Agentic buying for programmatic audio and podcasts uses the same mechanics as any other channel: a buyer agent reads a brief, finds inventory, negotiates terms within a mandate and executes the buy. What changes is the inventory. Streaming audio behaves much like online video, with an ad selected for each listener at the moment of playback. Podcasts are different: an ad can be read by the host, inserted dynamically when the episode is downloaded, or recorded permanently into the file. An agent can only buy audio well if its mandate tells these apart and the deal states exactly how delivery will be counted, because audio measurement rests on proxies rather than direct observation.

TL;DR. Programmatic audio covers streaming music and radio, digital simulcasts of broadcast stations and podcasts, and each behaves differently once an agent tries to buy it. Dynamically inserted ads are the natural fit for agent negotiation, because they are selected per request and can be targeted and counted per request. Host-read ads are closer to sponsorships: the script, the talent and category exclusivity need human approval on both sides, so an agent can prepare the deal but should not close it alone. Podcast delivery is usually counted in downloads, so the counting basis has to be agreed and written into the deal record before price means anything.

What counts as programmatic audio inventory?

Programmatic audio inventory is any audio ad slot bought through automated deals rather than a manual insertion order. The main sources are music and radio streaming, online simulcasts of broadcast radio, podcasts, and audio inside games and apps.

Streaming inventory is the closest to familiar programmatic. A listener's player requests an ad break, the request carries signals such as device, location and time, and the ad is chosen for that listener. Delivery is usually handled through VAST, the same IAB Tech Lab ad-serving template used for video.

Podcast inventory is less uniform. Ad positions fall before, during or after the episode (pre-roll, mid-roll and post-roll), and the way an ad reaches those positions determines whether an agent can buy it at all.

What is the difference between host-read and dynamically inserted podcast ads?

A host-read ad is spoken by the presenter, usually from a script or talking points agreed with the advertiser. A dynamically inserted ad is stitched into the episode by an ad server at the moment a listener downloads or streams it, so two listeners can hear different ads in the same slot.

The distinction matters because the two are sold on different terms. Dynamically inserted slots can be targeted by geography, time and device, priced per thousand, and traded through private marketplaces or programmatic guaranteed deals, which makes them suitable for agent negotiation.

Host-read ads involve a person's voice and reputation. Hosts commonly approve which brands they read for, the script is reviewed, and the seller may offer category exclusivity within an episode. Some host reads are recorded once and inserted dynamically; others are baked into the file and stay there for the life of the episode, reaching anyone who downloads it months later. A baked-in read is closer to a sponsorship than an impression buy. An agent can handle discovery and drafting for host-read packages, but approval of script, talent and exclusivity belongs with people.

Why is audio measurement harder for an agent to verify?

Audio measurement is harder because the counted event is rarely proof that a person heard the ad. Podcast delivery is usually reported in downloads, and a download shows that a file left the server, not that the episode was played as far as the mid-roll. IAB Tech Lab maintains podcast measurement guidelines that set out how downloads should be filtered and counted, which gives buyers a common reference, but a download remains a proxy.

Streaming audio can report ad starts and completions from the player, which is closer to delivery. There is still no audio equivalent of a viewable impression: a phone in a pocket plays the ad whether or not anyone is listening. Attribution is also indirect. Audio ads rarely carry a click, so outcomes are often inferred from promo codes, vanity web addresses, brand lift surveys or probabilistic matching of listener and site visit data.

For an agent, a price per thousand downloads and a price per thousand completed ad plays are different products, even when the rate card number looks the same. An agent comparing sellers has to normalise the counting basis before it compares price, which it can only do if each seller states that basis.

What does an agent's mandate need for audio buying?

An audio mandate needs more channel-specific detail than a display mandate, because more of the risk sits in definitions. A workable checklist:

Permitted inventory types: streaming, podcast dynamic insertion, host-read, baked-in, each allowed or excluded separately.
Counting basis: filtered downloads, server-side ad starts or player-reported completions, and which one prices are compared against.
Content suitability: a show-level allow list, plus rules for episodes whose content is not known at the time of booking.
Escalation rules: host-read and baked-in placements always routed to a person for approval.
Delivery window: the dates during which delivery counts, which matters for baked-in ads that keep accruing downloads.
Frequency: limits stated as approximate, since audio frequency control usually relies on device or network signals.
Spend threshold: the commitment above which a person must sign off.
Makegood trigger: what shortfall, measured how, entitles the buyer to replacement delivery.

How does a deal record protect an audio buy?

The deal record protects an audio buy by fixing the definitions that delivery reports will later be judged against. Consider a hypothetical £20,000 podcast campaign bought at a £25 CPM against filtered downloads, mid-roll dynamic insertion only, over four weeks. That buys 800,000 counted downloads. If the seller's end-of-campaign report counts unfiltered downloads, or includes pre-roll placements, the headline figure can look on target while the buyer received less than agreed.

Resolving that is straightforward only if both parties hold the same record of the counting basis, positions, targeting, window and makegood trigger. Otherwise the conversation becomes a comparison of logs.

Standards help with describing the offer. IAB Tech Lab's AAMP framework includes OpenProposal, which standardises how publishers define products and packages, so a podcast network could describe a dynamically inserted package and a host-read package in a form a buyer agent can compare. AdCP, a separate protocol from AgenticAdvertising.org, lists audio among the channels in its documentation's example of an agency buying through one workflow. The counting basis for a particular deal is still something buyer and seller have to agree and record.

Alkimi concentrates on that record. A neutral agentic advertising marketplace working within IAB Tech Lab's AAMP framework, it has buyer and seller agents agree terms on a DealSheet that is timestamped at agreement and held by both parties, with people keeping approval rights over each deal.

Frequently asked questions

Can AI agents buy podcast advertising?

Yes, most readily for dynamically inserted ads, which are selected per download or stream and can be targeted and priced like other programmatic inventory. Host-read and baked-in ads usually need human approval of script, talent and exclusivity.

What is a baked-in podcast ad?

A baked-in ad is recorded permanently into the episode file. It plays for anyone who downloads the episode, for as long as the episode is available, so delivery keeps accruing after the campaign dates.

How is podcast advertising measured?

Podcast delivery is usually counted in downloads, filtered according to IAB Tech Lab's podcast measurement guidelines. A download shows the file was delivered, not that the ad was heard.

What should an audio deal record include?

The inventory type, ad position, counting basis, targeting, delivery window, price, spend cap and makegood trigger, held identically by buyer and seller.

How Does Agentic Buying Work for Programmatic Audio and Podcasts?