What we measured
In September 2026 we took the RSS feed of each show, followed the first episode link the feed itself supplied, and measured what a crawler would actually find: server-rendered text with scripts and styles stripped, the presence of the word transcript, and whether the page carried PodcastEpisode structured data. Taking URLs from the feed rather than constructing them matters, because a guessed URL measures the guess.
The results are not close. Acquired's episode page carried 53,552 words, Talk Python's 43,114, and Python Bytes' 17,843, each with a full transcript in the page and PodcastEpisode markup present. At the other end, Darknet Diaries' episode page carried 448 words, Changelog's 809, The Verge's 1,552 and Lex Fridman's 2,326 — none with episode markup.
These are all established shows with real audiences. The difference is not production quality or popularity. It is a publishing decision about whether the episode gets a web page containing what was said.
For a guest, that decision determines almost everything about whether the appearance is findable afterwards, and it is made by the host before the guest is ever booked.
We also checked where each declared transcript actually lives. Across the ten feeds declaring transcripts, every feed serves all of its transcript files from a single host, and all of those files fetched successfully — 7KB to 206KB, in VTT, SRT, JSON, plain text and HTML. Declaration and existence line up. Reachability is the step that does not.
Words of server-rendered text, scripts and styles removed, September 2026
Source: Media Morpheus original measurement, September 2026. Episode URLs taken from each show's RSS feed.
One episode page per show. Text measured after removing script and style blocks. Not a sample of the medium.
What Google actually requires
Very little, and that is the point. Google's documentation on AI features states the condition plainly: to be eligible to appear as a supporting link, "a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements".
It adds, in the same passage, that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary".
So there is no podcast-specific lever. An episode page competes on the same terms as any other page: it must exist, be indexable, and be allowed to show a snippet.
Which means the failure mode is not sophisticated. A page with 448 words and no transcript has very little to match against a query, and a page a crawler is forbidden to fetch has nothing at all.
The blanket block nobody mentions
One result in our measurement was severe enough that we verified it twice. The whole of changelog.com's robots.txt is two lines: user-agent: * followed by disallow: /.
That is a total exclusion of every compliant crawler from the entire domain. As Google's robots.txt documentation explains, such a rule governs fetching rather than indexing, and a disallowed URL can still appear in results as a bare address — but nothing behind it can be read.
The detail that makes this worth reporting is what is being withheld. Changelog publishes more machine-readable transcripts than almost any show we examined: 777 transcript declarations on its main feed and 352 on JS Party, and we confirmed that every one of those 1,129 files is served from changelog.com itself.
So the show doing the most transcript work in our sample is also the one instructing every crawler to stay out. The transcripts exist, are declared in the feed, and download successfully when we fetch them directly — and they sit on the one domain in our sample that forbids compliant crawlers entirely.
A correction is worth recording here, because we nearly published a larger number. Practical AI, produced by the same publisher, declares 340 transcripts — but those are served from share.transistor.fm, which is open. An earlier pass of this analysis tested each show's homepage domain rather than the host actually serving the transcript file, and wrongly counted Practical AI and Diary of a CEO as blocked. Checking the file host rather than the brand halved the finding: two feeds are affected, not four.
Structured data follows the same split
PodcastEpisode markup appeared only on the pages that already had substantial text. Acquired, Talk Python and Python Bytes carried it; Darknet Diaries, Changelog, Lex Fridman, The Verge, Hidden Brain and Freakonomics did not.
That correlation is worth stating carefully. Markup does not create text, and text does not require markup. What the pattern suggests is that publishers who treat the episode page as a real web page tend to do both, and publishers who treat it as a player embed tend to do neither.
Markup is not a shortcut in any case. As Google's structured data guidelines state, using structured data "enables a feature to be present, it does not guarantee that it will be present", and it must be "a true representation of the page content".
Marking up a 448-word stub as a PodcastEpisode would be accurate and would not help, because the underlying page still says almost nothing.
The formats vary more than expected. Practical AI declares each episode three times over, in VTT, SRT and JSON; Darknet Diaries, Talk Python and Python Bytes use VTT; Acquired uses plain text; Changelog and JS Party use HTML. A system looking for a transcript therefore has to handle five different content types before it has read a single word.
Why guests are surprised
Because the appearance felt substantial and the artefact is thin.
A guest records for an hour, the conversation is genuinely good, and the episode goes out to a real audience. Then they search their own name a month later and find nothing, because the only durable public record is a page with a title, a paragraph and an audio player.
The audio contains everything that was said. As documented elsewhere in this series, the IAB's own measurement guidance describes podcast files as delivered to devices that cannot report consumption, and audio is not text. A crawler visiting an episode page reads the page, not the recording.
Nothing about this is a failure of the host. Publishing full transcripts costs money and effort, and plenty of excellent shows decide it is not worth it. It is simply a fact a guest should establish before recording rather than discover afterwards.
There is also no signal to the guest that any of this happened. Nothing in the booking conversation covers whether the episode gets a page with the conversation on it, whether that page is crawlable, or whether a transcript will exist. The guest learns the answer months later by searching their own name.
What this means for anyone booking appearances
Ask three questions before agreeing to record, all answerable in a couple of minutes from the show's public pages.
Does the show publish an episode page containing real text — show notes with substance, ideally a transcript? Open a recent episode and read it. If the page is a title and a player, that is what your appearance will produce.
Is the domain crawlable? Fetch the show's robots.txt. A wildcard disallow means nothing on that site can be read by search or AI crawlers, however good the content is.
Does the episode page carry PodcastEpisode markup? View source and search for it. Its absence is not fatal, and its presence indicates a publisher treating episodes as web pages rather than audio drops.
Media Morpheus books podcast appearances. What a booking reliably delivers is an appearance before a real audience. Whether Google or an AI assistant can find it afterwards depends almost entirely on decisions the host made before you arrived — and those decisions are public, checkable, and worth checking.
Methodology and limitations
This is an original measurement conducted in September 2026. For each show we retrieved the RSS feed, extracted the first episode link supplied by the feed itself, fetched that page with a desktop browser user agent, removed script and style blocks, and counted remaining words. We recorded occurrences of the word transcript and the presence of PodcastEpisode structured data, and separately retrieved each domain's robots.txt and parsed grouped user-agent blocks. Six limits apply. First, this is one episode page per show across a purposive selection of established podcasts; it is not a representative sample of podcasting and no population rate should be inferred from it. Second, word counts measure server-rendered text only: a page that renders content with JavaScript will under-count here, which is a real property for crawlers that do not execute scripts but is not the same as having no content. Third, counting occurrences of the word transcript indicates a signal, not verified transcript presence, though we confirmed fetchable transcript files separately for the feeds declaring them. Fourth, robots.txt records stated permission, not crawler behaviour. Fifth, EconTalk returned HTTP 403 to our requests and is excluded. Sixth, figures describe a single moment and these pages change.
Conclusion
The reason Google often does not show a podcast interview is usually mundane: there is very little there to show. Across twenty real episode pages we found a range from 276 to 53,552 words, with structured data present only at the text-heavy end and, in one case, an entire domain closed to crawlers by a two-line robots.txt while publishing hundreds of transcripts. Google's own condition is not demanding — indexed, snippet-eligible, no special optimization — which means the barrier is almost never sophistication. It is that a great many shows publish the conversation as audio and the web page as an afterthought, and a guest inherits whichever of those two approaches their host happens to take.
Frequently Asked Questions
Why doesn't my podcast interview appear in Google?
Most often because the episode page contains very little text. We measured pages ranging from 276 to 53,552 words; a page with a title and a player has almost nothing to match against a query.
What does Google actually require?
That a page "must be indexed and eligible to be shown in Google Search with a snippet". Google states there are no additional requirements or special optimizations for AI features.
Does a transcript make a difference?
It is the main thing that turns an episode into substantial readable text. The three pages in our measurement above 17,000 words all carried transcripts; the thin ones did not.
Can a show block search engines entirely?
Yes. Changelog's robots.txt is user-agent: * and disallow: / — a complete exclusion of compliant crawlers, despite the show publishing 777 transcript declarations.
Does PodcastEpisode markup help?
It describes the page; it does not create content. Google states structured data enables a feature rather than guaranteeing it, and must truly represent the page.
What should I check before recording?
Open a recent episode page and read it, fetch the show's robots.txt, and view source for PodcastEpisode markup. All three take minutes and all three are public.
Sources
- vendor research Media Morpheus, September 2026. Original measurement of episode pages, RSS feeds and robots.txt across 52 podcast feeds and twenty episode pages. Method and limitations stated above.
- official Google Search Central. AI features in Google Search, for the indexing and snippet eligibility condition.
- official Google Search Central. Introduction to robots.txt, for what a disallow rule does and does not control.