
The guidelines distinguish integrated (“baked-in”) ads recorded into the file, dynamically inserted ads chosen at file-request time, and sponsorship reads. Each carries different targeting, measurement, and suitability physics.
- 06
There is no cookie in the feed
An RSS request presents no cookie, device ID, or consent string — the decision inputs are request-level: IP, user agent, time. Identity-based targeting logic imported from the web does not transfer.
- 07
Audio rides shared rails
Audio’s own serving template, DAAST, is officially deprecated; per IAB Tech Lab, implementers should use VAST 4.1 or above, which designates audio via the adType attribute. Audio delivery is a tenant on video’s rails.
- 08
Attribution is mostly modeled
Promo codes undercount, pixels are not standard in RSS delivery, IP matching is a lossy join, and lift studies ride on design choices. Outcome claims in audio carry a methodology, or they carry doubt.
Ad delivery standards.
Four things carry audio delivery: the VAST rail that audio now rides, the deprecated template it replaced, the insertion technique that makes podcast inventory decisioned, and the feed layer where all of it is actually implemented.
- The current rail
VAST 4.x for audio
VAST 4.1 (released November 2018) integrated audio ad serving: an adType attribute on the Ad element designates video, audio, or hybrid (default video); DAAST’s Expires concept was ported in; and the official announcement describes every node of the 4.1 spec as clarified for audio versus video applicability. The current VAST version per the official standards page is 4.3 (December 2022) — public documentation attributes 4.2’s headline change to SIMID, not audio, so validate rather than assume audio behavior changed after 4.1.
- Deprecated — legacy
DAAST
The Digital Audio Ad Serving Template is officially retired. The IAB Tech Lab page — now under its archived standards path — states verbatim: “Digital Audio Ad Serving Template (DAAST) has been deprecated, and is being replaced by VAST. Please use VAST 4.1 or above.” DAAST stopped at the 1.x line (1.0 in 2014; a 1.1 document sits in the archive). Treat any DAAST dependency as technical debt to remove, not a capability to plan around.
- Server-side, request-time
Dynamic Ad Insertion
There is no standalone IAB DAI specification for podcasts. The official treatment lives inside the Podcast Measurement Technical Guidelines v2.2, Section 4.3: dynamically inserted ads are “targeted and dynamically inserted at the time of file request (rather than recorded directly into the audio file)”, with the ad server choosing the ad at request time. DAI is what makes podcast inventory decisioned — and what makes episode-level context a moving target.
- The distribution layer
Feeds and hosting
Podcast distribution runs on RSS enclosures served by hosting platforms — and IAB publishes no podcast RSS or feed-level technical specification. The feed and hosting layer is where DAI decisions, measurement logging, and filtering are actually implemented, which makes it implementation-sensitive: two hosts can serve the same episode and count it differently unless both follow the measurement guidelines.
The precision point
There is no “VAST for podcasts” specification and no standalone IAB DAI spec. VAST 4.1+ carries audio ad serving; dynamic insertion is defined inside the measurement guidelines. Saying it precisely is the first capability check.
Podcast measurement guidelines.
The IAB Podcast Measurement Technical Guidelines are the channel’s honest core: a server-log standard that defines what may be counted, what must be filtered, and what a “listener” can even mean when the only witnesses are request logs. The current version is 2.2 (May 2024), with a voluntary, audited compliance program alongside it.
- The central artifact
PTMG v2.2
The IAB Podcast Measurement Technical Guidelines define server-log-based measurement of downloads, listeners, and ad delivery. The current version is 2.2 (May 2024), published by IAB Tech Lab and developed by the Podcast Technical Working Group in partnership with the IAB Audio Committee. The lineage runs 1.0 (September 2016), 2.0 (December 2017), 2.1 (2021), 2.2 (May 2024) — a decade of converging on what a download honestly is.
- The mechanics
Downloads, listeners, filtering
A qualified download is a GET 200 that served at least one minute of content (ID3 and header bytes excluded), de-duplicated within a 24-hour window — fixed calendar day or rolling. Partial (206) responses count only if the one-minute rule is met with IP + user-agent de-duplication; HEAD and 304 are excluded. A listener is the unique IP + user-agent combination, IPv6 truncated to the first 64 bits. Filtering removes data-center and VPN IPs, self-declared bots, malformed user agents, unrealistic volumes, two-byte range probes, and Apple watchOS duplicates — with inclusion lists re-validated at least every 90 days.
- Voluntary, audited
The compliance program
IAB Tech Lab runs a Podcast Measurement Compliance certification program (in market since December 2018; NPR and RawVoice/Blubrry were first to certify). Per the Compliance Certification Guide v2.3 (April 2024): it is voluntary, runs “an in-depth multistep audit of individual implementations” via third-party auditors engaged by Tech Lab, carries an annual fee (no public amounts), and requires annual recertification — even with no changes, and at most one major guideline version behind. Eligibility covers platforms that use or distribute server-side measurement reporting; Tech Lab membership is not required, and the program is global. Certification standardizes counting — it does not turn downloads into listens.
| Metric | What it means | Watch-out |
|---|---|---|
| Download (qualified) | A GET 200 request that served at least one minute of content (ID3/header bytes excluded), de-duplicated by IP + user agent within a 24-hour window | It proves delivery of bytes — not playback, attention, or completion |
| Listener | The unique combination of IP address + user agent, with IPv6 truncated to the first 64 bits | A statistical proxy for a device — shared IPs, VPNs, and app updates distort it in both directions |
| Ad delivered | Ad bytes served within a qualified download — chosen at request time when dynamically inserted | Delivered is not heard: skips and abandons are invisible to server logs |
| Partial download (206) | Counted only when the one-minute threshold is met, with IP + user-agent de-duplication; HEAD and 304 excluded | Mishandled range requests are a classic inflation source — two-byte probe requests must be filtered out |
| Filtered traffic | Removal of data-center/VPN IPs, self-declared bots, malformed user agents, unrealistic 24-hour volumes, and watchOS duplicates; lists re-validated at least every 90 days | Filtering is list- and rule-based; the widely used player user-agent list is community-maintained (OPAWG), not an IAB product |
| Certified measurement | A platform passed a third-party audit of its server-side measurement against the current guidelines — voluntary, with annual recertification | A counting credential, not an outcome claim: certification does not validate listening, attention, or results |
Streaming audio vs podcast.
“Audio” names two different machines. Streaming platforms control playback and can observe it — under their own definitions, where supported. RSS podcasting controls neither playback nor the player, which is exactly why its measurement standard counts downloads. Conflating the two produces dashboards that compare a playback event to a file transfer.
| Dimension | Streaming audio | Podcasts (RSS) |
|---|---|---|
| Delivery model | Session-based streams in apps and web players, with ads requested around playback | File downloads via RSS enclosures — possibly long before, or without, listening |
| Ad decisioning | At or near play time, with VAST 4.x audio tags where supported | Baked-in at production, or dynamically inserted at file-request time per PTMG Section 4 |
| Measurement substrate | Playback-side events where the platform reports them — implementation-sensitive | Server logs: qualified downloads, filtered and de-duplicated per the guidelines |
| Listening confirmation | Possible where the player reports playback — definitions vary by platform | No standard playback telemetry in the RSS path: a download is not a listen |
| Identity signals | Logged-in or session context where supported, governed by consent rules | IP + user agent at the request — no cookies, device IDs, or consent strings in the feed |
| What buyers should ask | Which playback events exist, who reports them, and under what definitions | Whether counting follows PTMG v2.2 — and how DAI, filtering, and caching windows are implemented |
The discipline
Report streaming and RSS podcast metrics in separate columns, under their own definitions. A blended “audio impressions” number hides the fact that one column was observed and the other was served.
Brand safety and suitability.
Audio’s context is not a URL — it is an hour of human conversation. That makes suitability a content-understanding problem, makes transcripts the working substrate, and makes every vendor verdict implementation-sensitive. The adjacent disciplines — contextual intelligence and embeddings — are how spoken-word context becomes machine-readable.
- Spoken word is context
The transcript is the unit
Audio brand suitability is a content-understanding problem: the context is what is said, by whom, in what tone, across an hour of conversation. Transcript-based analysis — where transcripts exist and rights permit — is what makes that context machine-readable at scale; without it, suitability decisions run on show titles and category labels.
- No standard verdict
Suitability is implementation-sensitive
There is no audio-specific, cross-industry technical standard that hands down suitability verdicts. Vendors apply their own taxonomies, models, and thresholds to spoken-word content — so the working questions are which taxonomy, which model, what coverage of the catalog, and who audits the edge cases.
- The adjacency problem
Episode-level beats show-level
Dynamic insertion means the same ad can land across thousands of episodes — including back-catalog episodes recorded long before the campaign. Suitability assessed at show level misses episode-level context, and back-catalog insertion places today’s ad into yesterday’s conversation. Assess at the level the insertion actually happens.
Audio attribution and outcome measurement.
No standards body hands audio an attribution protocol. What exists is a set of paths with different confidence levels: promo codes and vanity URLs that undercount, pixels that fire only where playback contexts support them, modeled household matching, and lift studies whose estimates ride on design choices. The honest practice is to label each path for what it is — and to keep outcome claims inside what the evidence chain supports.
- Deterministic-ish
Promo codes and vanity URLs
Codes redeemed at checkout and dedicated landing paths are the channel’s classic proof — and both undercount. Listeners convert without the code, search the brand instead of typing the URL, and buy on another device. Treat these as a floor on response, not a measure of it.
- Where available
Pixels and web playback
Pixel-based confirmation exists where playback happens in contexts that can fire one — streaming and web players, where supported. It is not standard in RSS delivery, where the “impression” is a file leaving a server. Any vendor presenting pixel coverage as universal across podcasting is describing a different channel.
- Modeled
IP matching and lift studies
Household and device matching joins download IPs to identity graphs — a lossy, modeled join. Lift studies compare exposed and control groups, and their estimates ride on design choices: exposure definition, control construction, and window. Both are legitimate evidence when the methodology is disclosed — and neither is deterministic.
Caution
Most podcast attribution is modeled — methodology review required. Treat every outcome number as methodology-bearing evidence, not as an observation.
Agentic implications.
An agent buying audio does not buy a pageview or a session — it buys delivery into a channel where the strongest standard defines a download and the outcomes are modeled. The standards on this page are the constraint set that agentic reasoning has to satisfy. Six implications follow.
- 01
Optimize to what the metric proves
An agent optimizing podcast spend on downloads is optimizing delivery, not listening. Encode each metric’s meaning into the objective function — otherwise the workflow maximizes bytes served and reports it as audience.
- 02
The rail check is simple
Audio plans must compile to VAST 4.1-or-above audio (the adType attribute) on paths that support it, per the official deprecation direction. Any DAAST dependency in the toolchain is removal work, not a capability.
- 03
DAI inputs are request-level
At file-request time the decision sees an IP, a user agent, and a timestamp — not a person. Agents must target podcasts with request-level and content-level signals; web-style identity logic fails quietly here.
- 04
Trust counts only after the filter
Before comparing numbers across hosts, an agentic workflow should verify guideline-conformant counting — the 24-hour window, the one-minute rule, IVT filtering — or check certification status. Unfiltered counts are not comparable inputs.
- 05
Attribution needs a methodology field
Agentic reporting should carry how each number was made — code redemption, pixel where available, modeled match, lift design — alongside the number itself. A workflow that strips methodology turns estimates into assertions.
- 06
Transcripts unlock suitability at scale
Agents can read spoken word at scale — where transcripts exist and rights permit — making episode-level suitability tractable for the first time. Model-based judgments still need human-defined thresholds and an audit path.
Implementation lens.
The same standards land differently depending on where you sit in the chain. Select your company type for the version of this page that applies to you.
Require guideline-conformant counting on every podcast buy — and ask for compliance certification status where it matters. Insist on attribution methodology disclosure, keep outcome claims inside what modeled evidence supports, and set suitability expectations at the episode level, not the show label.
Build cross-host comparability into the workflow: which hosts follow which guideline version, how DAI is implemented, and what each reported metric means. Keep dashboards channel-honest — delivered and heard are different columns — and design lift studies before the flight, not after it.
Implement guideline-conformant logging and filtering, document how your DAI behaves, and treat certification as a trust asset buyers can check. Expose episode metadata and transcripts for suitability analysis where rights permit — transparency here is increasingly a sales asset.
Engineer the mechanics precisely: the one-minute rule with ID3 bytes excluded, 206/HEAD/304 handling, IP + user-agent listener logic with IPv6 truncation, IVT filtering, and 90-day list re-validation. Submit your user-agent value to the Tech Lab Spiders and Bots inclusion list, and weigh the voluntary audit.
Support VAST 4.x audio where applicable, report playback-side events under explicit published definitions, and keep consent governance in scope — streaming has the identity and telemetry signals the RSS path lacks, which makes its obligations larger, not smaller.
Carry VAST audio correctly, retire DAAST-era assumptions deliberately, and treat podcast DAI as a server-side decision with request-level inputs. Surface metric provenance to buyers — which numbers are guideline-conformant, which are platform-defined, which are modeled.
Separate deterministic-ish paths from modeled ones explicitly, publish match methodology and loss rates, design lift studies that survive review, and never present pixels as universal — they are not standard in RSS delivery. Your credibility is your methodology section.
No Fluff POV.
Audio is the channel that is most honest about its own limits — when you actually read the standards. The measurement guidelines define a download because a download is what server logs can defend; nothing in the official documentation claims it is a listen. Teams that respect that honesty get a channel with unusually clean definitions and a real, audited counting credential. Teams that paper over it ship dashboards where a file transfer impersonates attention — and agents scale the impersonation.
- Say it precisely: a download is a qualified delivery event under PTMG v2.2 — not a listen, and not an outcome.
- Use the current rail: VAST 4.1 or above for audio, per the official deprecation notice — treat DAAST as legacy to remove.
- Demand guideline-conformant counting before comparing hosts, and treat certification as a counting credential, not an outcome claim.
- Carry methodology with every attribution number: promo codes undercount, pixels are not standard in RSS, matches are modeled, lift rides on design.
- Assess suitability where insertion happens — at the episode, with transcripts where rights permit — not at the show label.
- Validate current status before building: the guideline line is 2.2 (May 2024), the compliance guide is v2.3 (April 2024), and audio delivery documentation sits in the VAST 4.x line.
The point
In podcasting, delivery is not the same as listening — and listening is not the same as outcome.
Primary sources to validate.
Audio, podcast, measurement, and platform references last validated: June 2026. Validate current official documentation before implementation.
Primary sources to validate 13 sources
- Digital Audio Ad Serving Template (DAAST) — deprecation notice ↗ Official standards page
The official DAAST page — now under the archived standards path — states verbatim: 'Digital Audio Ad Serving Template (DAAST) has been deprecated, and is being replaced by VAST. Please use VAST 4.1 or above.' Mentions DAAST versions 1.0 and 1.1 and links the VAST 4.1 audio announcement. Supports: DAAST deprecation status (official wording), DAAST version history (1.0, 1.1), VAST 4.1-or-above instruction.
- VAST 4.1 — Now With Audio Power! ↗ Official blog
Official blog explaining how VAST 4.1 absorbed audio ad serving: the separate DAAST spec 'can be deprecated to the mutual benefit of all industry constituents'; audio/video designation via the new adType attribute on the Ad element (values video, audio, hybrid; default video); DAAST's Expires concept ported into VAST 4.1; every node of the 4.1 spec clarified for audio vs video applicability. Supports: How VAST 4.1 supports audio, adType attribute (video/audio/hybrid), Rationale for retiring DAAST.
-
The final VAST 4.1 specification document; the title page states 'Released November 2018'. This is the VAST version that integrated audio ad serving from DAAST. Supports: Authoritative VAST 4.1 release date (November 2018), Canonical spec citation for VAST audio support.
- InteractiveAdvertisingBureau/vast — repository README ↗ Official GitHub
VAST repo README: VAST 4.1 released November 2018 with 'support for Ad Requests, Ad verification with Open Measurement, integration of DAAST for Audio ads'; VAST 4.2 (June 2019) added SIMID; VAST 4 (January 2016) added the mezzanine file for server-side ad insertion. Supports: 'Integration of DAAST for Audio ads' wording, VAST 4.x timeline, SIMID attribution to 4.2 (not audio).
- InteractiveAdvertisingBureau/daast — repository ↗ Official GitHub
Archive of DAAST artifacts: the DAAST 1.0.0 specification (2014), a DAAST 1.1 PDF, ad-categories spreadsheet, a VAST 3.0-to-DAAST transition schema, and XSDs. Confirms DAAST stopped at the 1.x line; the hosted 1.1 PDF is titled as a draft — final-release status of 1.1 was not verified. Supports: DAAST version artifacts (1.0, 1.1), Legacy/archive status.
- Podcast Measurement Technical Guidelines — IAB Tech Lab standards page ↗ Official standards page
Official landing page listing the current version as 2.2 (May 2024), with prior versions 2.1 (2021), 2.0 (December 2017), and 1.0 (September 2016). Links the v2.2 final PDF, the compliance guides, and the Podcast Measurement Compliance program page. Supports: Current guideline version (2.2, May 2024), Version history, Canonical download links.
-
The full v2.2 guidelines: server-log-based measurement of downloads, listeners, and ad delivery. Defines the 24-hour de-duplication window (fixed calendar day or rolling); the one-minute-of-content download threshold (ID3/header bytes excluded); HTTP status handling (count GET 200; 206 partials only if the one-minute rule is met with IP+UA de-dup; exclude HEAD and 304); IVT filtering (data centers, VPNs, bots, malformed user agents, 2-byte Range:0-1 probes); listener = unique IP address + User Agent; IPv6 truncated to the first 64 bits; inclusion lists re-validated at least every 90 days; Apple watchOS duplicate-download filtering. Section 4 distinguishes Integrated ('baked-in') Ads, Dynamically Inserted Ads, and Sponsorship Ads. Supports: Download definition and one-minute rule, 24-hour window, Listener (IP+UA) and IPv6 truncation, IVT filtering rules, Official DAI description (Section 4.3).
- Podcast Measurement Compliance — compliance program page ↗ Official standards page
Official program page: the certification certifies use of 'the common agreed-upon set of metrics for podcast content and ads' against the most up-to-date version of the Podcast Measurement Guidelines. Supports: Existence and scope of the compliance certification program.
-
Voluntary program certifying adherence to the guidelines via 'an in-depth multistep audit of individual implementations'; IAB Tech Lab engages third-party auditors. Annual fee (as of October 2022 — pricing via IAB Tech Lab, no public amounts); annual recertification required (recertify when 12 months have passed since the last certification test, even with no changes, or when more than one major guideline version behind); self-audits recorded at least twice a year. Eligibility: any podcast platform that uses or distributes server-side measurement reporting; Tech Lab membership not required; the program is global. Certified companies receive a seal, a compliant-companies listing, and an entry in the Transparency Center's Podcast Compliance API. Supports: Audit process and third-party auditors, Annual fee (no amounts) and annual recertification, Eligibility (server-side measurement reporting), Seal / listing / Podcast Compliance API.
- IAB Tech Lab Launches Podcast Measurement Compliance Program ↗ Official press release
The program entered the marketplace in December 2018; NPR and RawVoice/Blubrry were the first companies to earn Podcast Measurement Certification. Supports: Program launch date (December 2018), First certified companies.
- IAB/ABC International Spiders and Bots List ↗ Official standards page
Paid-subscription dual-file list (valid browsers/user agents plus known robots) for dual-pass IVT filtering, updated monthly. PTMG v2.2 recommends podcast platforms submit their user-agent header value to the IAB Tech Lab Spiders and Bots inclusion list so they are not treated as bots. IAB maintains no podcast-specific user-agent list. Supports: The official UA/bot list relevant to podcast filtering, Monthly update cadence, No podcast-specific IAB UA list.
-
The widely used open-source list of podcast player user-agent patterns, maintained by OPAWG — a community group of podcast developers. It is not an IAB or IAB Tech Lab product; the original opawg/user-agents repo is no longer updated and v2 is the maintained successor. Supports: Clarifying the common podcast UA list is community-maintained, not IAB.
- IAB Releases First 'Podcast Playbook' for Advertisers (2017) ↗ Official press release
IAB's first buyer's guide for podcast advertising (August 2017): audience demographics, listener behaviors, creative treatments, ad formats, delivery and targeting, and ad-effectiveness measurement. A marketing/buyer guide from IAB the trade body — not a technical specification and not an IAB Tech Lab standard. Supports: What buyer-facing podcast guidance IAB publishes, Playbook-is-not-a-standard framing.
Platform capabilities and naming change quickly. Last validated: June 12, 2026. Check current documentation before implementation.Audio, podcast, measurement, and platform references last validated: June 2026. Validate current official documentation before implementation.
Building audio, podcast, or agentic media workflows?
The operating work is to wire honest metric definitions, guideline-conformant counting, methodology-bearing attribution, and episode-level suitability into the buying path — before agents scale whatever the server logs happen to say.