Can a machine read audio at all?
Yes, and it has been able to for years. Automatic speech recognition is mature, cheap and widely deployed, and the technical barrier to turning a podcast into searchable text disappeared some time ago.
Google demonstrated this at scale. In 2019 it announced that it would surface individual podcast episodes in search results, using automatically generated transcripts to understand what episodes were about, so that someone searching for a niche subject or a particular interview could be shown a relevant episode rather than only a show's homepage.
The capability was real and the transcripts were imperfect in the ordinary ways — names and unusual terms transcribed incorrectly, coverage uneven across episodes and languages.
So the question is not whether audio can be indexed. It is whether any operator currently does it, keeps doing it, and surfaces the result — and that is a question about product decisions rather than technology.
Cost is worth noting too, because it shapes who bothers. Transcribing an hour of speech is now a trivial expense, which means a show publishing no transcript has generally decided not to rather than been unable to.
What happened to Google's podcast indexing?
The consumer product was discontinued. Google Podcasts shut down in the United States on 2 April 2024, and in remaining regions that June, with users directed to migrate their subscriptions to YouTube Music and an export window that ran into the following months.
This matters more than a product retirement usually would, because Google Podcasts was where much of the episode-level indexing surfaced. The transcripts that made episodes searchable were associated with that ecosystem, and its closure removed the most visible destination for the work.
We should be careful about what follows from this. The shutdown of an app is not proof that Google stopped indexing podcast content, and episode results may well still appear in Search. What we can say is that we located no current Google documentation describing how podcast episodes are indexed or surfaced today, which is a meaningful absence given how thoroughly Google documents other features.
The practical lesson is about dependency rather than capability. A discovery route owned entirely by one company can be withdrawn on that company's schedule, and a podcast strategy resting on a platform's indexing behaviour rests on something the podcaster does not control.
Podcasters who had built discovery around that route were given a migration path, not a replacement for the indexing itself.
What is actually indexable?
The page, not the file. This is the central practical point and it survives every platform change.
An audio file is opaque to anything that does not transcribe it, and transcription is a service somebody has to choose to run. A web page carrying the episode — a title, a description, show notes, guest names and ideally a full transcript — is ordinary text on an ordinary URL, indexable by every search engine and reachable by every crawler, including the AI crawlers documented elsewhere in this series.
This reframes what a podcast appearance produces. The conversation is the experience; the durable, findable artefact is whatever text exists about it. A show that publishes full transcripts creates a document that can be searched, quoted, cited and linked for years. A show that publishes a title and a play button creates almost nothing a machine can read.
It also puts a useful question in a guest's hands before booking. Whether the show publishes transcripts, and whether episode pages carry real text rather than an embedded player, determines almost everything about whether the appearance is findable afterwards.
There is a secondary benefit that has nothing to do with machines. A transcript makes an episode accessible to deaf and hard-of-hearing audiences, and to anyone who would rather scan than listen, which is a reason to publish one regardless of what any search system does with it.
Classification by machine readability, not a measurement
Source: Media Morpheus classification, based on the documented requirement that crawlers retrieve text and on schema.org PodcastEpisode property definitions.
Values rank machine readability as an ordinal category. They do not measure indexing rates, which no operator publishes.
What does the schema provide, and what does it lack?
Less than podcasters assume. The schema.org vocabulary defines a PodcastEpisode type, sitting in the hierarchy Thing, CreativeWork, Episode, PodcastEpisode, and it carries the properties one would expect: associatedMedia, partOfSeries, duration in ISO 8601 form, datePublished, description and audio.
What it does not define is a transcript property. For a medium whose entire machine readability depends on text, the standard vocabulary for describing an episode has no dedicated field for the text, which means transcripts are published as ordinary page content rather than as structured data.
The type is also less settled than its existence suggests. Schema.org marks it as being “in the 'new' area”, noting that implementation feedback and adoption can help improve the definitions, and reports usage on an estimated ten thousand to one hundred thousand domains as of version 30.1.
None of this makes markup pointless. It means the marginal value lies in the visible text rather than in the annotations around it — consistent with Google's own position that structured data establishes eligibility for presentation features and guarantees nothing, examined separately in this series.
Is audio quoted by AI answer engines?
There is no published evidence that it is, and no operator documents the question.
The reasoning that makes people assume it happens is sound as far as it goes. Assistants cite web pages; transcripts are web pages; therefore a transcript could be cited like any other document. That is true, and it is a statement about the transcript rather than about the audio.
What is unevidenced is anything stronger — that audio itself is processed and quoted, that podcast content is weighted differently from other text, or that an appearance on a show produces a measurable likelihood of being cited. As documented elsewhere in this series, no major operator publishes how sources are selected for citation, so no such claim can be calibrated by anyone outside those companies.
The honest formulation is narrow. A podcast appearance that results in a well-made transcript on a crawlable page has produced a document eligible to be retrieved and cited, on the same terms as any other page. An appearance that produces only an audio file and a player has produced something these systems have no documented way to read.
Beware of circular evidence here. A vendor demonstrating that an assistant can summarise a podcast has shown that a transcript exists and is reachable, not that podcasting confers any particular advantage.
What this means for anyone booking a podcast
Ask about the text before you agree to the conversation. Does the show publish full transcripts? Do episode pages carry substantive notes naming guests and topics? Is the episode page indexable, or is it behind a player on a platform you cannot see into?
Where a show does not publish transcripts, publish your own. A guest can put a transcript, a summary and the key claims on their own site, with a link to the episode — which creates an indexable document the guest controls, on a domain that will not be discontinued by somebody else's product decision.
Treat platform dependence as a risk you can price. Google Podcasts was a major discovery route until it was not, and the appearances that survived that change were the ones with text somewhere durable.
Media Morpheus books podcast appearances. What a booking reliably produces is an appearance on a real programme in front of its audience. Whether that appearance remains findable afterwards depends almost entirely on whether text about it exists, which is partly the show's choice and partly yours — and it is the part of the transaction most worth attending to.
Methodology and limitations
This report draws on schema.org's published definition of the PodcastEpisode type, including its property list, its status marker and its reported adoption at version 30.1; and on contemporaneous reporting from multiple established technology outlets of the Google Podcasts shutdown dates and migration to YouTube Music. Five limits apply. First, we located no current Google documentation describing how podcast episodes are indexed or surfaced in Search today, and the report therefore describes the 2019 announcement and the product's discontinuation without asserting present behaviour. Second, one source encountered in research gave a materially different shutdown date from the several established outlets that corroborate the 2024 dates used here; the outlier was discarded. Third, no operator publishes whether or how podcast transcripts are used in AI answer citation, and no measurement of that was conducted for this report. Fourth, the artefact comparison chart is our own ordinal classification based on machine readability, not a measurement of indexing outcomes. Fifth, platform behaviour and vocabulary definitions change without notice.
Conclusion
Audio is the least durable part of a podcast appearance. Machines have been able to transcribe speech for years, and Google demonstrated episode-level indexing at scale in 2019 — then discontinued the product that showcased it, leaving no current documentation of how episodes are handled. The standard vocabulary for describing an episode still has no transcript field, and is marked as new. What survives all of this is text: a transcript on a crawlable page is an ordinary document that any search or AI system can retrieve on ordinary terms, while an audio file behind a player is something they have no documented way to read. For a guest, the useful question is not whether AI will quote the podcast, but whether anyone wrote down what was said.
Frequently Asked Questions
Does Google index podcast audio?
Google announced automatic transcription and episode-level indexing in 2019, but discontinued Google Podcasts in 2024 and we located no current documentation describing how episodes are handled today.
When did Google Podcasts shut down?
In the United States on 2 April 2024, and in remaining regions that June, with users directed to migrate subscriptions to YouTube Music.
Do AI assistants quote podcast episodes?
There is no published evidence either way. A transcript published on a crawlable page is an ordinary document and can be cited as one; no operator documents whether audio itself is used.
Does the podcast schema support transcripts?
No. The schema.org PodcastEpisode type defines properties such as associatedMedia, partOfSeries, duration and audio, but no transcript property.
What makes a podcast appearance findable later?
Text. A full transcript and substantive show notes on an indexable episode page, rather than an audio file behind a player.
What if the show doesn't publish transcripts?
Publish your own on a site you control, with a summary, the key points and a link to the episode. That creates an indexable document not dependent on anyone else's product decisions.
Sources
- official Schema.org. PodcastEpisode type definition, version 30.1, including property list, status marker and reported domain adoption.
- trade press Contemporaneous technology press reporting of the Google Podcasts shutdown dates and migration to YouTube Music, corroborated across multiple established outlets.
- official Google Search Central. Structured data documentation, cited for the eligibility-not-guarantee position referenced in the discussion of markup.