Read this first
This is the short, plain version of a long research paper. The full paper is called Counted, Not Heard: The State of Audio Advertising in 2026. It has the sources, the tables and the math. This page has the same findings in fewer and shorter words. It checks the public proof behind what audio ads are sold on. It does not test if audio ads work. The counts of firms, sources and checks come from the same data file as the paper. The other numbers are copied from it. Each part links to the place in the paper that backs it up.
The research runs to 30 September 2026. It looks mostly at the US.
What audio ads are
You hear them all the time. A radio spot on the drive to work. An ad in a free music app. A host who reads an ad in the middle of a podcast. A short ad you can play in a game to earn a reward. All of these are audio ads. This paper covers radio on air and online, satellite radio, music and radio apps, podcasts and games. Chapter 1 sets out what counts.
What we found
1. The size of the market depends on who counts
There is no one number. The ad trade's own count says US firms took in $8.4 billion from online audio ads in 2025. Podcasts were $2,862.2 million of that. A forecast firm, as others report it, says US spend on digital audio will be $8.02 billion in 2026. That is less than the trade count for 2025. Yet it counts more kinds of ads. Local radio on air is put at $9.79 billion for 2026. Each number counts a different thing, so you cannot add them up. In the sources it screened, this research found no current public count that takes in national radio ads. Chapter 2 has the numbers.
2. Radio shrank and podcasts grew
Five radio firms file public reports that show their radio sales. In 2025, all five took in less from radio on air. iHeartMedia's radio sales fell 5.4%. Its podcast sales rose 25.6%, to $563.7 million. The firms do not report their sales the same way. So the reports this research read cannot show how much of the market each one holds. Chapter 3 goes through the firms.
3. The rules count downloads, not listens
Podcast ads are counted under one set of rules. They count four things: a download, a listener, an ad sent to a phone, and an ad the app says it played. None of them tells you if anyone heard the ad. None tells you if they went on to buy. Only four of the 34 firms on the public list are checked for the last one, the ad that played. Eleven of the 34 were last checked more than a year before this review. The rules ask for a check each year. The list does not say which checks are still current. This research found no podcast count that the US body that vets ratings has signed off. Chapter 4 explains the rules.
4. Most proof of sales rests on a home's IP address
Say you hear an ad on your phone at home and later buy the thing. How does the seller know? Often, the tools match the IP address of your home to the IP address of the buyer. That is how they link the two. They can't tell who in the home heard the ad. In the tools this research checked, 11 of the 12 that say how they match use a home's IP address or a tracking code on the buyer's site. One big tool says it finds a match 40% to 80% of the time. A match is not proof that the ad caused the sale. In the pages read, none of the tools shows how sure it is. In the sources it screened, this research found no fair test of whether audio ads cause sales, made public and run by someone with no stake in the result. It did not search every journal. Chapter 6 goes through it.
5. Claims that audio grabs more attention lack proof
Sellers say audio ads get more attention and annoy people less. This research looked for the proof behind each claim it could find. It found none that meets a fair bar. That bar is a sound test, a method made public, and a payer with no stake in the result. The most quoted study used a model to score attention. It then said the score was like attention to a screen. The best test found was run by Pandora on its own app. One more ad each hour cut listening time by about 2%. That shows ads have a cost. It does not show they grab more attention than other ads. Lack of proof is not proof that the claims are wrong. Chapter 7 has the studies.
6. In games, "no harm to players" is not shown
Two firms that sell audio ads in games are scored here: Audiomob and Odeeo. Both say their ads do not make players quit. This research set a fair test before it read any study. A game would need to split players at random, track who comes back after a week, and show the drop is less than one point. None of the claims found gives a control group, a time span or a margin of error. So none of them can show that players were not harmed. Chapter 9 looks at games.
7. The big firms own some pipes; how open they are is not clear
Ads bought by machine run through a few pipes. SiriusXM owns one, iHeartMedia owns another and Spotify owns a third. That is on record. But they are not the only way in. Five of the tools that ad buyers use name 43 audio sources. Only 15 of them belong to these three firms. That is about a third. Leave out one tool's long list, and it is about half. These are names on a list, not money spent. In the sources it screened, this research found no public rule that puts an owner's own ads first. It found no public price for using the pipes. Chapter 5 follows the pipes.
8. AI buying is ready on paper, and has barely started
Some new rules let AI agents buy ads for a brand. At least one set of these rules can describe an audio ad. But this research found just one real audio deal done that way. An agency bought just under $10,000 of streaming audio and podcast ads from iHeartMedia. Its press release does not say which rules it used. Several firms now sell tools that write or voice an ad with AI. Only one of those tools says the voice actor must agree and must approve each ad. Chapter 10 covers AI.
9. Buyers say the gap is proof, not the story
The author's own plan for audio sellers said the main problem was the story: buyers don't get why audio works. The two buyer surveys this research could use say something else. In both, the top problem was proof. More than half of the marketers in one survey said it was hard to measure what they got back. Problems with the story came third in one survey and fourth in the other. Two surveys are not many. One of them was run with Amazon, which sells audio ads. The other was run in Australia by a trade body for sellers. Neither speaks for all US buyers. Chapter 13 sets out each test.
Who shows their work
The paper scores 34 firms and tools on 12 things a buyer can check in public. There are two groups: firms that sell audio ads, and firms that sell the tech behind them. Each group is scored on its own. Scores show what a firm writes down in public. They do not show how well its ads work. A firm that says little can score low, even if it is good.
Among the sellers, Acast, Spotify and SiriusXM Media show the most in public. Most other places could still change: for 19 of the 24 sellers, facts they leave out could move them up or down a group. Among the tech firms, none makes the top group. Most radio firms say very little in public. They say little on how they count, how buyers book ads, or how they keep ads away from bad shows. Chapter 12 has the scores.
How this was done
AI tools did the research from one plan. Each piece was then sent to a second AI tool told to find what was wrong. The research opened 993 sources and logged 641 claims. The tests were written down before the research began. An outside AI model, run by the author, then checked the facts and the draft. Each fix it raised was checked against the source first. One AI pass set the scores and one more pass tuned them. Then a second AI scorer, which did not see those scores, scored the 17 firms nearest a line again. It gave the same score in 144 of 200 cases. Where the two differed most, a third check read the sources again and fixed 10 scores. No person scored any firm. This is not a peer-reviewed paper, and no one audited it. Chapter 14 explains the method.
About the author
The author works at Samba, a firm that measures TV and sells ad data. He helps lead a working group on how AI agents buy ads. One of the tech firms scored here helped launch the ad rules that group works on, and Spotify lists Samba as a firm it works with. No firm paid for, read or saw this paper before it came out. Every firm is scored under the same rules. The abstract has the full list.